<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Span Observer</title>
    <description>The latest articles on DEV Community by Span Observer (@spanobserver).</description>
    <link>https://dev.to/spanobserver</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4230759%2Fec482140-e2ca-44fb-b097-403595d07ccf.png</url>
      <title>DEV Community: Span Observer</title>
      <link>https://dev.to/spanobserver</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/spanobserver"/>
    <language>en</language>
    <item>
      <title>Log pattern analysis: from raw volume to reusable DQL queries</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 21:30:17 +0000</pubDate>
      <link>https://dev.to/spanobserver/log-pattern-analysis-from-raw-volume-to-reusable-dql-queries-1bpn</link>
      <guid>https://dev.to/spanobserver/log-pattern-analysis-from-raw-volume-to-reusable-dql-queries-1bpn</guid>
      <description>&lt;p&gt;Every on-call engineer knows the moment: an alert fires, you open the logs for the affected service, and the screen fills with tens of thousands of near-identical lines. Only a timestamp, a request ID or an IP address changes. The usual reflex is to stack filters, try full-text searches, then write a regular expression or a &lt;code&gt;parse&lt;/code&gt; by hand just to be able to count something. It is slow, and above all it is fragile when you are discovering the source at the same time as the incident. This is where log pattern analysis changes the way you approach the problem.&lt;/p&gt;

&lt;p&gt;Log pattern analysis, available in preview in the Dynatrace SaaS Logs app, tackles the problem from the other end. Instead of looking for a needle, you ask the platform to sort the haystack: lines that share the same structure are merged into a single pattern, and the parts that vary are isolated as typed tokens. A few dozen patterns replace hundreds of thousands of records.&lt;/p&gt;

&lt;p&gt;This article explains what the feature actually does, when it saves time, and above all how to turn a discovered pattern into something lasting: a DQL query, an OpenPipeline processor or a dashboard tile.&lt;/p&gt;

&lt;h2&gt;
  
  
  What log pattern analysis does
&lt;/h2&gt;

&lt;p&gt;The principle is similar to what some «&amp;nbsp;log clustering&amp;nbsp;» tools have done for a long time, but here it runs at query time, directly on the data stored in Grail. The analysis examines the content of each message, separates static text from dynamic values and replaces the latter with a token. Lines that share the same skeleton fall into the same group, with an occurrence count and a trend over time.&lt;/p&gt;

&lt;p&gt;The interesting part is that each token is typed. Dynatrace distinguishes timestamps, IP addresses, numbers and free text, among others. In practice, a thousand lines saying «&amp;nbsp;connection refused for user X from address Y&amp;nbsp;», scattered across a set of a hundred thousand logs, become a single pattern counted a thousand times, whose X and Y values you can then explore.&lt;/p&gt;

&lt;p&gt;Results are sorted by severity, then by frequency. You can narrow the scope in several ways:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  apply a &lt;strong&gt;segment&lt;/strong&gt; prepared by domain experts (for example «&amp;nbsp;payment platform logs&amp;nbsp;»);&lt;/li&gt;
&lt;li&gt;  add ad hoc &lt;strong&gt;filters&lt;/strong&gt; for a one-off exploration;&lt;/li&gt;
&lt;li&gt;  choose a &lt;strong&gt;time window&lt;/strong&gt; ranging from a few minutes to several days or weeks.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why it differs from a classic search
&lt;/h2&gt;

&lt;p&gt;A full-text search assumes you know what to look for. A DQL aggregation assumes the fields already exist, meaning someone has written the parsing. Pattern analysis assumes neither: it reveals the shape of the data before you have formed a hypothesis. That is exactly what is missing at the start of an investigation.&lt;/p&gt;

&lt;p&gt;It also rebalances frequency and importance. In a raw list, the chattiest message drowns out everything else. Once the lines are condensed, a group of twenty occurrences that appeared last night becomes as visible as a group of a million ordinary lines. Rare, new or changing signals stand out automatically.&lt;/p&gt;

&lt;h2&gt;
  
  
  Four situations where the tool pays off
&lt;/h2&gt;

&lt;h3&gt;
  
  
  An error spike to qualify
&lt;/h3&gt;

&lt;p&gt;A service’s failure rate climbs. Rather than scrolling through messages, run the analysis over the spike window. Within seconds you will know whether a single family of errors explains the rise, whether a new pattern appeared just before, and where to start.&lt;/p&gt;

&lt;h3&gt;
  
  
  The weak signal buried in noise
&lt;/h3&gt;

&lt;p&gt;An application looks healthy because the vast majority of its logs match the usual messages. A small group of intermittent timeouts to an external API goes unnoticed. Reduced to a handful of patterns, that group jumps out.&lt;/p&gt;

&lt;h3&gt;
  
  
  High-volume access logs
&lt;/h3&gt;

&lt;p&gt;In web or API access logs, almost every line is a 200 response. The analysis separates these ordinary requests from the atypical cases: unusual status codes, unexpected endpoints, repeated authentication failures. By expanding a pattern, you see the distribution of the IP addresses involved, which lets you spot, for example, a brute-force attempt against a database.&lt;/p&gt;

&lt;h3&gt;
  
  
  An unknown source to onboard
&lt;/h3&gt;

&lt;p&gt;A new application, or third-party software whose messages nobody really understands: the first question is «&amp;nbsp;what does this source contain?&amp;nbsp;». Patterns answer it by showing the event types produced, the fields that vary, how often, and which ones deserve parsing or enrichment.&lt;/p&gt;

&lt;h2&gt;
  
  
  In practice
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Step 1 — Scope the analysis
&lt;/h3&gt;

&lt;p&gt;Open the Logs app, apply a segment or a filter on the service or host concerned, and choose a time window that matches your question. For an incident, frame the start of the problem with some margin before it; to onboard a source, take several days so you capture rare messages.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 2 — Run the analysis and read the results
&lt;/h3&gt;

&lt;p&gt;Start the pattern analysis. Begin with high-severity patterns, then compare their trends: a pattern that starts exactly when the incident began is a better suspect than a large but stable one.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 3 — Explore the tokens
&lt;/h3&gt;

&lt;p&gt;Expand the pattern you have chosen and examine the distribution of its tokens. The question becomes «&amp;nbsp;which host, which ID, which code really matters?&amp;nbsp;». If a single IP address accounts for most occurrences, the investigation changes in nature.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 4 — Get the generated query
&lt;/h3&gt;

&lt;p&gt;For an interesting pattern, the app generates a &lt;a href="https://span.observer/en/dql-fundamentals-tutorial/" rel="noopener noreferrer"&gt;DQL query&lt;/a&gt; containing a parsing statement in Dynatrace Pattern Language (DPL). The result looks like this (simplified example, adapt it to the real pattern):&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;fetch logs
| filter matchesPhrase(content, "authentication failed")
| parse content, "LD 'user=' WORD:user LD 'from ' IPADDR:client_ip"
| summarize tentatives = count(), by:{client_ip, user}
| sort tentatives desc
| limit 20
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;You now have named fields (&lt;code&gt;user&lt;/code&gt;, &lt;code&gt;client_ip&lt;/code&gt;) that you can aggregate, filter or chart as a time series.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 5 — Make it permanent
&lt;/h3&gt;

&lt;p&gt;The same DPL pattern can be used wherever DQL can: Notebooks, Dashboards, Workflows and, above all, &lt;a href="https://span.observer/en/openpipeline-logs-grail-ingestion/" rel="noopener noreferrer"&gt;OpenPipeline processors&lt;/a&gt;. In the latter case, parsing happens at ingest and the extracted fields are available to every later query, with no recomputation. For tracking over time, a dashboard tile can rely on a query like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;fetch logs
| parse content, "LD 'from ' IPADDR:client_ip"
| filter isNotNull(client_ip)
| makeTimeseries echecs = count(), by:{client_ip}
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Start with a narrow scope&lt;/strong&gt;: a segment or a per-service filter gives more readable patterns than an analysis across all the logs in the environment.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Review the generated DPL&lt;/strong&gt; before promoting it to OpenPipeline: a pattern that is too generic can capture unrelated lines, and one that is too strict will miss some.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Parse at ingest what you use often&lt;/strong&gt;: if a pattern-based query runs in several dashboards, move the parsing into OpenPipeline rather than recomputing it on every view.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Share through segments&lt;/strong&gt;: a segment well defined by the team that owns a domain makes the analysis reproducible for everyone.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Watch for new patterns&lt;/strong&gt; after a deployment: they are often the first sign of a regression.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Availability and limits
&lt;/h2&gt;

&lt;p&gt;At the time of publication, the feature is in preview for Dynatrace SaaS customers with Grail and the DPS licensing model, subject to prior enrolment in the programme. As with any preview, the behaviour and interface may change: check the product documentation for up-to-date filtering options. Also note that the analysis relies on the textual structure of messages: logs that are already heavily structured as JSON will benefit more from classic JSON parsing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Do you still need to write regular expressions?
&lt;/h3&gt;

&lt;p&gt;Less often. For an interesting pattern, the app itself generates a DQL query with the matching DPL parsing statement. You still need to review and adjust it before reusing it in a dashboard or an OpenPipeline processor.&lt;/p&gt;

&lt;h3&gt;
  
  
  Who can use pattern analysis today?
&lt;/h3&gt;

&lt;p&gt;At the time of publication, the feature is in preview for Dynatrace SaaS customers with Grail and the DPS licensing model, after enrolling in the programme. Check the product documentation for the current status.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is the analysis useful on logs already structured as JSON?
&lt;/h3&gt;

&lt;p&gt;It relies on the textual structure of messages: for logs that are already heavily structured as JSON, classic JSON parsing is generally a better fit. The analysis is most valuable on free-text messages.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;Pattern analysis does not replace DQL; it is an on-ramp to it. It shortens the most thankless phase of an investigation, when you are discovering the shape of the data, and along the way it produces the parsing you would otherwise have had to write by hand. For a beginner team, it is a gateway to structured log analysis; for an experienced user, it is that many DPL patterns they no longer have to write. Once your sources are under control, the next step is making sure they keep arriving: that is the purpose of monitoring &lt;a href="https://span.observer/en/dynatrace-log-ingestion-monitoring/" rel="noopener noreferrer"&gt;log ingestion&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Further reading: &lt;a href="https://www.dynatrace.com/news/blog/from-millions-of-log-lines-to-actionable-patterns-in-seconds-with-logs-pattern-analysis/" rel="noopener noreferrer"&gt;From millions of log lines to actionable patterns in seconds with log pattern analysis&lt;/a&gt; (Dynatrace blog).&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Dynatrace is a trademark of Dynatrace LLC. SaaS features evolve quickly: check their availability in your environment and in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/log-pattern-analysis-dql/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/analyse-patterns-logs/" rel="noopener noreferrer"&gt;https://span.observer/fr/analyse-patterns-logs/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>dpl</category>
      <category>dql</category>
      <category>dynatraceintelligence</category>
      <category>dynatracesaas</category>
    </item>
    <item>
      <title>Crossplane Dynatrace: observability as code driven from Kubernetes</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 19:30:13 +0000</pubDate>
      <link>https://dev.to/spanobserver/crossplane-dynatrace-observability-as-code-driven-from-kubernetes-15bb</link>
      <guid>https://dev.to/spanobserver/crossplane-dynatrace-observability-as-code-driven-from-kubernetes-15bb</guid>
      <description>&lt;p&gt;Your platform team already runs clusters through Kubernetes manifests, yet Dynatrace alerting profiles and dashboards still get edited by hand in the UI. Configuration drifts and nobody can say who changed what. Crossplane Dynatrace answers one operational decision: describe observability configuration like any other Kubernetes object, and let a controller keep it in line.&lt;/p&gt;

&lt;p&gt;This article does not replace the Monaco and Terraform courses on this blog. It asks a single question: when should Dynatrace be driven from inside the cluster? It draws on &lt;a href="https://www.dynatrace.com/news/blog/observability-as-code-diy-with-crossplane/" rel="noopener noreferrer"&gt;an engineer’s write-up at the vendor&lt;/a&gt;, the Upjet repository and the Crossplane and Terraform documentation, all read on 10 November 2026. I did not deploy the setup myself, so anything unverified is flagged as such.&lt;/p&gt;

&lt;h2&gt;
  
  
  Crossplane Dynatrace: which problem does it solve?
&lt;/h2&gt;

&lt;p&gt;The problem is not creating configuration, it is keeping it faithful. A script run from a laptop or a CI job applies the desired state once, then goes quiet. A manual edit made the next day stays until the next run.&lt;/p&gt;

&lt;p&gt;Kubernetes has a native answer: the operator pattern. You declare an object, a controller compares actual and desired state continuously, and it corrects the gap. Crossplane applies that principle to external systems. A &lt;em&gt;provider&lt;/em&gt; adds a service’s resources as Kubernetes objects, and the provider pod watches them and creates whatever is missing (Crossplane v2.4 documentation).&lt;/p&gt;

&lt;p&gt;For observability the benefit is concrete: teams already deploying through GitOps (see &lt;a href="https://span.observer/en/argo-cd-gitops-kubernetes-guide/" rel="noopener noreferrer"&gt;Argo CD and GitOps on Kubernetes&lt;/a&gt;) can put their alerts in the same repository, under the same code review.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the provider comes from: Terraform under the hood
&lt;/h2&gt;

&lt;p&gt;I cannot cite an official Crossplane provider for Dynatrace: the repository address in my sources did not respond when I checked, so I promise no package to install. The approach the author describes is to generate one.&lt;/p&gt;

&lt;p&gt;The tool is Upjet. Its README presents it as a framework that turns a Terraform provider into a Crossplane provider, producing CRDs, controllers and example manifests from the Terraform schema. According to the same README it powers the AWS, Azure and GCP providers and more than fifty community ones.&lt;/p&gt;

&lt;p&gt;For Dynatrace, the raw material is the &lt;code&gt;dynatrace-oss/dynatrace&lt;/code&gt; Terraform provider, whose documentation covers hundreds of resources. The write-up generates four (alerting, dashboard, metric events, email notification) with provider version 1.66.0 pinned in the Makefile. That is a demonstration choice, not an Upjet limit.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4tq9fydnx7wdtd0ovqmq.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4tq9fydnx7wdtd0ovqmq.png" alt="Excerpt from the Dynatrace Terraform provider documentation: the dynatrace\_alerting resource with an AVAILABILITY severity rule" width="800" height="365"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This HCL block comes from the &lt;code&gt;dynatrace_alerting&lt;/code&gt; resource page. Upjet reads that resource’s schema and derives a Kubernetes object whose fields mirror Terraform’s. The API token needs the &lt;code&gt;settings.read&lt;/code&gt; and &lt;code&gt;settings.write&lt;/code&gt; scopes, as stated on that page.&lt;/p&gt;

&lt;h2&gt;
  
  
  Steps: from generation to the first object
&lt;/h2&gt;

&lt;p&gt;Here is the sequence as the write-up describes it, summarised rather than replayed.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; &lt;strong&gt;Name the Terraform provider to wrap&lt;/strong&gt;: source, version and binary name, in the Makefile.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Wire up authentication&lt;/strong&gt;: the tenant URL and API token are passed to the Terraform provider through the client configuration.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Pick an external identifier strategy&lt;/strong&gt; for each resource; the write-up uses &lt;code&gt;IdentifierFromProvider&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Register each resource group&lt;/strong&gt; with a short name, then run &lt;code&gt;make generate&lt;/code&gt;. The write-up reports «&amp;nbsp;Generated 4 resources&amp;nbsp;» and CRDs written to &lt;code&gt;package/crds&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Apply the CRDs, create a secret&lt;/strong&gt; holding the URL and token, then a &lt;code&gt;ProviderConfig&lt;/code&gt; that references it.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Create an object&lt;/strong&gt; (a dashboard, say) and let the controller sync it.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;In the demonstration the operator runs locally with &lt;code&gt;make run&lt;/code&gt;. That suits learning but is not a production model. For lasting use, the provider is installed as a package through a &lt;code&gt;Provider&lt;/code&gt; object whose &lt;code&gt;spec.package&lt;/code&gt; field names the image.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7qwjig3f4r584j2hksxb.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7qwjig3f4r584j2hksxb.png" alt="Crossplane Provider manifest with the spec.package field, an example from the official documentation" width="800" height="239"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This documentation example installs the AWS S3 provider, not Dynatrace; it only shows the shape of the manifest. The write-up also leaves its full Dashboard object out (a «&amp;nbsp;More config…&amp;nbsp;» placeholder), so I reproduce no Dynatrace manifest that I could not read in full.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the operator model changes day to day
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Drift is corrected, not just detected
&lt;/h3&gt;

