Activity

1. vojtajina

   added

   windows

   on Mar 16, 2014
2. vojtajina commented on Mar 16, 2014

   

   vojtajina

   on Mar 16, 2014

   Contributor

   More actions

   @sylvain-hamel thanks for reporting this. It would be great if you can look into it a bit more.

   The changelog of all the changes is here.

   There was a significant refactoring in the launcher, I think that might have introduced some issue:
   e6f36ca
3. sylvain-hamel commented on Mar 16, 2014

   

   sylvain-hamel

   on Mar 16, 2014

   ContributorAuthor

   More actions

   Yep, I'll take a closer look. You can assign this issue to me if you want.
4. sylvain-hamel commented on Mar 19, 2014

   

   sylvain-hamel

   on Mar 19, 2014

   ContributorAuthor

   More actions

   I found the change that causes this new behavior: karma 0.10 does not call webServer.close() whereas karma 0.12 does.

   server.js (karma 0.10)

     var disconnectBrowsers = function(code) {
       var sockets = socketServer.sockets.sockets;
       Object.getOwnPropertyNames(sockets).forEach(function(key) {
         sockets[key].removeAllListeners('disconnect');
       });
       globalEmitter.emitAsync('exit').then(function () {
           done(code || 0);
       });
     };
   

   server.js - karma 0.12

     var disconnectBrowsers = function(code) {
       var sockets = socketServer.sockets.sockets;
       Object.getOwnPropertyNames(sockets).forEach(function(key) {
         sockets[key].removeAllListeners('disconnect');
       });
       globalEmitter.emitAsync('exit').then(function () {
         webServer.close(function () {
           webServer.removeAllListeners();
           processWrapper.removeAllListeners();
           done(code || 0);
         });
       });
     };
   

   node's Server.close waits for connections to close but when running with IE11, there are pending connections even after killing the iexplore.exe process.

   More info on IE:

   When you launch iexplorer.exe, that spawns two processes. You can witness that by just running iexplorer.exe from the command-line and you'll see two processes in Task Manager. However

   • On IE10: If you kill the first process using Task Manager, the second process goes away.
   • On IE11: If you kill the first process using Task Manager, the second process remains there (apparently for no reason).

   My understanding of the root cause:

   • My guess is that with the arrival of IE11, they use the second process to act as some kind of proxy for the first process and handles connections to the server.
   • Because killing the first process no longer kills the second one AND because Karma now waits the server to close, karma ends up waiting indefinitely for the server to close.

   I see 3 options to fix that:

   1. stop calling webserver.close() in karma, as it used to be in karma 0.10
   2. implement a webserver close timeout in disconnectBrowsers, something like this (I tried it and it works):

     var disconnectBrowsers = function(code) {
       var sockets = socketServer.sockets.sockets;
       Object.getOwnPropertyNames(sockets).forEach(function(key) {
         sockets[key].removeAllListeners('disconnect');
       });
       globalEmitter.emitAsync('exit').then(function () {
         webServer.close(function () {
           webServer.removeAllListeners();
           processWrapper.removeAllListeners();
           done(code || 0);
         });
         ////////////////////////////////////////////////////////////////////////////
         // call done, which exits the current process after 3 sec
         ////////////////////////////////////////////////////////////////////////////
         setTimeout(done, 3000); 
       });
     };
   

   x
   3. Implement a hack in karma-ie-launcher to find and kill the second iexplorer process when the first one is killed by karma.

   I think that option 2 is the best because it is the simplest and protects karam against future clients (browsers) that would keep connections to the karma server.

   What do you think?
5. dignifiedquire commented on Mar 19, 2014

   

   dignifiedquire

   on Mar 19, 2014

   Member

   More actions

   I agree with the suggestion of 2. My experience with server.close is a very mixed and it should always have a timeout to kill any connections that are left.
6. sylvain-hamel

   added 3 commits that reference this issue on Mar 21, 2014

   fix(web-server): implement a timeout on webServer.close()

   ...

   This issue will close once commit 32cbfa4 is merged into the 'master' branch.

   32cbfa4

   fix(web-server): implement a timeout on webServer.close()

   ...

   This issue will close once commit ad5fa5e is merged into the 'master' branch.

   ad5fa5e

   fix(web-server): implement a timeout on webServer.close()

   ...

   This issue will close once commit fe3dca7 is merged into the 'master' branch.

   fe3dca7
7. sylvain-hamel commented on Mar 27, 2014

   

   sylvain-hamel

   on Mar 27, 2014

   ContributorAuthor

   More actions

   We've been running with #973 for a week now. It works fine.

   However, each time our ci is triggered, a new extra process leaks out (as explained in a previous comment). After a few days, we start getting ERROR [launcher]: Cannot start IE. If I kill the 80 iexplore.exe processes and restart the build and then it works again. I don't know why exactly that prevents a new iexplore.exe from being spawned but it doesn’t really matter. 80 inactive processes is not good, even if it worked, at some point, the CI server would run out of memory and fail.

   I really think #973 is valid and necessary (solution 2). However I'll also implement solution 3 in karma-ie-launcher to get rid of the extra processes that accumulate.

   In the meantime, can you guys merge #973.
8. sylvain-hamel

   closed this as completedin #973on Apr 1, 2014