-
Bug
-
Resolution: Fixed
-
Major
-
Jenkins
I'm using the ec2-plugin to launch Windows nodes in EC2. The instances are launched correctly, and the slave.jar is started over WinRM once a WinRM connection was established.
Running a simple freestyle job containing a single batch step (echo TEST) works perfectly fine and finishes immediately.
Running a similar pipeline job (bat 'echo TEST') takes 22 minutes (!!!) to finish. Adding the same node to jenkins manually via JNLP allows this job to also finish immediately. I can consistently reproduce these behaviours.
Here's some log output from the durable step plugin and the winrm package of the ec2-plugin: https://gist.github.com/anonymous/074223af8b5942b0363f
There's more log information available, but it looks mostly the same: A lot of InterruptedExceptions all over the place.
As running the slave over JNLP instead of WinRM seems to work perfectly fine, I believe this to be some weird interaction between the WinRM connection and the Filesystem Monitors in the Durable Task plugin.
- links to
I was not aware there was a WinRM-based launcher. Is that in the EC2 plugin? I remember suggesting it to kohsuke that one could be created to supersede the DCOM launcher, and christ66 started to play with it, but kohsuke came up with a list of reasons why it would never work well enough to be usable. Maybe you just should not use it?