&lt;p&gt;A controller replays reconciliation in a loop. If someone deletes an alerting profile in the UI, the Kubernetes object stays the reference. Ask your team: do you really want the screen to be read-only in practice?&lt;/p&gt;

&lt;h3&gt;
  
  
  The cost of the Terraform engine
&lt;/h3&gt;

&lt;p&gt;The write-up flags an operating point: each reconciliation of an Upjet provider calls Terraform, with a &lt;code&gt;plan&lt;/code&gt; then an &lt;code&gt;apply&lt;/code&gt;, which can weigh on large estates. The Upjet README does describe three execution modes, two of them without the Terraform CLI (Plugin SDK v2 and Plugin Framework). I did not verify which one a generated Dynatrace provider would use, so measure before you scale up.&lt;/p&gt;

&lt;h3&gt;
  
  
  Translation limits
&lt;/h3&gt;

&lt;p&gt;The write-up notes that Kubernetes labels have no Terraform equivalent, so you handle them yourself. It also warns that Crossplane v2, still a proposal when it was written (February 2025, updated September 2025), might change composite resources and providers. Check compatibility before pinning a version.&lt;/p&gt;

&lt;h2&gt;
  
  
  Taking back control: management policies
&lt;/h2&gt;

&lt;p&gt;Crossplane lets you limit what the controller may do to each object through the &lt;code&gt;managementPolicies&lt;/code&gt; field. The documentation says the default gives the controller full control and an empty list pauses the object. It also states that the feature is beta and that the provider decides whether to support it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwbeasvqzeoisxyl9ohlj.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwbeasvqzeoisxyl9ohlj.png" alt="Excerpt from the Crossplane documentation showing managementPolicies with the values Create, Delete and Observe" width="799" height="197"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;For observability this is a useful guardrail: an alerting profile set to &lt;code&gt;Observe&lt;/code&gt; only reads. Check the generated provider’s documentation to confirm the feature is enabled before relying on it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Crossplane, Terraform or Monaco: how to decide
&lt;/h2&gt;

&lt;p&gt;Crossplane is not a Terraform replacement here, since it reuses it. The decision lies elsewhere.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;You already run Kubernetes clusters and GitOps&lt;/strong&gt;: a controller in the cluster fits your chain, and state lives in the Kubernetes API.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;You have no central Kubernetes platform&lt;/strong&gt;: Terraform or Monaco stay simpler. The &lt;a href="https://span.observer/en/monaco-vs-terraform-dynatrace-configuration/" rel="noopener noreferrer"&gt;Monaco vs Terraform comparison&lt;/a&gt; covers that choice.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;You want to manage Dynatrace objects with the Terraform provider directly&lt;/strong&gt;: see &lt;a href="https://span.observer/en/dynatrace-terraform-provider-guide/" rel="noopener noreferrer"&gt;Terraform for alerts and dashboards&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;On targets, the Terraform provider page accepts a SaaS environment URL or a platform URL, but I did not verify how a generated Crossplane provider behaves against a Managed cluster. Treat Managed and SaaS separately and test on a throwaway environment.&lt;/p&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Start with a single resource type and a minimal token scope.&lt;/li&gt;
&lt;li&gt;  Keep the token secret out of the repository and out of manifests.&lt;/li&gt;
&lt;li&gt;  Use &lt;code&gt;Observe&lt;/code&gt; before allowing &lt;code&gt;Update&lt;/code&gt; and &lt;code&gt;Delete&lt;/code&gt; on existing objects.&lt;/li&gt;
&lt;li&gt;  Watch the controller’s load: the Upjet documentation lists dedicated Prometheus metrics, including a gauge of active Terraform invocations.&lt;/li&gt;
&lt;li&gt;  Keep an inventory of what code manages and what remains manual.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Is Crossplane Dynatrace available as a ready-to-install package?
&lt;/h3&gt;

&lt;p&gt;I could not confirm it. The source write-up generates the provider with Upjet; I did not verify that a published, maintained image exists.&lt;/p&gt;

&lt;h3&gt;
  
  
  Should we drop Terraform?
&lt;/h3&gt;

&lt;p&gt;No. The generated provider reuses the Terraform provider’s logic, and the write-up says the two tools are not mutually exclusive.&lt;/p&gt;

&lt;h3&gt;
  
  
  What are the risks at scale?
&lt;/h3&gt;

&lt;p&gt;The write-up cites the load from repeated &lt;code&gt;plan&lt;/code&gt; and &lt;code&gt;apply&lt;/code&gt; runs and the lack of a label equivalent. Test volumes with your own resources.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;Choose Crossplane when the cluster is already your control plane, not out of fashion. Begin with one read-only resource, measure the load, then widen. For more on the API side, read &lt;a href="https://span.observer/en/monaco-managed-to-saas-configuration/" rel="noopener noreferrer"&gt;the configuration migration from Managed to SaaS&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Brands mentioned belong to their owners. Tools change quickly: check commands and options against the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/crossplane-dynatrace-observability-as-code/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/crossplane-dynatrace-observabilite-as-code/" rel="noopener noreferrer"&gt;https://span.observer/fr/crossplane-dynatrace-observabilite-as-code/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>configurationascode</category>
      <category>crossplane</category>
      <category>dynatrace</category>
    </item>
    <item>
      <title>Site Reliability Guardian AWS: validate every EC2 tag change</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 17:30:18 +0000</pubDate>
      <link>https://dev.to/spanobserver/site-reliability-guardian-aws-validate-every-ec2-tag-change-3nnm</link>
      <guid>https://dev.to/spanobserver/site-reliability-guardian-aws-validate-every-ec2-tag-change-3nnm</guid>
      <description>&lt;p&gt;On a Tuesday, someone edits the &lt;code&gt;env&lt;/code&gt; tag on an EC2 instance, from &lt;code&gt;canary&lt;/code&gt; to &lt;code&gt;prod&lt;/code&gt;. No deployment, no pull request, yet the machine’s role has just changed. Nobody checks that the service still meets its objectives. Pairing Site Reliability Guardian AWS events closes that gap: a tag change becomes the signal that starts a validation. The vendor describes the idea in &lt;a href="https://www.dynatrace.com/news/blog/srg-aws-tag-changes/" rel="noopener noreferrer"&gt;a blog post&lt;/a&gt;; what follows is an operator’s reading, checked against the documentation on 10 November 2026.&lt;/p&gt;

&lt;p&gt;I start from the decision, not the tool: which infrastructure changes deserve an automatic check, and which check? I did not replay the setup on an AWS account. Anything I could not confirm is marked «&amp;nbsp;unverified&amp;nbsp;».&lt;/p&gt;

&lt;p&gt;On the Dynatrace side, everything here applies to &lt;strong&gt;Dynatrace SaaS&lt;/strong&gt;. I read no equivalent page for Managed, so do not assume it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Site Reliability Guardian AWS: when should you validate?
&lt;/h2&gt;

&lt;p&gt;Site Reliability Guardian is an app that compares indicators with thresholds. A «&amp;nbsp;guardian&amp;nbsp;» groups objectives, each computed by a DQL query. The documentation separates two families: Lifecycle guardians, built on software-lifecycle events, and Business guardians, built on business events. For an AWS tag, you pick the second: there is no deployment event, so you create a business event.&lt;/p&gt;

&lt;p&gt;A naming check, as requested: the documentation page I read still says «&amp;nbsp;Site Reliability Guardian&amp;nbsp;» and «&amp;nbsp;guardian&amp;nbsp;». I found no rename and no deprecation notice there. The source post does not mention one either. If your tenant shows something different, trust the interface.&lt;/p&gt;

&lt;p&gt;To decide, ask three questions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  Does the tag drive anything (routing, monitoring, autoscaling groups, backup policy)?&lt;/li&gt;
&lt;li&gt;  Can a tag change degrade a service without any deployment tool noticing?&lt;/li&gt;
&lt;li&gt;  Do health objectives already exist for that service? If not, start there, as in &lt;a href="https://span.observer/en/site-reliability-guardian-tutorial/" rel="noopener noreferrer"&gt;the release validation tutorial&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The event’s path from AWS to Dynatrace
&lt;/h2&gt;

&lt;p&gt;The chain has four links. EventBridge receives the tag event, a rule filters it, an API destination posts it to Dynatrace, and Dynatrace stores it as a business event and triggers a workflow. Each link can break alone, so test each one alone.&lt;/p&gt;

&lt;h3&gt;
  
  
  Link 1: the EventBridge filter
&lt;/h3&gt;

&lt;p&gt;AWS publishes tag changes under the source &lt;code&gt;aws.tag&lt;/code&gt;, with the type &lt;code&gt;Tag Change on Resource&lt;/code&gt;. The EventBridge documentation states best-effort delivery: expect that an event may be missed or arrive twice. The pattern below is limited to EC2 instances.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo9x1dyofr4oiq58lxd79.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo9x1dyofr4oiq58lxd79.png" alt="EventBridge event pattern in JSON: source aws.tag, detail-type Tag Change on Resource, service ec2 and resource-type instance" width="800" height="281"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The source post filters on the same values. The &lt;code&gt;changed-tag-keys&lt;/code&gt; and &lt;code&gt;version&lt;/code&gt; fields, which would let you target a single tag key, come from my memory of the AWS documentation: unverified, so check them before use.&lt;/p&gt;

&lt;h3&gt;
  
  
  Link 2: dress the event as a CloudEvent
&lt;/h3&gt;

&lt;p&gt;The business events ingest API accepts plain JSON or CloudEvents. For CloudEvents, four fields are mandatory: &lt;code&gt;specversion&lt;/code&gt;, &lt;code&gt;id&lt;/code&gt;, &lt;code&gt;source&lt;/code&gt; and &lt;code&gt;type&lt;/code&gt;. The &lt;code&gt;type&lt;/code&gt; becomes &lt;code&gt;event.type&lt;/code&gt; in Dynatrace, and the workflow will filter on it. An EventBridge input transformer builds this document.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5bmc2psvj4h5szn6w17i.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5bmc2psvj4h5szn6w17i.png" alt="EventBridge input transformer: an InputPathsMap, then an InputTemplate that produces a CloudEvent of type ec2.tag.change carrying the event's resources and detail" width="800" height="365"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This example is my own and I did not run it: unverified. The source post instead uses a transformer that passes &lt;code&gt;detail&lt;/code&gt;, &lt;code&gt;id&lt;/code&gt;, &lt;code&gt;region&lt;/code&gt;, &lt;code&gt;resources&lt;/code&gt;, &lt;code&gt;source&lt;/code&gt; and &lt;code&gt;time&lt;/code&gt;. Both approaches work as long as the mandatory fields are present. For the API destination’s header, the Dynatrace documentation lists &lt;code&gt;application/cloudevent+json&lt;/code&gt; as an accepted type.&lt;/p&gt;

&lt;h3&gt;
  
  
  Link 3: authentication
&lt;/h3&gt;

&lt;p&gt;The post creates an access token with the &lt;code&gt;bizevents.ingest&lt;/code&gt; scope and stores it in the EventBridge connection («&amp;nbsp;API Key&amp;nbsp;» mode), preferably through AWS Secrets Manager. The ingest page also mentions a platform token (&lt;code&gt;openpipeline:bizevents:ingest&lt;/code&gt;) and an OAuth client (&lt;code&gt;storage:events:write&lt;/code&gt;). It adds that the newer platform URL, on &lt;code&gt;apps.dynatrace.com&lt;/code&gt;, accepts OAuth only. Which mode suits your tenant? Unverified: look at your environment URL before choosing.&lt;/p&gt;

&lt;p&gt;Whatever the mode, apply least privilege: ingest scope only, a dedicated token, a planned rotation. For the rest of the SaaS API, see &lt;a href="https://span.observer/en/dynatrace-saas-api-reference/" rel="noopener noreferrer"&gt;the API schemas&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to check that it works
&lt;/h2&gt;

&lt;p&gt;Change a tag on a test instance, then query business events in a Notebook. If the query returns nothing, the problem sits between AWS and Dynatrace; leave the guardian alone for now.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw4h22qbsj3sp1wa636xk.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw4h22qbsj3sp1wa636xk.png" alt="DQL query fetch bizevents filtered on event.type ec2.tag.change, sorted by date, then the same filter used as the workflow trigger" width="800" height="299"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Once events arrive, move on to the guardian. For syntax, reread &lt;a href="https://span.observer/en/dql-commands-filter-summarize-timeseries/" rel="noopener noreferrer"&gt;the DQL filter, summarize and timeseries commands&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Wire the guardian to this trigger
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; Create a guardian for the service that depends on the tag. The post starts from the «&amp;nbsp;Four Golden Signals&amp;nbsp;» template; adapt the objectives to your service.&lt;/li&gt;
&lt;li&gt; Choose the Business type, then click &lt;strong&gt;Automate&lt;/strong&gt;. Dynatrace generates a workflow whose trigger filters &lt;code&gt;bizevents&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt; Set the filter to &lt;code&gt;event.type == "ec2.tag.change"&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt; Add a wait before the action. The post uses 600 seconds so the system can settle: an immediate check measures the old state, not the new one. Tune the delay to how fast your fleet converges.&lt;/li&gt;
&lt;li&gt; Change a tag and follow the workflow run, then the guardian result.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The validation action exposes &lt;code&gt;guardian_id&lt;/code&gt;, &lt;code&gt;validation_id&lt;/code&gt;, &lt;code&gt;validation_url&lt;/code&gt; and &lt;code&gt;validation_status&lt;/code&gt;. The last one takes &lt;code&gt;info&lt;/code&gt;, &lt;code&gt;pass&lt;/code&gt;, &lt;code&gt;warning&lt;/code&gt;, &lt;code&gt;fail&lt;/code&gt; or &lt;code&gt;error&lt;/code&gt;, and the overall result follows the most severe objective. So chain a next step on &lt;code&gt;fail&lt;/code&gt;: a notification, a ticket, or reverting the tag. For workflows in general, see &lt;a href="https://span.observer/en/dynatrace-workflows-automation/" rel="noopener noreferrer"&gt;the workflows module&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What can go wrong
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Bursts: ten instances retagged at once trigger ten validations. Plan a narrower filter or some grouping.&lt;/li&gt;
&lt;li&gt;  Permissions: the workflow needs &lt;code&gt;automation:workflows:*&lt;/code&gt; rights, the guardian needs &lt;code&gt;app-settings:objects:*&lt;/code&gt; rights, and validation needs read rights on the queried data, plus &lt;code&gt;storage:events:write&lt;/code&gt; to write its own business events.&lt;/li&gt;
&lt;li&gt;  Adaptive thresholds: they only take effect after at least five validations; until then the result stays «&amp;nbsp;info&amp;nbsp;». A rarely changed tag will feed them slowly, so prefer static thresholds.&lt;/li&gt;
&lt;li&gt;  Limits: 1,000 guardians and 50 objectives per guardian, according to the documentation.&lt;/li&gt;
&lt;li&gt;  Best-effort delivery on the AWS side: do not make the guardian your only safeguard.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Habits worth adopting
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  One event type per signal (&lt;code&gt;ec2.tag.change&lt;/code&gt;, not a catch-all), so filters stay readable.&lt;/li&gt;
&lt;li&gt;  A token dedicated to ingestion, kept in a vault.&lt;/li&gt;
&lt;li&gt;  A trial on a disposable instance before production, one link at a time.&lt;/li&gt;
&lt;li&gt;  Objectives tied to your SLOs; for alerting, &lt;a href="https://span.observer/en/slo-burn-rate-alerts-dynatrace/" rel="noopener noreferrer"&gt;burn-rate alerts&lt;/a&gt; complete the picture.&lt;/li&gt;
&lt;li&gt;  A quarterly review: which tag changes actually failed a validation?&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Do I need an agent or a full AWS integration?
&lt;/h3&gt;

&lt;p&gt;Not for this scenario. EventBridge only has to post the event to the business events ingest API. A CloudWatch integration remains useful for metrics; see &lt;a href="https://span.observer/en/dynatrace-aws-integration-cloudwatch/" rel="noopener noreferrer"&gt;the article on AWS and CloudWatch&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does it work for resources other than EC2?
&lt;/h3&gt;

&lt;p&gt;The post says the approach extends to other AWS resources. The exact list of types that emit a tag event is unverified: check the AWS documentation.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is Site Reliability Guardian still current?
&lt;/h3&gt;

&lt;p&gt;The documentation I read on 10 November 2026 uses it as the app name, with no mention of a rename or deprecation. Reread it before building a long project on it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;A tag is an operations decision, so it deserves a validation. Filter tightly in EventBridge, prove each link, let the system settle before judging, then hook an action to &lt;code&gt;fail&lt;/code&gt;. Try it on a test instance first.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC (nor with Amazon Web Services). Trademarks belong to their owners. Tools change fast: check commands and options in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/site-reliability-guardian-aws-tag-changes/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/site-reliability-guardian-aws-tags-ec2/" rel="noopener noreferrer"&gt;https://span.observer/fr/site-reliability-guardian-aws-tags-ec2/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>aws</category>
      <category>businessevents</category>
      <category>dynatracesaas</category>
      <category>eventbridge</category>
    </item>
    <item>
      <title>Davis forecast in Dashboards: decide before capacity runs out</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 15:30:13 +0000</pubDate>
      <link>https://dev.to/spanobserver/davis-forecast-in-dashboards-decide-before-capacity-runs-out-18kd</link>
      <guid>https://dev.to/spanobserver/davis-forecast-in-dashboards-decide-before-capacity-runs-out-18kd</guid>
      <description>&lt;p&gt;Thursday, 4 p.m.: the head of operations asks whether the archive database will still have room at the end of the quarter. You have a free-space curve, but no answer. A Davis forecast in a dashboard exists for exactly this moment: it turns a past curve into a numeric estimate, with a margin of error, before you order storage or decide to do nothing.&lt;/p&gt;

&lt;p&gt;This article starts from the decision, not from the feature. When does a forecast really help? How do you attach it to a tile? And how do you avoid trusting a line that has no value? The advice applies to Dynatrace SaaS with Grail and the Dashboards app. I have not checked Dynatrace Managed, so I promise nothing there.&lt;/p&gt;

&lt;p&gt;A word on method: the facts come from the Dashboards documentation and the Davis for Workflows page, read on 10 November 2026. span.observer has not tested these settings on a tenant, so anything missing from those pages is marked «&amp;nbsp;unverified&amp;nbsp;».&lt;/p&gt;

&lt;h2&gt;
  
  
  When a Davis forecast changes a decision
&lt;/h2&gt;

&lt;p&gt;A forecast is only worth something if it leads to a dated action. Three situations justify one:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;A resource running out&lt;/strong&gt;: disk, quota, connection count. The question is «&amp;nbsp;in how many days?&amp;nbsp;».&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;A load that keeps climbing&lt;/strong&gt;: traffic, message queues. The question is «&amp;nbsp;do we add capacity before the next campaign?&amp;nbsp;».&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;A budget drifting&lt;/strong&gt;: ingested log volume, monthly cost. The question is «&amp;nbsp;where will we be at month end?&amp;nbsp;».&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Conversely, an erratic curve with no trend or rhythm has nothing to predict. In that case, a threshold or anomaly detection is better, and the documentation offers both in the same tile section: auto-adaptive threshold, seasonal baseline or static threshold.&lt;/p&gt;

&lt;p&gt;A post on the official Dynatrace blog, &lt;a href="https://www.dynatrace.com/news/blog/better-dashboarding-with-dynatrace-davis-ai/" rel="noopener noreferrer"&gt;Better dashboarding with Dynatrace Davis AI&lt;/a&gt;, presents the idea from the product side. Here I stick to the walkthrough and the operational pitfalls.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: write a query worth predicting
&lt;/h2&gt;

&lt;p&gt;The forecast is computed on the tile result. A poor query therefore gives a poor forecast. Start with a regular series over a window long enough to contain one full cycle of your activity.&lt;/p&gt;

&lt;p&gt;Here is a free disk space tile per host and disk, over seven days, with one point per hour:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5dwkgv48h86de33wlcn0.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5dwkgv48h86de33wlcn0.png" alt="DQL timeseries query on dt.host.disk.free, grouped by host and disk, over seven days with a one-hour interval, limited to 50 series" width="800" height="239"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Two choices matter. Grouping gives one series per disk, so you can spot the failing disk instead of averaging the whole fleet. And the 50-series limit keeps the chart readable, because the documentation says the analyzer accepts up to 1000 series, set through the &lt;strong&gt;Limit&lt;/strong&gt; option.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: turn on the Forecast analyzer for the tile
&lt;/h2&gt;

&lt;p&gt;In the Dashboards app, the documented procedure is short:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; Run the tile query.&lt;/li&gt;
&lt;li&gt; Expand the &lt;strong&gt;Davis AI&lt;/strong&gt; section of the options panel.&lt;/li&gt;
&lt;li&gt; In the &lt;strong&gt;Analyzers&lt;/strong&gt; list, pick &lt;strong&gt;Forecast&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt; Switch the toggle at the top from «&amp;nbsp;AI data analysis is not active&amp;nbsp;» to «&amp;nbsp;AI data analysis is active&amp;nbsp;»: analyzers are off by default.&lt;/li&gt;
&lt;li&gt; Choose the &lt;strong&gt;Davis AI analysis&lt;/strong&gt; visualization.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Two settings drive the forecast. &lt;strong&gt;Data points to predict&lt;/strong&gt; sets how many future points you get; the page warns that more points mean a less reliable forecast and a longer run. &lt;strong&gt;Forecast offset&lt;/strong&gt; ignores the last N points and forecasts them instead, which is handy for testing the model against what really happened.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F14an1pg7m5x6wr3k59z7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F14an1pg7m5x6wr3k59z7.png" alt="Davis AI section settings of a tile, using the documentation labels: Forecast, Data points to predict, Forecast offset, Limit and the Davis AI analysis visualization" width="800" height="239"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;On permissions, you need &lt;code&gt;davis:analyzers:read&lt;/code&gt; to see the configuration and &lt;code&gt;davis:analyzers:execute&lt;/code&gt; to run the analyzer. If a colleague sees the tile without the forecast, check their IAM policies first.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: read the forecast with suspicion
&lt;/h2&gt;

&lt;p&gt;The &lt;strong&gt;Davis AI analysis&lt;/strong&gt; visualization can show the chart, the table or both (the &lt;strong&gt;Visible sections&lt;/strong&gt; option). Keep both while you form an opinion.&lt;/p&gt;

&lt;p&gt;The Grail and Davis AI capacity post says the model is chosen automatically from the characteristics of the series, and that the result is probabilistic, with a predicted value and bounds. So remember that a forecast is a range, not a single date. Decide on the unfavourable bound.&lt;/p&gt;

&lt;p&gt;This is the test I apply before telling anyone about a forecast:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Backtest&lt;/strong&gt;: set &lt;strong&gt;Forecast offset&lt;/strong&gt; to predict points you already know, then compare.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Sensible horizon&lt;/strong&gt;: do not go beyond a third of the history window. That is a personal rule of thumb, not a vendor recommendation.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Past events&lt;/strong&gt;: a recent migration or clean-up breaks the trend; exclude the period or note it on the tile.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 4: get out of the tile with a workflow
&lt;/h2&gt;

&lt;p&gt;A dashboard wakes nobody up. If the decision must be automatic, the Davis for Workflows documentation describes the &lt;strong&gt;Generic Forecast Analysis&lt;/strong&gt; analyzer, used in an «&amp;nbsp;Analyze data&amp;nbsp;» action. Its example predicts disk capacity from &lt;code&gt;avg(dt.host.disk.free)&lt;/code&gt; over seven days in 120 bins, with 100 points to predict.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fq27uie8zyq6skaysw7sr.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fq27uie8zyq6skaysw7sr.png" alt="Parameters of the Generic Forecast Analysis example from the documentation: query on dt.host.disk.free, 120 bins, 100 data points to predict and checks that analysisStatus equals OK and forecastQualityAssessment equals VALID" width="799" height="260"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The key point is the safeguard: the example only uses a series if &lt;code&gt;analysisStatus&lt;/code&gt; is &lt;code&gt;OK&lt;/code&gt; and &lt;code&gt;forecastQualityAssessment&lt;/code&gt; is &lt;code&gt;VALID&lt;/code&gt;. Adopt that habit. Without it, you will alert on forecasts the engine itself considers doubtful. The page also warns that the output volume can be very large, so filter before acting.&lt;/p&gt;

&lt;p&gt;One useful caveat: that page does not mention dashboards and routes capacity setup through Notebooks. I therefore cannot say that a workflow can be triggered from a tile: unverified.&lt;/p&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  One tile, one decision: write the expected action in the tile title («&amp;nbsp;order storage before…&amp;nbsp;»).&lt;/li&gt;
&lt;li&gt;  Limit the series; a forecast over 1000 curves cannot be read.&lt;/li&gt;
&lt;li&gt;  Keep anomaly detection for series without a trend, and forecasting for those that have one.&lt;/li&gt;
&lt;li&gt;  Review the forecast weekly at first: a drifting model shows up quickly.&lt;/li&gt;
&lt;li&gt;  Give the tile an owner and a runbook, as you would for an &lt;a href="https://span.observer/en/slo-dashboard-leadership-dynatrace/" rel="noopener noreferrer"&gt;SLO dashboard for management&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does a Davis forecast work on every query?
&lt;/h3&gt;

&lt;p&gt;It applies to a numeric time series. A tile that returns a list of rows is a poor candidate: turn it into a series first with &lt;code&gt;timeseries&lt;/code&gt; or &lt;code&gt;makeTimeseries&lt;/code&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need to write DQL to get a forecast?
&lt;/h3&gt;

&lt;p&gt;You need a query that produces the series, but the analysis itself is set through the interface. If DQL is new to you, see &lt;a href="https://span.observer/en/davis-copilot-natural-language-limits/" rel="noopener noreferrer"&gt;Davis CoPilot and query generation&lt;/a&gt;, and always review the generated query.&lt;/p&gt;

&lt;h3&gt;
  
  
  How many points can I predict?
&lt;/h3&gt;

&lt;p&gt;The Dashboards page gives no ceiling; it only says that more points make the forecast less reliable. No numeric limit is verified here.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;Start from a capacity question, write a regular query, turn on the Forecast analyzer, then test it on the past before showing it to anyone. To learn how to build the tiles themselves, go through &lt;a href="https://span.observer/en/dynatrace-saas-dashboard-dql-examples/" rel="noopener noreferrer"&gt;the SaaS dashboard examples in DQL&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Trademarks belong to their owners. Tools change quickly: check the options in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/davis-forecast-dashboards-capacity-planning/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/previsions-davis-tableaux-de-bord-dynatrace/" rel="noopener noreferrer"&gt;https://span.observer/fr/previsions-davis-tableaux-de-bord-dynatrace/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>capacity</category>
      <category>dashboards</category>
      <category>davis</category>
      <category>dql</category>
    </item>
    <item>
      <title>Azure Data Factory Dynatrace: spot failing pipelines</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 13:30:16 +0000</pubDate>
      <link>https://dev.to/spanobserver/azure-data-factory-dynatrace-spot-failing-pipelines-419k</link>
      <guid>https://dev.to/spanobserver/azure-data-factory-dynatrace-spot-failing-pipelines-419k</guid>
      <description>&lt;p&gt;A nightly pipeline fails at 3 a.m., nobody is told, and by 9 a.m. the finance dashboard shows yesterday’s figures. Wiring Azure Data Factory Dynatrace together answers that scenario concretely: which pipeline failed, since when, and whether the fault lies in the activity or in the machine running it. The vendor post, &lt;a href="https://www.dynatrace.com/news/blog/transforming-azure-data-factory-operations-with-dynatrace/" rel="noopener noreferrer"&gt;Transforming Azure Data Factory operations with Dynatrace&lt;/a&gt;, sets out that ambition; what follows is an operator’s reading, cross-checked against the documentation.&lt;/p&gt;

&lt;p&gt;I start from an operations decision: what should happen when a pipeline does not finish? Your answer determines which signals you need. The scope is Dynatrace SaaS; I read nothing about Managed on this topic, so I promise nothing there.&lt;/p&gt;

&lt;p&gt;A note on method: span.observer has not connected this integration to a real factory. The facts come from the Dynatrace documentation and the post cited, read on 10 November 2026. Anything I could not read is marked «&amp;nbsp;not verified&amp;nbsp;».&lt;/p&gt;

&lt;h2&gt;
  
  
  Azure Data Factory Dynatrace: two sources of signal
&lt;/h2&gt;

&lt;p&gt;To monitor Azure Data Factory with Dynatrace, two streams complement each other. The first is the classic Azure service: Dynatrace queries the Azure Monitor API about every 5 minutes and derives metrics. The second is the factory’s diagnostics, which produce one log per run. The vendor post talks about ingesting logs and metrics from Azure Monitor and correlating them.&lt;/p&gt;

&lt;p&gt;The difference is simple: metrics tell you &lt;em&gt;how many&lt;/em&gt; (how many runs failed this hour), logs tell you &lt;em&gt;which one and why&lt;/em&gt; (which pipeline, which error message). You build an alert on the former and run an investigation in the latter. For Azure logs in general, see &lt;a href="https://span.observer/en/dynatrace-azure-integration-monitor-logs/" rel="noopener noreferrer"&gt;sending Azure Monitor logs to Dynatrace&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Bring the Data Factory metrics into Dynatrace
&lt;/h2&gt;

&lt;p&gt;The integration documentation asks for Dynatrace 1.199 or later and an environment ActiveGate 1.195 or later. It lists no permission specific to the service: the general Azure connection model applies (a read-only service principal, an ActiveGate that reaches &lt;code&gt;management.azure.com&lt;/code&gt; and &lt;code&gt;login.microsoftonline.com&lt;/code&gt;). The custom-role templates on the general page do include the read action &lt;code&gt;Microsoft.DataFactory/*/read&lt;/code&gt;.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; Check that an ActiveGate with the Azure module is available, under Deployment Status then ActiveGates.&lt;/li&gt;
&lt;li&gt; Create or reuse the Azure connection: Settings, Cloud and virtualization, Azure, Connect new instance.&lt;/li&gt;
&lt;li&gt; In the connection’s Services section, add Azure Data Factory (V1, V2) with Manage services then Add service.&lt;/li&gt;
&lt;li&gt; Open the Dashboards page: a preset dashboard appears there with the recommended metrics.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Two useful details from the page. On a service that is already monitored, the preset dashboard may not show up until you resave the credentials (Save on the instance). And the dashboard cannot be edited: clone it, then edit the copy.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which metrics to read first
&lt;/h2&gt;

&lt;p&gt;The page lists dozens of metrics for version 2. Not all deserve your attention on day one. These are the ones the documentation flags as recommended.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcwl3oft6l61vsopqcxzs.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcwl3oft6l61vsopqcxzs.png" alt="The seven Azure Data Factory V2 metrics flagged as recommended in the Dynatrace documentation, with their dimensions" width="799" height="260"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Read them layer by layer. Pipeline runs (succeeded or failed) answer «&amp;nbsp;did the job finish?&amp;nbsp;». Activity runs, with the &lt;code&gt;PipelineName&lt;/code&gt; dimension, answer «&amp;nbsp;which step breaks?&amp;nbsp;». CPU and memory of the integration runtime answer «&amp;nbsp;is the machine coping?&amp;nbsp;». The &lt;code&gt;FailureType&lt;/code&gt; dimension separates failure families; I did not verify its values, so look at them in your own data before building a rule on them.&lt;/p&gt;

&lt;p&gt;The optional metrics add queue length and task pick-up delay. They help when a pipeline slows down without failing.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fj35ia2mxz2wqhirlpr48.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fj35ia2mxz2wqhirlpr48.png" alt="Seven optional Azure Data Factory V2 metrics from the Dynatrace documentation: runtime queue length, pick-up delay, failed triggers, resource count" width="799" height="260"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Mind the cost: monitored services consume Davis data units, and Azure may charge beyond one million API calls a month. Enable the optional ones only for a specific question.&lt;/p&gt;

&lt;h2&gt;
  
  
  From the late pipeline to the failed pipeline
&lt;/h2&gt;

&lt;p&gt;The vendor post details what logs add: the status and duration of each run, the time spent in the queue, the actual execution time, and a message field with runtime detail, errors included. Separating queue from execution is valuable: a pipeline that is slow because it waits on a saturated runtime is not fixed like one that is slow because of its source query.&lt;/p&gt;

&lt;p&gt;For those logs to exist, the factory must send its diagnostic settings to a destination Dynatrace can read. The exact category and field names are not in the pages I read: not verified, so check your Azure portal and your log view before writing a query.&lt;/p&gt;

&lt;h2&gt;
  
  
  Tie the pipeline to a business stake
&lt;/h2&gt;

&lt;p&gt;The post’s most interesting idea is to step beyond the technical. A Lookup activity reads a business value at the source (the example given is the day’s revenue in a spreadsheet) and passes it on as a custom user property. You can then see, next to pipeline health, whether the day’s figure actually arrived. The failure is no longer «&amp;nbsp;pipeline X is red&amp;nbsp;» but «&amp;nbsp;the sales report is incomplete&amp;nbsp;».&lt;/p&gt;

&lt;p&gt;If that appeals, look at your &lt;a href="https://span.observer/en/dynatrace-business-events-kpi-tracking/" rel="noopener noreferrer"&gt;business events and business KPIs&lt;/a&gt;: it is the same logic, giving a technical signal a business meaning.&lt;/p&gt;

&lt;h2&gt;
  
  
  Decide the reaction: alert, then automatic retry?
&lt;/h2&gt;

&lt;p&gt;Back to the opening decision. For the alert, rely on pipeline failures rather than a duration threshold: see &lt;a href="https://span.observer/en/create-a-dynatrace-alert-step-by-step/" rel="noopener noreferrer"&gt;how to create an alert step by step&lt;/a&gt;. For recovery, the post describes Workflows that call the Azure Data Factory REST API to rerun failed pipelines automatically, including those owned by other teams whose retry policy is not configured.&lt;/p&gt;

&lt;p&gt;One safeguard: rerunning without understanding can double writes. Keep automatic retries for idempotent pipelines, cap the number of attempts and log every rerun. The Workflow details (trigger, authentication, request body) are not in the post: not verified.&lt;/p&gt;

&lt;p&gt;In an Azure context, a failure can also come from the network. On that front, &lt;a href="https://span.observer/en/azure-snat-exhaustion-dynatrace-alert/" rel="noopener noreferrer"&gt;SNAT exhaustion on Azure&lt;/a&gt; is a cause to rule out before blaming the pipeline.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to check that it works
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; In Dashboards, open the preset Data Factory dashboard: pipeline runs should appear.&lt;/li&gt;
&lt;li&gt; Deliberately run a test pipeline that fails (missing source): &lt;code&gt;PipelineFailedRuns&lt;/code&gt; should rise with the pipeline name.&lt;/li&gt;
&lt;li&gt; Check that the log’s error message matches the one in the Azure portal.&lt;/li&gt;
&lt;li&gt; If nothing appears, resave the connection credentials and wait a few five-minute cycles.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Start with the seven recommended metrics and add the rest on demand.&lt;/li&gt;
&lt;li&gt;  Tag your factories in Azure: the connection accepts tag filters, which limits scope and cost.&lt;/li&gt;
&lt;li&gt;  Separate waiting from execution when analysing a slowdown.&lt;/li&gt;
&lt;li&gt;  Give each alert a pipeline owner, not a generic team.&lt;/li&gt;
&lt;li&gt;  Keep automatic retries for replayable jobs.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does Azure Data Factory Dynatrace monitoring work on Managed?
&lt;/h3&gt;

&lt;p&gt;I only read the general Azure integration documentation and the Data Factory service page, with no Managed detail. Treat the topic as not verified on Managed.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need an agent installed on the factory?
&lt;/h3&gt;

&lt;p&gt;Not for metrics: they come from the Azure Monitor API through an ActiveGate. A self-hosted integration runtime remains a machine to monitor separately.&lt;/p&gt;

&lt;h3&gt;
  
  
  Are metrics enough?
&lt;/h3&gt;

&lt;p&gt;They are enough to alert. To know why a pipeline fails, you need the error message from the logs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;First decide what a failed pipeline must trigger, then pick the signals: metrics to alert, logs to understand, business values to prioritise. Test on a staging factory before production.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC (nor with Microsoft). Brands mentioned belong to their owners. Tools change quickly: check commands and options in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/azure-data-factory-dynatrace-monitoring/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/azure-data-factory-dynatrace-pipelines/" rel="noopener noreferrer"&gt;https://span.observer/fr/azure-data-factory-dynatrace-pipelines/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>alerting</category>
      <category>azure</category>
      <category>azuredatafactory</category>
      <category>dynatrace</category>
    </item>
    <item>
      <title>Tiveo: a free IPTV player for Android TV and Google TV</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 11:30:18 +0000</pubDate>
      <link>https://dev.to/spanobserver/tiveo-a-free-iptv-player-for-android-tv-and-google-tv-1g6c</link>
      <guid>https://dev.to/spanobserver/tiveo-a-free-iptv-player-for-android-tv-and-google-tv-1g6c</guid>
      <description>&lt;p&gt;&lt;strong&gt;An app built by us, and free.&lt;/strong&gt; Tiveo is an IPTV player for Android TV developed by the team behind this blog. We present it here with enthusiasm, but we promise nothing beyond what it does.&lt;/p&gt;

&lt;p&gt;You have an IPTV subscription, an Android TV or Google TV set, and a remote whose buttons barely work in the app that came with the service. What you need is an IPTV player for Android TV designed for the living room: readable from three metres away, quick, and usable without a keyboard. That is the need Tiveo answers.&lt;/p&gt;

&lt;p&gt;One point first, to be clear: Tiveo is a player only. It provides no channels, no playlists and no content. You bring your own subscription, add it once, and the app does the rest. It is completely free: no price, no in-app purchase.&lt;/p&gt;

&lt;p&gt;The 22-second Tiveo tour: home, live TV, guide, movies and settings&lt;/p&gt;

&lt;p&gt;Prefer the YouTube version? &lt;a href="https://youtu.be/2_JoVCULfbs" rel="noopener noreferrer"&gt;Watch the demo on YouTube&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/warezfr/tiveo-releases/releases/latest" rel="noopener noreferrer"&gt;Download Tiveo (APK, free)&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Why a dedicated IPTV player for Android TV
&lt;/h2&gt;

&lt;p&gt;A TV app is judged by its remote control. How many presses to jump from one channel to another? To find the film you started yesterday? To see what is on tonight? Tiveo starts from those everyday gestures and builds everything around them.&lt;/p&gt;

&lt;p&gt;The same app also runs on Android phones, with a touch interface. It is designed for the television first, though, so a few TV-specific flows are not available on phones.&lt;/p&gt;

&lt;h2&gt;
  
  
  Home: pick up where you left off
&lt;/h2&gt;

&lt;p&gt;The home screen opens on a large featured billboard. Below it, rows follow your last activity: Continue watching, On now, Matches, Favourite channels, My list, and the latest films and series. Each profile can switch these rows on or off. Tiveo can also start straight on your last channel, a chosen channel or My list.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxmo1uo20q0xoxh28y5x1.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxmo1uo20q0xoxh28y5x1.png" alt="Tiveo home screen on a television: a large featured film, a Featured row and an On now row of live channels" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The screenshots in this article use demo channels and open films from the Blender Foundation, released under Creative Commons licences.&lt;/p&gt;

&lt;h2&gt;
  
  
  Live TV with preview and programme info
&lt;/h2&gt;

&lt;p&gt;The Live TV section shows channels as a list, a grid or with a preview on top. The focused channel plays in a live mini preview. For every channel you see the current programme, the next one and the time left.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ft6o3sipcpra6awdipyvk.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ft6o3sipcpra6awdipyvk.png" alt="Tiveo Live TV screen: channel list on the left, live preview on the right with the current programme and the next two" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  Categories, Recent, All channels and Favourites (hold OK to add, then reorder).&lt;/li&gt;
&lt;li&gt;  Type a channel number on the remote; up and down switch channel.&lt;/li&gt;
&lt;li&gt;  Sort by provider order, name or number.&lt;/li&gt;
&lt;li&gt;  A favourite can be the channel Tiveo starts on.&lt;/li&gt;
&lt;li&gt;  Fast zapping options: preload the next channel, banner duration, zapping timeout.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The TV guide and replay
&lt;/h2&gt;

&lt;p&gt;The guide is a full grid by channel and time, from 1 to 14 days (7 by default). If your guide data is shifted, a setting corrects the time by up to 12 hours either way. You can set a reminder on an upcoming programme, either as a notification or as an automatic switch to the channel.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0418am5l3n0tijeav8wx.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0418am5l3n0tijeav8wx.png" alt="Tiveo programme guide: grid of channels and programmes by time, with the current-time line and details of the selected programme" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Replay completes the guide. You pick a channel that keeps an archive, a day, then a programme, which plays from the start. Note that this depends on your provider. If archives do not exist, Tiveo says so plainly.&lt;/p&gt;

&lt;p&gt;Tiveo also spots matches in your guide data. A Sports page groups them as Live, Today and Tomorrow, with a button to watch on the right channel.&lt;/p&gt;

&lt;h2&gt;
  
  
  Movies and series: resume, seasons and episodes
&lt;/h2&gt;

&lt;p&gt;Movies and series are browsed by category, sorted by provider order, name, date added or rating. A detail page gives the synopsis, cast, director, rating and similar titles.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fn5ajek6q1ncvm7dusnng.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fn5ajek6q1ncvm7dusnng.png" alt="Series page in Tiveo: Resume S1 E2 button, My list button and the episode list of season 1 with a watched mark" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  Resume where you left off, or start from the beginning.&lt;/li&gt;
&lt;li&gt;  Series with seasons, specials, watched marks and codes such as «&amp;nbsp;S1 E3&amp;nbsp;».&lt;/li&gt;
&lt;li&gt;  The next episode starts automatically, with a countdown you can cancel.&lt;/li&gt;
&lt;li&gt;  My list to save titles for later.&lt;/li&gt;
&lt;li&gt;  Clear, plain-language messages when a server returns an error.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  One search box and settings within reach
&lt;/h2&gt;

&lt;p&gt;A single search box covers channels, programmes (now and upcoming), movies and series. Results are grouped by type.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frweci0o0ewc3jnjxl5uv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frweci0o0ewc3jnjxl5uv.png" alt="Tiveo search screen: typing &amp;quot;spri&amp;quot; shows the film Spring under Movies" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Settings form a hub of cards with its own search, plus a Quick list where you pin the settings you use most. Four themes are offered (Night blue by default, White, Aubergine, Teal), three text sizes and a reduce-motion switch. The interface language can follow the system, or be French or English.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqycfzz5jstz8c7rxynrz.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqycfzz5jstz8c7rxynrz.png" alt="Tiveo settings: list of sections on the left, Quick list on the right with Theme, Subtitles, Aspect ratio and Hardware decoding" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;For playback, you can set hardware decoding, a quality cap (Auto, 1080, 720, 480), the aspect ratio, audio tracks and subtitles. The player reconnects on its own when a live stream drops, with a configurable timeout and number of retries. On phones, picture-in-picture is available.&lt;/p&gt;

&lt;h2&gt;
  
  
  Profiles, parental control and backup
&lt;/h2&gt;

&lt;p&gt;Several profiles can coexist, each with its own name, colour, favourites, history, resume points, theme and settings. A kid profile hides locked categories and needs a PIN to leave. The 4-digit parental PIN locks out after wrong attempts and offers a recovery option. Adult categories are hidden by default.&lt;/p&gt;

&lt;p&gt;You can export sources, profiles, settings, favourites, history and resume points to a file, then restore them on another device. Source passwords and addresses are left out unless you choose to include them, and PIN codes are never exported.&lt;/p&gt;

&lt;h2&gt;
  
  
  Installing Tiveo on Android TV, step by step
&lt;/h2&gt;

&lt;p&gt;Tiveo is distributed as an APK, outside the official store. You therefore need to allow the installation of unknown apps. Here is how.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; Download the APK from the &lt;a href="https://github.com/warezfr/tiveo-releases/releases/latest" rel="noopener noreferrer"&gt;GitHub releases page&lt;/a&gt;, either on the TV itself or on a phone or computer, then copy it to the TV.&lt;/li&gt;
&lt;li&gt; On the TV, open Settings then Privacy (or Security &amp;amp; restrictions). Allow «&amp;nbsp;Install unknown apps&amp;nbsp;» for the app that will open the APK, for example a file manager.&lt;/li&gt;
&lt;li&gt; Open the APK with that app and confirm the installation.&lt;/li&gt;
&lt;li&gt; Launch Tiveo. The wizard has you pick a theme and add your first source.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Tiveo can check for new versions on its own and offer to install them. For that, allow «&amp;nbsp;Install unknown apps&amp;nbsp;» for Tiveo as well. You need Android 8.0 or later, and a remote with a D-pad for TV use.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which sources to add
&lt;/h2&gt;

&lt;p&gt;Tiveo accepts four kinds of source. The wizard tests the connection before saving, and sources can refresh in the background.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Xtream Codes&lt;/strong&gt; (server address, username, password): live TV, movies, series, guide and replay.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;M3U or M3U8 playlist&lt;/strong&gt;, with an optional separate guide address: live TV, plus the guide if you provide one.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Stalker / MAG portal&lt;/strong&gt; (portal address and MAC address): live TV, movies and series.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;XMLTV guide&lt;/strong&gt;: automatic with Xtream, or an optional extra address for any source.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;You can save several sources and set a default. However, only one source is shown at a time.&lt;/p&gt;

&lt;h2&gt;
  
  
  Privacy and limits: what to know
&lt;/h2&gt;

&lt;p&gt;Let us be frank about what Tiveo is not. It is not a service: no channels, playlists or content are provided or sold. Use only sources you are legally entitled to use.&lt;/p&gt;

&lt;p&gt;Replay depends on your provider keeping archives. If a weather chip is shown, the data comes from MET Norway, and you can turn it off. The repository holds only the compiled releases, distributed as they are through GitHub Releases, all rights reserved. Free does not mean open source.&lt;/p&gt;

&lt;p&gt;To report a bug or suggest an idea, open an issue on the repository’s &lt;a href="https://github.com/warezfr/tiveo-releases/issues" rel="noopener noreferrer"&gt;GitHub Issues&lt;/a&gt;, mentioning your device, Android version and Tiveo version (Settings, Application, About).&lt;/p&gt;

&lt;p&gt;This blog covers observability and AI, but we also build our own tools. If you like the idea, read our presentation of &lt;a href="https://span.observer/en/selecto-ai-windows-action-bar-selected-text/" rel="noopener noreferrer"&gt;Selecto AI Windows&lt;/a&gt;, another free app from the team. To understand what makes a video stream smooth, see also our guide to &lt;a href="https://span.observer/en/xbox-cloud-gaming-latency-measure-network/" rel="noopener noreferrer"&gt;measuring network latency&lt;/a&gt; on a streaming service, and our &lt;a href="https://span.observer/en/self-hosted-open-source-ai-stack/" rel="noopener noreferrer"&gt;self-hosted open-source AI stack&lt;/a&gt; for home-lab fans.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does Tiveo provide channels or an IPTV subscription?
&lt;/h3&gt;

&lt;p&gt;No. Tiveo is a player only, with no channels, no playlists and no content. You need your own subscription or playlist, one you are entitled to use.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is Tiveo really free?
&lt;/h3&gt;

&lt;p&gt;Yes. The app has no price and no in-app purchase. It is developed by our team and distributed free of charge.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need the Google Play Store to install it?
&lt;/h3&gt;

&lt;p&gt;No. Tiveo installs from an APK downloaded from the GitHub releases page. You must allow unknown apps, as described above. It runs on Android 8.0 and later.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion: give Tiveo a try
&lt;/h2&gt;

&lt;p&gt;Tiveo bets on the essentials: a home that resumes where you left off, a full guide, a single search and clear settings. Download the APK from the &lt;a href="https://github.com/warezfr/tiveo-releases/releases/latest" rel="noopener noreferrer"&gt;releases page&lt;/a&gt;, add your source, and tell us what you think.&lt;/p&gt;

&lt;h2&gt;
  
  
  Support us
&lt;/h2&gt;

&lt;p&gt;Tiveo is developed by our team and stays free. If it is useful to you, two simple gestures help it get known: leave a &lt;a href="https://github.com/warezfr/tiveo-releases" rel="noopener noreferrer"&gt;star on GitHub&lt;/a&gt; (one click with your GitHub account) and share it with someone who needs it, a colleague, a friend or a community. Every star and every share makes the project more visible, and keeps us motivated to carry on.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Independent blog, not affiliated with Dynatrace LLC. This article is not about Dynatrace: Tiveo is an app developed by this blog’s team.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/iptv-player-for-android-tv-tiveo-free/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/lecteur-iptv-android-tv-tiveo-gratuit/" rel="noopener noreferrer"&gt;https://span.observer/fr/lecteur-iptv-android-tv-tiveo-gratuit/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>androidtv</category>
      <category>apk</category>
      <category>freeapp</category>
      <category>googletv</category>
    </item>
    <item>
      <title>Dynatrace SaaS 1.350: what changes and what to check</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 09:30:21 +0000</pubDate>
      <link>https://dev.to/spanobserver/dynatrace-saas-1350-what-changes-and-what-to-check-2cdc</link>
      <guid>https://dev.to/spanobserver/dynatrace-saas-1350-what-changes-and-what-to-check-2cdc</guid>
      <description>&lt;p&gt;When a new release lands on your tenant, the real decision is not «&amp;nbsp;what should I read?&amp;nbsp;» but «&amp;nbsp;who has to act this week, and on what?&amp;nbsp;». Dynatrace SaaS 1.350 began rolling out on 4 November 2026, according to the official release notes page (last updated 6 November 2026). It lists 32 feature updates, three breaking changes and 26 fixes.&lt;/p&gt;

&lt;p&gt;This article starts from an on-call question: which dashboards, queries and alerts on your team might behave differently on Monday? It applies to Dynatrace SaaS; Managed follows another schedule and is not covered. The previous release is covered in &lt;a href="https://span.observer/en/dynatrace-saas-1-349-production-checks/" rel="noopener noreferrer"&gt;the 1.349 article&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;A note on honesty: span.observer has not tested 1.350 on a tenant. The facts come from the release notes; the checks below are suggestions, not measured results.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dynatrace SaaS 1.350: three breaking changes, three owners
&lt;/h2&gt;

&lt;p&gt;Give each breaking change a named owner and a date. The summary below restates what the page says.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvvlnvo4kvgj8w6wn3nic.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvvlnvo4kvgj8w6wn3nic.png" alt="YAML summary of the three Dynatrace SaaS 1.350 breaking changes: security events and the default\_security bucket, Billing Usage Events versioning, stricter dashboard validation" width="800" height="386"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The first two touch your queries. Billing Usage Events change their version format: &lt;code&gt;event.version&lt;/code&gt; becomes a three-part number such as &lt;code&gt;1.0.0&lt;/code&gt;. Any DQL query that filters on the old value must drop that filter, or it will stop returning what you expect.&lt;/p&gt;

&lt;p&gt;The security events change is the most distant but the most useful to prepare for. It is announced for SaaS 1.353, planned for December 2026. These events become configurable in OpenPipeline and move to a new &lt;code&gt;default_security&lt;/code&gt; bucket with a new default retention period. Act if you query &lt;code&gt;default_securityevents&lt;/code&gt; or &lt;code&gt;default_securityevents_builtin&lt;/code&gt;, or need to keep data beyond 35 days.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dashboards: stricter validation
&lt;/h2&gt;

&lt;p&gt;The third breaking change shows up on screen. Invalid &lt;code&gt;tiles.*.querySettings&lt;/code&gt; (query limits) and &lt;code&gt;tiles.*.davis&lt;/code&gt; (analyse and alert) settings are removed. The affected dashboards are mainly those created or changed through the API or an external tool.&lt;/p&gt;

&lt;p&gt;If you manage dashboards as code, with Terraform for example, review them before someone notices a tile without its limit. To recover a lost setting, the page points to Version history, then Edit JSON. For query examples inside tiles, see &lt;a href="https://span.observer/en/dynatrace-saas-dashboard-dql-examples/" rel="noopener noreferrer"&gt;our SaaS dashboard DQL examples&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;A second trap comes from the fixes: time series tiles with two metrics now assign the left and right axes by field order. An existing chart may therefore show swapped axes. It is not an outage, but a dashboard on the wall during an incident must stay readable.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to check it works
&lt;/h2&gt;

&lt;p&gt;Here are two checks you can do in ten minutes. I ran neither; adapt them to your environment.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhdmrbroia1e9q2781vgr.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhdmrbroia1e9q2781vgr.png" alt="DQL query on dt.system.events counting Billing Usage Events by event.version, then the two security bucket names to search for in saved queries" width="799" height="260"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; &lt;strong&gt;Billing&lt;/strong&gt;: list the &lt;code&gt;event.version&lt;/code&gt; values and find saved queries that still filter on them.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Security&lt;/strong&gt;: search your dashboards, notebooks and workflows for the two old bucket names. Each hit is a December task.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;Dashboards&lt;/strong&gt;: open the API-fed dashboards and compare query limits before and after.&lt;/li&gt;
&lt;li&gt; &lt;strong&gt;OTLP metrics&lt;/strong&gt;: if you enabled Advanced OTLP Metrics Dimensions, note that &lt;code&gt;dt.&lt;/code&gt;-prefixed metrics sent over OTLP are now dropped, according to the fixes list.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  What the release brings to operations
&lt;/h2&gt;

&lt;p&gt;Among the 32 feature updates, a few change an on-call engineer’s work more than the others.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Maintenance windows&lt;/strong&gt;: they can suppress problem creation for their scope. Handy against migration noise, as long as the scope is narrow. Window start and end events are also ingested into Grail.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Dashboards&lt;/strong&gt;: base queries shared across tiles, named collapsible rows (collapsed rows skip their queries) and PDF export through Print view, in Chromium-based browsers only. See also &lt;a href="https://span.observer/en/dql-smartscape-topology/" rel="noopener noreferrer"&gt;DQL and Smartscape&lt;/a&gt;, since &lt;code&gt;K8S_SERVICE&lt;/code&gt; nodes gain an &lt;code&gt;ip&lt;/code&gt; attribute.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Workflows&lt;/strong&gt;: actions to send an event to PagerDuty and an ITOM event to ServiceNow, with templates, plus create, read, update and delete actions for documents. Script limits and pitfalls: &lt;a href="https://span.observer/en/dynatrace-workflows-examples-javascript-limits/" rel="noopener noreferrer"&gt;Workflow examples&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Go&lt;/strong&gt;: static monitoring no longer needs ptrace on Linux; it will replace the opt-in «&amp;nbsp;Go improved static monitoring&amp;nbsp;» feature.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Permissions&lt;/strong&gt;: &lt;code&gt;storage:smartscape:read&lt;/code&gt; can be scoped by &lt;code&gt;storage:type&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;OpenPipeline&lt;/strong&gt;: forwarding to object storage covers more event types (security events, bizevents, Davis events and problems).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;One change touches the API: the OneAgent on-a-host auto-update API returns &lt;code&gt;400&lt;/code&gt; if the requested version is not higher than the installed one. A rollback script aiming at an older version will therefore fail.&lt;/p&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Separate what breaks (three changes), what looks different (axes, tiles) and what is optional (the rest).&lt;/li&gt;
&lt;li&gt;  Test the new Workflow actions on a low-stakes case before any PagerDuty or ServiceNow send in production.&lt;/li&gt;
&lt;li&gt;  Re-read the release notes before each rollout wave: the page is updated after publication.&lt;/li&gt;
&lt;li&gt;  Plan the move to &lt;code&gt;default_security&lt;/code&gt; now rather than the night before 1.353.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  When was Dynatrace SaaS 1.350 released?
&lt;/h3&gt;

&lt;p&gt;The page states a rollout start on 4 November 2026. It gives no date for each feature or each tenant.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is the security events change already active?
&lt;/h3&gt;

&lt;p&gt;No, according to the page: it is announced for 1.353, planned for December 2026. Release 1.350 is the warning to prepare your queries and retention.&lt;/p&gt;

&lt;h3&gt;
  
  
  Are my API-created dashboards at risk?
&lt;/h3&gt;

&lt;p&gt;Only if their query limit or Davis analysis settings are invalid. Those are then removed; Version history lets you recover them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;For 1.350, three actions are enough: drop version filters from billing queries, inventory the old security buckets before December, and review API-pushed dashboards. The rest can wait for a test. Sources: &lt;a href="https://docs.dynatrace.com/docs/whats-new/saas/sprint-350" rel="noopener noreferrer"&gt;Dynatrace SaaS 1.350 release notes&lt;/a&gt; and the &lt;a href="https://www.dynatrace.com/news/blog/whats-new-in-dynatrace-saas-version-1-350/" rel="noopener noreferrer"&gt;vendor’s blog post&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Dynatrace is a trademark of Dynatrace LLC. Versions and dates change: check them in the official documentation for your environment.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/dynatrace-saas-1-350-what-to-check/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/dynatrace-saas-1-350-quoi-verifier/" rel="noopener noreferrer"&gt;https://span.observer/fr/dynatrace-saas-1-350-quoi-verifier/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>dashboards</category>
      <category>dql</category>
      <category>dynatracesaas</category>
      <category>openpipeline</category>
    </item>
    <item>
      <title>Dynatrace Session Replay on Grail: see what the user experienced, with traces and logs alongside</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Mon, 16 Nov 2026 07:30:18 +0000</pubDate>
      <link>https://dev.to/spanobserver/dynatrace-session-replay-on-grail-see-what-the-user-experienced-with-traces-and-logs-alongside-i29</link>
      <guid>https://dev.to/spanobserver/dynatrace-session-replay-on-grail-see-what-the-user-experienced-with-traces-and-logs-alongside-i29</guid>
      <description>&lt;p&gt;«&amp;nbsp;Payment isn’t going through.&amp;nbsp;» The support ticket says nothing more. On the backend, no error spike; on the frontend, a few JavaScript exceptions among thousands of others. Without seeing the customer’s screen, the team is left guessing, and the ticket risks ending up as «&amp;nbsp;cannot reproduce&amp;nbsp;». This is exactly the case Dynatrace Session Replay was rebuilt for.&lt;/p&gt;

&lt;p&gt;Session recording tools have met this need for a long time, but they usually live on their own: you watch the video in one tool, look for the trace in another, and copy a request ID into a third. The new Dynatrace Session Replay, now available to all SaaS customers on Grail, changes the game: every recording is stored alongside the logs, traces, events and RUM data of the same platform.&lt;/p&gt;

&lt;p&gt;This article looks at what this brings in practice, how to get started, and what to watch out for if you still use Session Replay Classic.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dynatrace Session Replay: from film to investigation
&lt;/h2&gt;

&lt;p&gt;A replay on its own shows &lt;em&gt;what&lt;/em&gt; the user did. It does not say &lt;em&gt;why&lt;/em&gt; the button did not respond. The value of the new version lies in its foundation on Grail: a session is no longer a video file, it is a set of queryable events linked to the rest of the telemetry.&lt;/p&gt;

&lt;p&gt;In practice, this opens several investigation paths, in both directions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  from a session event to the matching distributed trace, span by span;&lt;/li&gt;
&lt;li&gt;  from a log line to the user sessions concerned;&lt;/li&gt;
&lt;li&gt;  from a backend incident to the sessions active at the same time;&lt;/li&gt;
&lt;li&gt;  from any DQL query to Notebooks, Dashboards, Workflows or Investigations.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;No more copying request IDs from one tool to another: the frontend-backend link is built in.&lt;/p&gt;

&lt;h3&gt;
  
  
  What the player shows
&lt;/h3&gt;

&lt;p&gt;During playback, the chronological list of events scrolls next to the recording: requests, errors, user actions, navigation. A timeline groups these events and lets you jump straight to a specific moment; a visual marker flags the key moments. The usual controls are there: playback speed, skipping inactive periods, going back a few seconds. Each event expands to show its detailed attributes, including the trace ID, and a search bar lets you navigate long sessions without watching them in full.&lt;/p&gt;

&lt;h3&gt;
  
  
  A summary by Dynatrace Intelligence
&lt;/h3&gt;

&lt;p&gt;The &lt;strong&gt;Summarize Session&lt;/strong&gt; button asks Dynatrace Intelligence to read the session’s RUM events and produce a plain-language summary: what the user experienced, the anomalies and errors detected, and what happened beneath the surface. For a twenty-minute session, it is often the quickest way to know whether it is worth watching.&lt;/p&gt;

&lt;h2&gt;
  
  
  Beyond the incident
&lt;/h2&gt;

&lt;p&gt;Extended retention of Session Replay data is available in preview. It opens up uses beyond immediate troubleshooting: handling customer complaints received weeks after the events, compliance audits, fraud detection. Because digital experience data stays in Grail, you can also analyse trends across session history with DQL, as you would for logs with &lt;a href="https://span.observer/en/log-pattern-analysis-dql/" rel="noopener noreferrer"&gt;log pattern analysis&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Another often underrated use: non-technical teams. Customer support and business teams readily use a replay to understand a complaint, provided access is properly scoped.&lt;/p&gt;

&lt;h2&gt;
  
  
  In practice
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Step 1 — Enable recording, ideally as configuration as code
&lt;/h3&gt;

&lt;p&gt;For a few applications, enabling it manually is enough. Beyond that, use Dynatrace configuration-as-code tooling to apply the same recording policy to all your frontend applications, without tuning each application individually. This is also the right time to review masking rules for user input: a replay must never expose a card number or sensitive personal data.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 2 — Find the right sessions
&lt;/h3&gt;

&lt;p&gt;Open the &lt;strong&gt;Users &amp;amp; Sessions&lt;/strong&gt; app and apply the filter for sessions with a replay, which are clearly flagged in the list. Facets let you narrow down by user ID, country, browser, operating system, frontends visited or number of errors. For a support ticket, the user ID and the time range are usually enough.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 3 — Go further with DQL
&lt;/h3&gt;

&lt;p&gt;When facets are not enough, query user events directly. Example: count the sessions of real users who reached a checkout page in the last 24 hours, excluding bots and synthetic tests.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;fetch user.events, from: now()-24h
| search "checkout"
| filter dt.rum.user_type == "real_user"
| summarize sessions = countDistinct(dt.rum.session.id)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;By replacing the last line with &lt;code&gt;dedup dt.rum.session.id&lt;/code&gt;, you get the list of distinct sessions, which you can open one by one. These results appear directly in a Notebook (see the &lt;a href="https://span.observer/en/dynatrace-notebooks-dashboards-dql/" rel="noopener noreferrer"&gt;training module on Notebooks and Dashboards&lt;/a&gt;), which keeps the analysis and its conclusions in the same place. The semantic dictionary in the documentation lists the available fields.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 4 — Pivot to the backend
&lt;/h3&gt;

&lt;p&gt;In the replay, find the failing action, expand the event, then follow the link to the distributed trace. You will see whether the request reached the backend, which service responded and with which error. If the request does not appear, the problem is most likely on the client side: that is already a diagnosis. For this link to work with services instrumented by OpenTelemetry, the chain must be complete: see &lt;a href="https://span.observer/en/opentelemetry-end-to-end-dynatrace/" rel="noopener noreferrer"&gt;end-to-end OpenTelemetry&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 5 — Migrate from Session Replay Classic
&lt;/h3&gt;

&lt;p&gt;If you use Session Replay Classic, your existing configuration stays active and data is not billed twice for double storage. However, data access permissions must be granted for the new experience. Run both in parallel during the transition, and use the opportunity to define who, between support and technical teams, should see what.&lt;/p&gt;

&lt;h3&gt;
  
  
  Best practices
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;  Validate masking rules with your DPO before extending recording, especially if extended retention is being considered.&lt;/li&gt;
&lt;li&gt;  Grant permissions in a targeted way: support needs replays, not necessarily application logs.&lt;/li&gt;
&lt;li&gt;  Use the session summary to triage, then watch the recording to confirm.&lt;/li&gt;
&lt;li&gt;  Always filter on &lt;code&gt;dt.rum.user_type&lt;/code&gt; to keep bots and synthetic tests out of your analyses.&lt;/li&gt;
&lt;li&gt;  User Interactions, which represent each interaction as a standalone event, are in Early Access and must be enabled separately: test them on a pilot application.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does Session Replay record what users type?
&lt;/h3&gt;

&lt;p&gt;Masking rules decide what is recorded. Configure them before extending recording, so that no card data or sensitive personal data ever appears in a replay.&lt;/p&gt;

&lt;h3&gt;
  
  
  Should we drop Session Replay Classic straight away?
&lt;/h3&gt;

&lt;p&gt;No. The existing configuration stays active and both can run side by side during the transition; you only need to grant data access permissions for the new experience.&lt;/p&gt;

&lt;h3&gt;
  
  
  How do I find the session of a customer who opened a ticket?
&lt;/h3&gt;

&lt;p&gt;In the Users &amp;amp; Sessions app, filter on sessions with a replay, then on the user ID and the ticket’s time range. For finer criteria, query user events with DQL.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;The new Session Replay turns session recording from visual evidence into the entry point of a full investigation. Because it is built on Grail, a session naturally links to the trace, log or host that explains the observed behaviour, and vice versa. For teams that juggled a replay tool and an observability platform, it changes daily work; for everyone else, it is a good reason to finally look at what their users see.&lt;/p&gt;

&lt;p&gt;Further reading: &lt;a href="https://www.dynatrace.com/news/blog/investigate-user-sessions-end-to-end-with-the-new-session-replay/" rel="noopener noreferrer"&gt;Investigate user sessions end-to-end with the new Session Replay&lt;/a&gt; (Dynatrace blog).&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Dynatrace is a trademark of Dynatrace LLC. SaaS features evolve quickly: check their availability in your environment and in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/dynatrace-session-replay-grail/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/session-replay-grail/" rel="noopener noreferrer"&gt;https://span.observer/fr/session-replay-grail/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>digitalexperience</category>
      <category>distributedtracing</category>
      <category>dql</category>
    </item>
    <item>
      <title>OpenTelemetry end-to-end: from user click to host, without stitching the pieces together</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Sun, 15 Nov 2026 21:30:22 +0000</pubDate>
      <link>https://dev.to/spanobserver/opentelemetry-end-to-end-from-user-click-to-host-without-stitching-the-pieces-together-2k2c</link>
      <guid>https://dev.to/spanobserver/opentelemetry-end-to-end-from-user-click-to-host-without-stitching-the-pieces-together-2k2c</guid>
      <description>&lt;p&gt;Many teams have adopted OpenTelemetry for a good reason: no longer depending on a proprietary agent to instrument their services. The project has now graduated from the CNCF and has become the common framework for producing traces, metrics and logs. But once the SDKs are in place and the Collector is deployed, one question soon comes back: why is investigating still so laborious? That is the whole point of taking OpenTelemetry end-to-end, from the user’s click to the host.&lt;/p&gt;

&lt;p&gt;The answer fits in one sentence: OpenTelemetry standardises the &lt;strong&gt;shape&lt;/strong&gt; of signals, not the &lt;strong&gt;relationships&lt;/strong&gt; between them. A trace knows which service it comes from; it does not know which host that service runs on, which user session triggered it, or which AI model it calls downstream. Without that context, the on-call engineer makes the connections by hand, across several tabs.&lt;/p&gt;

&lt;p&gt;Dynatrace SaaS tackles the problem from the other end: it accepts OTel telemetry as is and automatically builds the topology that connects it. Let us look at what this brings in practice and which settings determine the result.&lt;/p&gt;

&lt;h2&gt;
  
  
  Four seams to stitch back together
&lt;/h2&gt;

&lt;p&gt;In an architecture instrumented with OpenTelemetry, context breaks almost always happen in the same places.&lt;/p&gt;

&lt;h3&gt;
  
  
  Between the browser and the first service
&lt;/h3&gt;

&lt;p&gt;Dynatrace frontend monitoring (web, mobile, cross-platform) connects to OTel-instrumented backends through standard W3C propagation, i.e. the &lt;code&gt;traceparent&lt;/code&gt; and &lt;code&gt;tracestate&lt;/code&gt; headers. No additional instrumentation is needed along the backend chain. You can then start from a session in &lt;strong&gt;User Events and Sessions&lt;/strong&gt; (including its &lt;a href="https://span.observer/en/dynatrace-session-replay-grail/" rel="noopener noreferrer"&gt;replay on Grail&lt;/a&gt;), open the associated trace in &lt;strong&gt;Distributed Tracing&lt;/strong&gt;, or go the other way from a backend span to the user action that caused it. The condition, and it is often forgotten: the trace context must actually be propagated all the way to the service.&lt;/p&gt;

&lt;h3&gt;
  
  
  Between services
&lt;/h3&gt;

&lt;p&gt;The cloud-native &lt;strong&gt;Service Map&lt;/strong&gt; continuously discovers connections at runtime: upstream and downstream services, database queries, messaging flows, with health status overlaid. Service-to-database mapping is built from OTel data, with no manual entity modelling; it relies on dependencies actually observed rather than on tags.&lt;/p&gt;

&lt;h3&gt;
  
  
  Between the application and the infrastructure
&lt;/h3&gt;

&lt;p&gt;For teams already running a Collector on their hosts, host and process metrics feed the &lt;strong&gt;Infrastructure &amp;amp; Operations&lt;/strong&gt; app, with correlated logs and alert templates. Infrastructure entities join Smartscape, which makes it quick to tell an application problem from a platform problem. The documentation lists the supported Collector distributions; they must include the &lt;code&gt;hostmetrics&lt;/code&gt; receiver and the &lt;code&gt;resourcedetection&lt;/code&gt; processor.&lt;/p&gt;

&lt;h3&gt;
  
  
  Between classic services and AI
&lt;/h3&gt;

&lt;p&gt;Community conventions for AI, such as OpenInference and &lt;a href="https://span.observer/en/langfuse-openllmetry-llm-observability/" rel="noopener noreferrer"&gt;OpenLLMetry&lt;/a&gt;, which belong to the OpenTelemetry ecosystem, allow Dynatrace to place models, agents and agent services in Smartscape, next to traditional services. Captured production telemetry can then serve as input to online evaluations of response quality, safety and performance, depending on the chosen configuration.&lt;/p&gt;

&lt;h2&gt;
  
  
  OneAgent or OpenTelemetry: no need to choose
&lt;/h2&gt;

&lt;p&gt;Smartscape treats OneAgent and OpenTelemetry as equivalent sources: an OTel-instrumented service appears with the same depth as a service observed by OneAgent. That is a practical argument for hybrid environments, which are very common: one team keeps OneAgent on its legacy VMs, another instruments its microservices with OTel, and the topology stays unified.&lt;/p&gt;

&lt;p&gt;A note on scope: the Service Map, the frontend-backend link and host monitoring through the Collector are offered to customers on the Dynatrace Platform Subscription (DPS).&lt;/p&gt;

&lt;h2&gt;
  
  
  In practice: OpenTelemetry end-to-end in Dynatrace
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Step 1 — Check propagation on the frontend side
&lt;/h3&gt;

&lt;p&gt;When the frontend calls an API on another domain, the browser only sends trace headers if the API explicitly allows them. Check your APIs’ CORS response:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight http"&gt;&lt;code&gt;&lt;span class="nc"&gt;Access&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Control&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Allow&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Headers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;traceparent&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;tracestate&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;content&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="k"&gt;type&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Also check that your gateways and reverse proxies do not strip these headers. A session that leads to no backend trace almost always comes down to this.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 2 — Complete the host Collector
&lt;/h3&gt;

&lt;p&gt;Here is a minimal configuration that adds host metrics and resource detection, then exports to the environment’s OTLP endpoint. Dynatrace expects metrics with delta temporality, hence the conversion processor:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;receivers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;hostmetrics&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;collection_interval&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;60s&lt;/span&gt;
    &lt;span class="na"&gt;scrapers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;cpu&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
      &lt;span class="na"&gt;memory&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
      &lt;span class="na"&gt;disk&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
      &lt;span class="na"&gt;filesystem&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
      &lt;span class="na"&gt;network&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
      &lt;span class="na"&gt;process&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="na"&gt;processors&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;resourcedetection&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;detectors&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s"&gt;env&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s"&gt;system&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
  &lt;span class="na"&gt;cumulativetodelta&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="na"&gt;batch&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="na"&gt;exporters&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;otlphttp&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;endpoint&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;https://&amp;lt;environment-id&amp;gt;.live.dynatrace.com/api/v2/otlp&lt;/span&gt;
    &lt;span class="na"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;Authorization&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Api-Token ${env:DT_API_TOKEN}"&lt;/span&gt;
&lt;span class="na"&gt;service&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;pipelines&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;metrics&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;receivers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s"&gt;hostmetrics&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
      &lt;span class="na"&gt;processors&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s"&gt;resourcedetection&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s"&gt;cumulativetodelta&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s"&gt;batch&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
      &lt;span class="na"&gt;exporters&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s"&gt;otlphttp&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The token must have the ingest scopes matching the signals sent. Adapt the detectors to your context (cloud, Kubernetes) using the Collector documentation.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 3 — Check that spans arrive in Grail
&lt;/h3&gt;

&lt;p&gt;Before judging the topology, make sure spans arrive with the right resource attributes. In a Notebook (for a syntax refresher, see &lt;a href="https://span.observer/en/dql-fundamentals-tutorial/" rel="noopener noreferrer"&gt;DQL fundamentals&lt;/a&gt;):&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;fetch spans, from: now()-30m
| summarize spans = count(), by: { service.name }
| sort spans desc
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A missing service, or one saddled with a generic name, signals a &lt;code&gt;service.name&lt;/code&gt; not set in the SDK or the instrumentation configuration.&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 4 — Walk the full chain
&lt;/h3&gt;

&lt;p&gt;Pick a real user journey, for example a slow product search. Start from the session, open the trace, identify the service and its database query in the Service Map, then move up to the host in Infrastructure &amp;amp; Operations. If a link is missing, you know exactly where to step in. The Dynatrace Playground lets you do this exercise on demo data before touching your own environment.&lt;/p&gt;

&lt;h3&gt;
  
  
  Best practices
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;  Follow the OpenTelemetry semantic conventions for resource attributes: entity construction relies on them.&lt;/li&gt;
&lt;li&gt;  Test W3C propagation end to end, including CORS and proxies, before concluding that a frontend-backend link «&amp;nbsp;does not work&amp;nbsp;».&lt;/li&gt;
&lt;li&gt;  Keep a single, well-documented Collector pipeline per host type rather than diverging variants.&lt;/li&gt;
&lt;li&gt;  In mixed environments, let OneAgent and OTel coexist: the topology reconciles them.&lt;/li&gt;
&lt;li&gt;  For AI workloads, adopt a common convention (OpenInference or OpenLLMetry) from the start rather than home-grown attributes.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Do you have to choose between OneAgent and OpenTelemetry?
&lt;/h3&gt;

&lt;p&gt;No. Smartscape treats both as equivalent sources: an OTel-instrumented service appears with the same depth as a service observed by OneAgent, and the topology stays unified in a hybrid environment.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why does a user session lead to no backend trace?
&lt;/h3&gt;

&lt;p&gt;Almost always because the W3C context is not propagated: the API called on another domain does not allow the &lt;code&gt;traceparent&lt;/code&gt; and &lt;code&gt;tracestate&lt;/code&gt; headers in its CORS response, or a gateway strips them along the way.&lt;/p&gt;

&lt;h3&gt;
  
  
  Which Collector distribution should you use for host metrics?
&lt;/h3&gt;

&lt;p&gt;One of the distributions listed in the Dynatrace documentation, as long as it includes the &lt;code&gt;hostmetrics&lt;/code&gt; receiver and the &lt;code&gt;resourcedetection&lt;/code&gt; processor. Add the conversion to delta temporality, which Dynatrace expects for metrics.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;OpenTelemetry has settled the question of format; the value now lies in connecting the signals. By automatically building the topology from OTel telemetry, from the browser to the host and AI components, Dynatrace SaaS avoids the manual stitching that slows down every incident. The remaining work is mostly hygiene: propagate context, name resources correctly and equip the Collector with the right components.&lt;/p&gt;

&lt;p&gt;Further reading: &lt;a href="https://www.dynatrace.com/news/blog/connect-opentelemetry-signals-end-to-end-with-dynatrace/" rel="noopener noreferrer"&gt;Connect OpenTelemetry signals end-to-end with Dynatrace&lt;/a&gt; (Dynatrace blog).&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with Dynatrace LLC. Dynatrace is a trademark of Dynatrace LLC. SaaS features evolve quickly: check their availability in your environment and in the official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/opentelemetry-end-to-end-dynatrace/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/opentelemetry-bout-en-bout-dynatrace/" rel="noopener noreferrer"&gt;https://span.observer/fr/opentelemetry-bout-en-bout-dynatrace/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>distributedtracing</category>
      <category>dynatrace</category>
      <category>dynatracesaas</category>
      <category>grail</category>
    </item>
    <item>
      <title>AgentLens AI agent audit trail: hash chaining, OpenTelemetry and MCP</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Sun, 15 Nov 2026 19:30:14 +0000</pubDate>
      <link>https://dev.to/spanobserver/agentlens-ai-agent-audit-trail-hash-chaining-opentelemetry-and-mcp-33i</link>
      <guid>https://dev.to/spanobserver/agentlens-ai-agent-audit-trail-hash-chaining-opentelemetry-and-mcp-33i</guid>
      <description>&lt;p&gt;An AI agent deleted a file, called a tool and burned through tokens. Six months later, who can prove that last night’s log was not edited? The AgentLens AI agent audit trail answers with a hash-chained journal. The &lt;a href="https://github.com/agentkitai/agentlens" rel="noopener noreferrer"&gt;agentkitai/agentlens repository&lt;/a&gt; is small and young. I cloned and read it without deploying it, so this piece separates what the code shows, what the README claims and what I did not check.&lt;/p&gt;

&lt;p&gt;The project is worth a look for one reason. Most agent observability tools measure. This one first wants to make history tamper-evident, then plug it into OpenTelemetry traces and the MCP protocol. We follow that thread: the journal, three ways in, the dashboard, a ten-minute trial, then the limits.&lt;/p&gt;

&lt;p&gt;Facts as of 9 November 2026. The repository is MIT-licensed (I read the LICENSE file, author Amit Paz). The 50 latest commits I fetched run from 2 July to 1 August 2026. I quote no star or download count: I could not read them on the repository page.&lt;/p&gt;

&lt;h2&gt;
  
  
  AgentLens AI agent audit trail: a journal that checks itself
&lt;/h2&gt;

&lt;p&gt;The README calls AgentLens a flight recorder for agents. Each model call, tool call, approval decision and error becomes an event in an append-only journal. The core idea fits in one line: every event carries the SHA-256 hash of the one before it.&lt;/p&gt;

&lt;p&gt;The code in &lt;code&gt;packages/core/src/hash.ts&lt;/code&gt; backs this up. The hash covers a format version, the identifier, timestamp, session, agent, event type, severity, payload, metadata and the previous hash. A session starts from &lt;code&gt;null&lt;/code&gt;. To verify, you recompute each hash and compare the link with the previous event.&lt;/p&gt;

&lt;p&gt;In practice, editing a row, deleting one or swapping two breaks verification at the exact spot. Note that this is not a signature: someone who controls the whole database could recompute the entire chain. The project mentions a signed export for auditors (&lt;code&gt;GET /api/audit/verify/export&lt;/code&gt;); I did not test it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhzzyuey93ee22qibdvfc.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhzzyuey93ee22qibdvfc.png" alt="Excerpt from the AgentLens README: demo/aha.sh script, chain valid then broken after one event is altered" width="800" height="281"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Source : &lt;a href="https://github.com/agentkitai/agentlens/blob/main/README.md" rel="noopener noreferrer"&gt;github.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The capture above reproduces the output the README gives for &lt;code&gt;./demo/aha.sh&lt;/code&gt;. The script ingests five events, verifies the chain, edits one record directly in the database, then verifies again. I did not run it: this is the output the project advertises.&lt;/p&gt;

&lt;h2&gt;
  
  
  Three ways to feed the audit trail
&lt;/h2&gt;

&lt;h3&gt;
  
  
  OpenTelemetry, no SDK
&lt;/h3&gt;

&lt;p&gt;If your agent already emits spans following the OpenTelemetry GenAI conventions, point the OTLP exporter at the server, which listens on port 3400:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:3400
export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://localhost:3400/v1/traces
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;According to the README, &lt;code&gt;chat&lt;/code&gt; spans become call and response pairs, &lt;code&gt;execute_tool&lt;/code&gt; becomes a tool call and &lt;code&gt;invoke_agent&lt;/code&gt; an agent event. One trace maps to one session. Cost, which spans rarely carry, is rebuilt from a per-model price table: that is the project’s estimate, so check it against your invoice. In production, the &lt;code&gt;OTLP_AUTH_TOKEN&lt;/code&gt; variable requires a token on these routes.&lt;/p&gt;

&lt;h3&gt;
  
  
  The MCP server, on the agent side
&lt;/h3&gt;

&lt;p&gt;The &lt;code&gt;@agentkitai/agentlens-mcp&lt;/code&gt; package goes into the agent’s MCP configuration, for instance Claude Desktop or Cursor. The &lt;code&gt;docs/guide/mcp-integration.md&lt;/code&gt; guide describes a stdio transport and tools such as &lt;code&gt;agentlens_session_start&lt;/code&gt;, &lt;code&gt;agentlens_log_event&lt;/code&gt; and &lt;code&gt;agentlens_log_llm_call&lt;/code&gt;. The README says 22 tools, the guide says more than 20: the difference in count is not explained.&lt;/p&gt;

&lt;p&gt;A reliability caveat applies. Here the agent itself decides to call the logging tool. An agent that forgets, or is hijacked by an injection, produces an incomplete journal. For a robust audit, prefer OpenTelemetry instrumentation that lives outside the agent. MCP tool risks are covered in our article on &lt;a href="https://span.observer/en/mcp-tool-poisoning-detection/" rel="noopener noreferrer"&gt;MCP tool poisoning&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  The Python and TypeScript SDKs
&lt;/h3&gt;

&lt;p&gt;To read sessions and analytics, &lt;code&gt;pip install agentlensai&lt;/code&gt; or &lt;code&gt;npm install @agentkitai/agentlens-sdk&lt;/code&gt; give you a client with an API key. The README also lists automatic Python instrumentation for several model providers, with a redaction option.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the dashboard shows
&lt;/h2&gt;

&lt;p&gt;The session detail view shows a timeline: tool calls, an error, tracked cost, session end. A Chain Valid badge shows the chain state. The capture comes from the repository (demo folder, MIT licence) and shows demo data: a 45-second Research Agent session costing 0.3420 dollars.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxn6bgpmxwjjnforqx3ro.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxn6bgpmxwjjnforqx3ro.png" alt="AgentLens session detail: timeline of nine events, one tool error, cost breakdown and Chain Valid badge" width="800" height="862"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Source : &lt;a href="https://github.com/agentkitai/agentlens/blob/main/README.md" rel="noopener noreferrer"&gt;github.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The Health page gives each agent a score out of 100 built from several dimensions. I did not read the formula, so I do not judge it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmu4wasxe37c4vxj7clfm.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmu4wasxe37c4vxj7clfm.png" alt="AgentLens Health Overview page: code-reviewer and data-pipeline agents with health scores out of 100" width="800" height="513"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Source : &lt;a href="https://github.com/agentkitai/agentlens/blob/main/README.md" rel="noopener noreferrer"&gt;github.com&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Trying AgentLens in ten minutes
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;p&gt;Install Docker, then run the image the project publishes. According to the README, &lt;code&gt;AUTH_DISABLED=true&lt;/code&gt; is for local trials only.&lt;br&gt;
&lt;/p&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;docker run -p 3400:3400 -e AUTH_DISABLED=true -e JWT_SECRET=dev-secret ghcr.io/agentkitai/agentlens
&lt;/code&gt;&lt;/pre&gt;

&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Open &lt;code&gt;http://localhost:3400&lt;/code&gt;. Without Docker, the README offers &lt;code&gt;npx @agentkitai/agentlens-server&lt;/code&gt; (Node.js 20 or later).&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;To see the chain break, clone the repository and run &lt;code&gt;./demo/aha.sh&lt;/code&gt; (Docker required).&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Then point an instrumented agent at port 3400 with the OTLP variables above.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;For a shared setup, the repository ships a Docker Compose file with Postgres, Redis, authentication and TLS. By default SQLite is enough to test.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limits and open questions
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Maturity.&lt;/strong&gt; The repository is young and only one author appears in the licence. I could not read the stars or the issue pace.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Compliance.&lt;/strong&gt; The README cites Article 12 of the EU AI Act. That is not a certification: a hash chain helps record-keeping, it replaces neither legal analysis nor a retention policy.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Integrity.&lt;/strong&gt; Without an external anchor for the head hash, a fully rewritten database stays undetectable.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Data.&lt;/strong&gt; Prompts and outputs can contain secrets. Turn on redaction and encrypt the disk.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Dynatrace.&lt;/strong&gt; Nothing links AgentLens to Dynatrace. It receives OTLP, but it is neither an exporter nor an official integration.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Instrument the agent from outside (OpenTelemetry) rather than asking it to log itself.&lt;/li&gt;
&lt;li&gt;  Record the head hash regularly in a third-party system to anchor the chain.&lt;/li&gt;
&lt;li&gt;  Keep the same conversation identifier in your APM and the audit trail so you can correlate traces and evidence.&lt;/li&gt;
&lt;li&gt;  Turn off &lt;code&gt;AUTH_DISABLED&lt;/code&gt; as soon as the server leaves your workstation.&lt;/li&gt;
&lt;li&gt;  Tie spend to alerts: see &lt;a href="https://span.observer/en/ai-agent-cost-tokens-opentelemetry/" rel="noopener noreferrer"&gt;the token cost of an AI agent&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For local coding-agent monitoring, compare with &lt;a href="https://span.observer/en/observal-ai-coding-agent-observability/" rel="noopener noreferrer"&gt;Observal&lt;/a&gt;, and for the limits to impose on agents, read &lt;a href="https://span.observer/en/ai-agent-guardrails-production/" rel="noopener noreferrer"&gt;guardrails for AI agents in production&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does AgentLens replace an observability tool such as Dynatrace?
&lt;/h3&gt;

&lt;p&gt;No. It keeps a record of what an agent did and shows that the record has not changed. SLOs, infrastructure correlation and root cause analysis stay with your observability platform.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does the hash chain prevent tampering?
&lt;/h3&gt;

&lt;p&gt;It makes tampering detectable, not impossible. Whoever rewrites the whole chain produces a consistent journal. Anchor the head hash outside the database to close that door.&lt;/p&gt;

&lt;h3&gt;
  
  
  Can I use it in production?
&lt;/h3&gt;

&lt;p&gt;That is your call: the project is young and I have not stress-tested it. Start with a non-critical agent, with authentication, redaction and backups.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;AgentLens offers a simple, useful idea: an agent journal you can verify again. Read it for the hash-chain pattern, try it with the demo, and do not promote it to production before testing its limits. For the span attributes to emit, see the &lt;a href="https://opentelemetry.io/docs/specs/semconv/gen-ai/" rel="noopener noreferrer"&gt;OpenTelemetry GenAI conventions&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This blog is independent and not affiliated with Dynatrace LLC. AgentLens belongs to its authors. The captures come from the project repository (MIT licence); I did not run the software.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/agentlens-ai-agent-audit-trail/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/agentlens-audit-agents-ia/" rel="noopener noreferrer"&gt;https://span.observer/fr/agentlens-audit-agents-ia/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>agentlens</category>
      <category>aiagents</category>
      <category>aicompliance</category>
      <category>aiobservability</category>
    </item>
    <item>
      <title>AgentOps Center: running multi-agent systems with OpenTelemetry and SigNoz</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Sun, 15 Nov 2026 17:30:14 +0000</pubDate>
      <link>https://dev.to/spanobserver/agentops-center-running-multi-agent-systems-with-opentelemetry-and-signoz-5c80</link>
      <guid>https://dev.to/spanobserver/agentops-center-running-multi-agent-systems-with-opentelemetry-and-signoz-5c80</guid>
      <description>&lt;p&gt;A multi-agent workflow takes 45 seconds to answer, no exception was thrown, and every dashboard is green. What do you do next? That operations question is what AgentOps Center sets out to answer. It is a small &lt;a href="https://github.com/swapitsneil/AgentOps-Center" rel="noopener noreferrer"&gt;GitHub project&lt;/a&gt; built for a SigNoz hackathon. I read it file by file without deploying it, so this article keeps three things apart: what the code does, what the README claims, and what I could not check.&lt;/p&gt;

&lt;p&gt;The repository is more useful as a pattern than as a product. It shows how to name spans in a system of cooperating agents, where to send them, and how to make an assistant’s answer depend on real telemetry. We will walk through those three decisions, then the limits.&lt;/p&gt;

&lt;p&gt;A note on method: the last commit I read is dated 26 July 2026. I quote no star count because I did not read one from the source. The README says MIT, yet I found no LICENSE file in the tree, so check before reusing anything.&lt;/p&gt;

&lt;h2&gt;
  
  
  The problem AgentOps Center targets: slow is not broken
&lt;/h2&gt;

&lt;p&gt;Classic APM counts requests and HTTP errors. An agent workflow behaves differently. It loops, replays a tool, or waits on a model provider that is rate limiting it. Nothing breaks, everything slows down, and the token bill climbs.&lt;/p&gt;

&lt;p&gt;The README lists the questions operators need to settle: which node looped, which tool returned a rate limit, what each sub-agent cost. Answering them takes spans per agent, per model call and per tool, all linked in one trace. That is what the code builds.&lt;/p&gt;

&lt;h2&gt;
  
  
  Decision 1: how to name spans in a multi-agent system
&lt;/h2&gt;

&lt;p&gt;The file &lt;code&gt;backend/instrumentation/agent_spans.py&lt;/code&gt; declares attribute constants. It reuses the &lt;code&gt;gen_ai.*&lt;/code&gt; keys for model calls, then adds house keys for agents and tools.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvmycvfat81uc9byvkwc3.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvmycvfat81uc9byvkwc3.png" alt="Excerpt of agent\_spans.py: constants gen\_ai.system, gen\_ai.request.model, gen\_ai.usage.input\_tokens, then agent.name, agent.workflow\_id and tool.name" width="800" height="532"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The split is sensible. Standard keys let you compare models and providers, while house keys carry business logic: which agent, which workflow. A workflow identifier on every span then groups everything together, even when the spans come from different libraries. To see what a span carries, read &lt;a href="https://span.observer/en/anatomy-of-an-opentelemetry-span/" rel="noopener noreferrer"&gt;the anatomy of an OpenTelemetry span&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Two cautions apply:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  The OpenTelemetry GenAI conventions are moving quickly. The specification page now points to a dedicated repository that I could not read (the server returned an error when I checked). &lt;code&gt;gen_ai.system&lt;/code&gt; may have been replaced by a newer provider attribute: unverified.&lt;/li&gt;
&lt;li&gt;  &lt;code&gt;gen_ai.usage.cost_usd&lt;/code&gt; is the project’s own invention, not a key I know to be standard. Dashboards built on it will stay specific to your code.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;One more design detail: &lt;code&gt;gen_ai.prompt&lt;/code&gt; and &lt;code&gt;gen_ai.completion&lt;/code&gt; are cut to 500 characters. Full prompts in spans are expensive and expose data, so truncation is a fair compromise. Still, set your own policy before production. See also &lt;a href="https://span.observer/en/opentelemetry-attributes-dynatrace/" rel="noopener noreferrer"&gt;OpenTelemetry attributes in Dynatrace&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Decision 2: where the telemetry goes
&lt;/h2&gt;

&lt;p&gt;The &lt;code&gt;setup.py&lt;/code&gt; module creates a tracer provider and a meter provider, exporting over OTLP gRPC to &lt;code&gt;localhost:4317&lt;/code&gt; by default, which &lt;code&gt;OTEL_EXPORTER_OTLP_ENDPOINT&lt;/code&gt; overrides. It turns on FastAPI and httpx instrumentation, plus LangChain through OpenInference, which also covers LangGraph according to a code comment. Sampling is fixed at 100 %, flagged in the code as a demo setting.&lt;/p&gt;

&lt;p&gt;The repository’s Collector receives OTLP and writes straight to ClickHouse using SigNoz exporters, one pipeline per signal.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwld57gsn62wjvc267pf3.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwld57gsn62wjvc267pf3.png" alt="Excerpt of otel-collector/config.yaml: exporters clickhousetraces, signozclickhousemetrics and clickhouselogsexporter, then the traces pipeline with memory\_limiter and batch" width="800" height="428"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This layout is for self-hosted SigNoz. With another backend, Dynatrace for instance, only the exporter changes: the OTLP protocol and the attributes stay. Our &lt;a href="https://span.observer/en/opentelemetry-collector-dynatrace-setup/" rel="noopener noreferrer"&gt;OpenTelemetry Collector to Dynatrace guide&lt;/a&gt; shows that variant. I have not tested this repository against Dynatrace.&lt;/p&gt;

&lt;p&gt;For operations, note the &lt;code&gt;memory_limiter&lt;/code&gt; and the batch of 10,000 items every 10 seconds. Good instinct: an agent system can emit thousands of spans per run. The README’s smoke tests report between 2,340 and 3,559 spans per run.&lt;/p&gt;

&lt;h2&gt;
  
  
  Decision 3: let an assistant diagnose, but only from evidence
&lt;/h2&gt;

&lt;p&gt;This is the most interesting idea. The Root Cause Copilot does not ask a model cold. An evidence engine first calls the SigNoz MCP server to fetch traces, error logs, metrics and alerts. It then builds a context labelled by source tool and grades its own confidence.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://signoz.io/docs/ai/signoz-mcp-server/" rel="noopener noreferrer"&gt;SigNoz documentation&lt;/a&gt; confirms that this MCP server exposes &lt;code&gt;signoz_search_traces&lt;/code&gt;, &lt;code&gt;signoz_search_logs&lt;/code&gt;, &lt;code&gt;signoz_query_metrics&lt;/code&gt; and &lt;code&gt;signoz_list_alert_rules&lt;/code&gt;, among others. HTTP mode is switched on with &lt;code&gt;TRANSPORT_MODE=http&lt;/code&gt;, and the endpoint is &lt;code&gt;/mcp&lt;/code&gt;. For the wider picture, read &lt;a href="https://span.observer/en/monitor-mcp-agents-opentelemetry/" rel="noopener noreferrer"&gt;monitoring MCP agents with OpenTelemetry&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The grading logic lives in &lt;code&gt;reasoning.py&lt;/code&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  MCP server unreachable: LOW confidence, with in-memory local context only;&lt;/li&gt;
&lt;li&gt;  traces plus another signal (logs, metrics or alerts): HIGH;&lt;/li&gt;
&lt;li&gt;  a single signal: MEDIUM;&lt;/li&gt;
&lt;li&gt;  no usable data: NONE.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A critical reading: the rule counts signals, it does not measure correctness. One condition in the code even grades HIGH whenever the MCP server answers with at least two signals of any kind. The badge says «&amp;nbsp;I found material to back my answer&amp;nbsp;», not «&amp;nbsp;my answer is right&amp;nbsp;».&lt;/p&gt;

&lt;h2&gt;
  
  
  Testing observability with injected faults
&lt;/h2&gt;

&lt;p&gt;The chaos engine (&lt;code&gt;chaos/injector.py&lt;/code&gt;) draws faults at random, at an intensity between 0 and 1: timeouts, 429 or 500 errors, tool failures, a 2 to 8 second slowdown, agent crashes, and corrupted output. The README lists five, while the code has six, adding invalid output. Each fault creates an error span carrying &lt;code&gt;chaos.mode&lt;/code&gt; and &lt;code&gt;chaos.injected&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;For an SRE the value is clear: check that your alerts and dashboards see the degradation before a real one arrives. The &lt;code&gt;chaos.injected&lt;/code&gt; marker also stops you mistaking a drill for an incident. For the matching safeguards, read &lt;a href="https://span.observer/en/ai-agent-guardrails-production/" rel="noopener noreferrer"&gt;AI agent guardrails in production&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is real and what is scripted
&lt;/h2&gt;

&lt;p&gt;Here is the main limit. The workflow’s incidents are staged. In &lt;code&gt;graph.py&lt;/code&gt;, depending on words found in the scenario text, the code emits Redis, SQL and third-party API spans with hard-coded values: 48 connections out of 50, a 429 status from a payment service, and waits faked with &lt;code&gt;asyncio.sleep&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3l7pd6520o7x8zmk128y.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3l7pd6520o7x8zmk128y.png" alt="Excerpt of graph.py: redis\_span, db\_span and third\_party\_api\_span emitted from scenario keywords, with active\_connections=48 and status\_code=429 written in the code" width="800" height="386"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;That is fine for a demonstration, but it changes how to read the README screenshots. They prove the telemetry pipeline works, not that the tool detects real outages. Model pricing is also a table inside the code (four models, a default price otherwise). For genuine cost tracking, see &lt;a href="https://span.observer/en/ai-agent-cost-tokens-opentelemetry/" rel="noopener noreferrer"&gt;AI agent cost per token with OpenTelemetry&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The README itself admits further gaps in its improvement list: no authentication or multi-tenancy, no built-in alerting, no long-term retention. I did not reuse the README screenshots here, because the repository has no licence file.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reusing the pattern: step by step
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; Define your attribute keys: a workflow identifier, an agent name, a tool name, and the &lt;code&gt;gen_ai.*&lt;/code&gt; keys for models.&lt;/li&gt;
&lt;li&gt; Wrap every model call and every tool in a span, setting the error status and recording the exception.&lt;/li&gt;
&lt;li&gt; Send everything over OTLP to a Collector, with &lt;code&gt;memory_limiter&lt;/code&gt; and batching.&lt;/li&gt;
&lt;li&gt; To run it locally, the README says to copy &lt;code&gt;.env.example&lt;/code&gt;, supply a Groq or OpenAI key, then run &lt;code&gt;docker compose up -d&lt;/code&gt; (nine containers are announced).&lt;/li&gt;
&lt;li&gt; Check the services: &lt;code&gt;localhost:3000&lt;/code&gt; (UI), &lt;code&gt;8000&lt;/code&gt; (API) and &lt;code&gt;8080&lt;/code&gt; (SigNoz), according to the README.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;To see how a graph of agents turns into a trace in code, our article on &lt;a href="https://span.observer/en/langgraph-agents-opentelemetry-tracing/" rel="noopener noreferrer"&gt;LangChain and LangGraph tracing&lt;/a&gt; covers the instrumentation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Best practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Put a workflow identifier on every span, and in the logs too.&lt;/li&gt;
&lt;li&gt;  Use standard keys for what you compare and documented house keys for the rest.&lt;/li&gt;
&lt;li&gt;  Sample in production: 100 % suits a demo, not a fleet of agents.&lt;/li&gt;
&lt;li&gt;  Never put full prompts or secrets in attributes.&lt;/li&gt;
&lt;li&gt;  Mark injected faults so they cannot be confused with real incidents.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Does AgentOps Center work with Dynatrace?
&lt;/h3&gt;

&lt;p&gt;It is written for SigNoz: the Collector exports to ClickHouse and the copilot queries the SigNoz MCP server. The spans themselves are standard OpenTelemetry and could go elsewhere, but I have not tested that route.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does the confidence badge guarantee a good diagnosis?
&lt;/h3&gt;

&lt;p&gt;No. It measures how much telemetry was found, not whether the conclusion is correct. Always read the traces it cites.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is it production ready?
&lt;/h3&gt;

&lt;p&gt;The README presents it as a hackathon demonstration and lists authentication, alerting and retention among future work. Treat it as a source of ideas.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;AgentOps Center answers a real operations question with a clear pattern: well-named spans, an OTLP pipeline, and an assistant forced to cite its sources. Its incidents are staged and the code is hackathon code, but the architecture reads in an hour and transposes well. Take the ideas, not the code as it stands.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with the AgentOps Center project, its authors, SigNoz or Dynatrace LLC. Trademarks belong to their respective owners. Tools change quickly: check commands and options in their official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/agentops-center-multi-agent-observability-signoz/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/agentops-center-multi-agents-signoz/" rel="noopener noreferrer"&gt;https://span.observer/fr/agentops-center-multi-agents-signoz/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>agentopscenter</category>
      <category>aiagents</category>
      <category>aiobservability</category>
      <category>chaosengineering</category>
    </item>
    <item>
      <title>Observal AI coding agent observability: sessions and traces</title>
      <dc:creator>Span Observer</dc:creator>
      <pubDate>Sun, 15 Nov 2026 15:30:13 +0000</pubDate>
      <link>https://dev.to/spanobserver/observal-ai-coding-agent-observability-sessions-and-traces-56j5</link>
      <guid>https://dev.to/spanobserver/observal-ai-coding-agent-observability-sessions-and-traces-56j5</guid>
      <description>&lt;p&gt;A developer tells you: «&amp;nbsp;the agent ran for an hour, burned a pile of tokens and shipped nothing.&amp;nbsp;» There is no error code and no stack trace, only a conversation. For that case, Observal AI coding agent observability means keeping the session itself, which is the ground Observal covers: an open source repository (Apache 2.0) showing about 4.3k stars on the GitHub page I read on 9 November 2026. This post approaches it through the operational decision it forces on you: what to keep from an agent session, where, and for whom.&lt;/p&gt;

&lt;p&gt;A note on method: span.observer did not deploy Observal. Everything below comes from the README, the repository documentation and the screenshots it publishes; anything I could not read is marked «&amp;nbsp;unverified&amp;nbsp;».&lt;/p&gt;

&lt;h2&gt;
  
  
  Observal AI coding agent observability: what is it?
&lt;/h2&gt;

&lt;p&gt;First trap: the name moved. The README now calls itself «&amp;nbsp;Agentic Resource Registry&amp;nbsp;» and states «&amp;nbsp;Formerly Observal: same code, same maintainers, same Apache 2.0 license&amp;nbsp;». The command is still &lt;code&gt;observal&lt;/code&gt; and old links redirect. Today the project describes itself as a self-hosted registry of resources for coding agents (skills, MCP servers, hooks, prompts, sandboxes) with a session-tracing layer on top. If you were looking for an LLM benchmark-style evaluation platform, that is not what the documentation describes (see limits below).&lt;/p&gt;

&lt;p&gt;Facts taken from the repository: Apache 2.0 licence, Python on the server and CLI (FastAPI, Typer), React for the UI, PostgreSQL 16 for the registry, ClickHouse for telemetry, Redis for the queue. The latest commit on the main branch is dated 7 November 2026 and &lt;code&gt;pyproject.toml&lt;/code&gt; carries version &lt;code&gt;1.17.0-dev.0&lt;/code&gt;; I could not read a published release on the page, so I quote no stable number. The fork count (about 1.1k) comes from the same page.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it records from an agent session
&lt;/h2&gt;

&lt;p&gt;The mechanism is easy to state. A hook or extension in the harness (Claude Code, Codex, Cursor, Copilot, Kiro, OpenCode, Pi, Goose, Antigravity according to the README) reads the conversation transcript and sends it to the server. The CLI first writes to a durable local outbox, then posts to &lt;code&gt;/api/v1/ingest/session&lt;/code&gt;; the server returns the highest acknowledged position, and delivery is described as «&amp;nbsp;at-least-once with effectively-once storage&amp;nbsp;». &lt;code&gt;observal reconcile&lt;/code&gt; replays what is missing from the local source.&lt;/p&gt;

&lt;p&gt;Each record becomes an event: prompt, response, tool call, result, token update, error, subagent. Per-session aggregates (tokens, models, tools, duration) feed the dashboards. The documentation is frank about a useful limit: detail depends on what the harness writes, and Observal does not intercept MCP traffic or invent fields.&lt;/p&gt;

&lt;h2&gt;
  
  
  Interface tour: sessions, turns, calls
&lt;/h2&gt;

&lt;p&gt;The next screenshot comes from the repository (docs/img, covered by the project’s Apache 2.0 licence). It shows a Pi session: duration, 17 turns, input and output tokens, cache reads, tool call count, models, then the expandable list of turns with their duration.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgo7ocdoozmtv6siaj9wu.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgo7ocdoozmtv6siaj9wu.png" alt="Session detail in Observal: duration, number of turns, tokens, tool calls and the list of turns" width="799" height="395"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Source : &lt;a href="https://github.com/BlazeUp-AI/Observal/blob/main/README.md" rel="noopener noreferrer"&gt;github.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The session list gives the overview: platform, input and output tokens, tools called, duration, and a «&amp;nbsp;Score&amp;nbsp;» column that stays empty in this demo capture. It is the only evaluation hint I read there; I do not know when or how that score is filled.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7ip2s7fmgq9rggi4qz98.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7ip2s7fmgq9rggi4qz98.png" alt="Observal session list with user, platform, tokens, tools, duration and score column" width="800" height="496"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Source : &lt;a href="https://github.com/BlazeUp-AI/Observal/blob/main/README.md" rel="noopener noreferrer"&gt;github.com&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  From complaint to diagnosis: the documented method
&lt;/h2&gt;

&lt;p&gt;The «&amp;nbsp;Debug agent failures&amp;nbsp;» guide walks through four moves, which I paraphrase: find the session (filters by harness, agent, model, period), read the events in order, inspect tool call detail, then patch delivery gaps with &lt;code&gt;observal reconcile&lt;/code&gt;. Its pattern table is the most telling part: the same tool re-called with identical arguments points to a loop; a call with no result suggests an abrupt stop or an incomplete transcript; an expected MCP call that never shows up means the model did not pick it, or the harness did not record it.&lt;/p&gt;

&lt;p&gt;The commands in the capture below are those of the README and that guide.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhlg8o1zmmxkqjaz2rwhk.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhlg8o1zmmxkqjaz2rwhk.png" alt="Observal commands: auth login, doctor --patch, ops traces with --limit and --turn, reconcile" width="800" height="219"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Installing Observal: server first, CLI second
&lt;/h2&gt;

&lt;p&gt;Per the README there are two parts: a server to host (API, web UI, databases) and a CLI on each workstation. The server needs Docker Engine 24 or later with Compose v2; the &lt;code&gt;install-server.sh&lt;/code&gt; script downloads a Compose package, generates secret files, binds ports to loopback by default and starts ten services. For the CLI: an &lt;code&gt;install.sh&lt;/code&gt; script (macOS, Linux), &lt;code&gt;install.ps1&lt;/code&gt; (Windows) or &lt;code&gt;uv tool install observal-cli&lt;/code&gt; on Python 3.11 or later. Then:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;observal auth login
observal doctor --patch
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The second command detects the harness and installs the telemetry hooks. An operator’s tip: these installs go through &lt;code&gt;curl | bash&lt;/code&gt;. Download the script, read it, and try it on a throwaway machine before pointing real sessions at it. I did not run these commands.&lt;/p&gt;

&lt;h2&gt;
  
  
  Observing Observal itself
&lt;/h2&gt;

&lt;p&gt;The project applies observability to its own stack, without OpenTelemetry. According to its self-observability page: &lt;code&gt;/livez&lt;/code&gt; and &lt;code&gt;/readyz&lt;/code&gt; probes (the latter checks PostgreSQL, ClickHouse and Redis), JSON logs with a request ID carried by the &lt;code&gt;X-Request-ID&lt;/code&gt; header, and Prometheus metrics on &lt;code&gt;/metrics&lt;/code&gt; (request counter, latency histogram, memory, CPU). A Prometheus and a Grafana ship in the Compose file.&lt;/p&gt;

&lt;p&gt;On the agent side, the Claude Code integration page tells you &lt;em&gt;not&lt;/em&gt; to set the &lt;code&gt;OTEL_*&lt;/code&gt; variables or &lt;code&gt;CLAUDE_CODE_ENABLE_TELEMETRY&lt;/code&gt;: Observal does not consume OTLP telemetry and relies only on its hooks and reconciliation. If you want OpenTelemetry metrics on your agents, that is a different pipeline; see &lt;a href="https://span.observer/monitor-coding-agents-enterprise/" rel="noopener noreferrer"&gt;monitoring coding agents in the enterprise&lt;/a&gt;. The project documents no Dynatrace integration, and we assume none.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limits and open points
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Evaluation&lt;/strong&gt;: I found no benchmark or quality test set. There are LLM-generated «&amp;nbsp;Insights&amp;nbsp;» reports (through LiteLLM, with the provider you choose) summarising what helps or hinders, which is usage analysis, not a correctness measure.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Privacy&lt;/strong&gt;: a session contains your prompts, command output and sometimes paths or secrets. Decide on retention (the databases page mentions retention policies) and on who may read what before deploying.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Maturity&lt;/strong&gt;: a development version, a changelog with a busy «&amp;nbsp;Unreleased&amp;nbsp;» section, and the Leaderboard is announced there as off by default. Expect movement.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Coverage&lt;/strong&gt;: detail varies by harness; a poor transcript gives a poor trace.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Weight&lt;/strong&gt;: ten services to watch agents is a real infrastructure project.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Good practices
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Start with one harness and one pilot team, on an isolated server.&lt;/li&gt;
&lt;li&gt;  Keep ports on loopback or behind an authenticated proxy, as the default does.&lt;/li&gt;
&lt;li&gt;  Monitor the stack with &lt;code&gt;/readyz&lt;/code&gt; and Prometheus; learn to repair gaps with &lt;code&gt;observal reconcile&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;  Pair session traces with cost tracking: see &lt;a href="https://span.observer/ai-agent-cost-tokens-opentelemetry/" rel="noopener noreferrer"&gt;AI agent cost from OpenTelemetry token metrics&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;  Put guardrails upstream, as in &lt;a href="https://span.observer/ai-agent-guardrails-production/" rel="noopener noreferrer"&gt;AI agent guardrails in production&lt;/a&gt;. For a neighbouring repository, read &lt;a href="https://span.observer/kagent-ai-agents-kubernetes/" rel="noopener noreferrer"&gt;kagent, AI agents on Kubernetes&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Is Observal still the project’s name?
&lt;/h3&gt;

&lt;p&gt;The README presents it as Agentic Resource Registry, «&amp;nbsp;formerly Observal&amp;nbsp;», with the same code base and licence; the CLI is still called &lt;code&gt;observal&lt;/code&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does Observal evaluate the quality of an agent’s answers?
&lt;/h3&gt;

&lt;p&gt;I did not read any automatic evaluation feature: the project records sessions and produces usage reports. A score column exists in the UI, with no explanation I could verify.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need OpenTelemetry to use it?
&lt;/h3&gt;

&lt;p&gt;No: ingestion goes through hooks and reconciliation, and the documentation advises against enabling Claude Code’s OTLP telemetry for this integration.&lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;Observal answers a concrete question: what happened in that agent session? Its strength is the kept, replayable trace; its limits are a heavy stack, little visible evaluation and a project still on the move. For a team that lets agents work on its code, it deserves an isolated trial, with a clear decision about the data it keeps.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;span.observer is an independent community blog, not affiliated with the Observal project, its maintainers or Dynatrace LLC. Trademarks belong to their owners. Tools change quickly: check commands and options in their official documentation.&lt;/em&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://span.observer/en/observal-ai-coding-agent-observability/" rel="noopener noreferrer"&gt;span.observer&lt;/a&gt;. French version: &lt;a href="https://span.observer/fr/observal-observabilite-agents-de-code/" rel="noopener noreferrer"&gt;https://span.observer/fr/observal-observabilite-agents-de-code/&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>aiagents</category>
      <category>aiobservability</category>
      <category>claudecode</category>
      <category>mcp</category>
    </item>
  </channel>
</rss>
