ATH does not startup with BROWSER=chrome-container (or firefox-container) anymore #1090
Labels
Comments
6 tasks
uhafner
changed the title
|
In order to fix this I need much more information that is available from GitHub. the container logs should say why the browser could not start - but these are not available (at least as far as I can tell). And it does not completely break as https://github.com/jenkinsci/warnings-ng-plugin/actions/runs/4967128160/jobs/8888944269?pr=1470 shows that things did work
yes it is flaky - but to understand why without any more information is required. Additionally - I am not sure that the Diagnostic rule works correctly here - and could hide issues where a test fails by trying to capture diagnostics when it is not possible. |
|
I'm not sure if I can follow. You already started work on an additional PR jenkinsci/warnings-ng-plugin#1507 where you could see those problems on your machine. Or does that PR work now in your setup and just fails on GitHub? I tried to align my PR jenkinsci/warnings-ng-plugin#1470 with the changes of your PR but also without success. What information do you need from me? Aren't the failing PRs in the warnings and git-forensics plugins not a good start to investigate what is broken in the Selenium 4 container tests? |
This is also a good point, from the log it seems to use firefox even though |
I could reproduce container failing to start consistently and I got to the point where the container would start correctly. |
There are however copious amounts of warnings produced as popper2 adjunct does not exist. |
|
I can not reproduce this locally inside a Linux VM, and am only lead to the conclusion that something is wrong in the github actions environment which I have no way to debug. |
|
I have updated to 5588.vd13b_52985008 and run the tests to double check the container and it works flawlessly - but it is useing firefox for whatever reason...? (and version 4.6.0 but still it works)... |
|
works with chrome also. diff --git a/ui-tests/pom.xml b/ui-tests/pom.xml
index cd2c5a919..6fe83f104 100644
--- a/ui-tests/pom.xml
+++ b/ui-tests/pom.xml
@@ -48,7 +48,7 @@
<dependency>
<groupId>org.jenkins-ci</groupId>
<artifactId>acceptance-test-harness</artifactId>
- <version>5581.vfd8e43f46a_03</version>
+ <version>5588.vd13b_52985008</version>
</dependency>
<dependency>
<groupId>com.fasterxml.jackson.core</groupId>
@@ -223,7 +223,7 @@
<environmentVariables>
<PLUGINS_DIR>../plugin/target/test-classes/test-dependencies</PLUGINS_DIR>
<JENKINS_VERSION>${jenkins.version}</JENKINS_VERSION>
- <BROWSER>firefox-container</BROWSER>
+ <BROWSER>chrome-container</BROWSER>
</environmentVariables>
<rerunFailingTestsCount>1</rerunFailingTestsCount>
<includes> |
|
the github runners have 7GB of ram. the These Jenkins's for test at first glance do not appear to have any memory limits set either. With all this unconstrained memory usage I would not be surprised if something is getting OOMKiller by the OS and would explain the randomness failures and failed to connect. |
Ah, thanks, this is a bug indeed. Opened jenkinsci/bootstrap5-api-plugin#213 as a fix. |
The interesting thing is, that the old container worked flawlessly. Now everything is flaky: in jenkinsci/warnings-ng-plugin#1470 I have a green build for one commit (jenkinsci/warnings-ng-plugin@83a315c) and then again some failures. Are there some additional logs available that may explain the problem? The error message actually does not help very much. And jenkinsci/git-forensics-plugin#602 is green as well. Really strange... |
there are docker logs for the container - https://github.com/jenkinsci/acceptance-test-harness/blob/c141c9a74536b66600d0951eb3445bb49b927d15/src/main/java/org/jenkinsci/test/acceptance/FallbackConfig.java#L228C46-L246 some things to possibly try
I'm wondering if the issue at heart is that the tests are flaky and something is trying to capture diagnostics at a point in time that is too late (causing the original error to get masked). Can you reproduce this locally at all? |
the new container starts an extra service IIRC - it could just be that the memory was on the limit and it has not just tipped over. OT: the container-firefox is pretty much how I run locally on windows (except I am manually launching the selenium container because docker host networking is baad) |
Yes, I have seen these. They do not contain anything useful.
In my linux based virtual machine all tests fail with the errors from above. I'll see if I can find something useful... |
can you share them anyway? @jeromepochat saw some similar failures in a CI environment (not using the |
|
Here are the logs when I run one of my tests in a Linux based virtual machine. I can reproduce these failures as soon as I switch to the new ATH. When switching back to the old ATH everything works like a charm. I wonder if it is worth to switch locally to the old 3.1 docker images to see if the problem is in the image change or in the source code changes of the ATH? The test failure is here: |
can you run the tests differently? ie using the other thing that jumpes out is the devbox Ip |
|
do you have a funky firewall or anything like that, other than that a broken docker version or some other Linux kernel bug? The following is not setting anything special - relying on the project setup for everything (the tip of jenkinsci/warnings-ng-plugin#1470) followed by |
|
@uhafner can you run a I observed for the first time today an almost similar stack trace - where I start selenium myself and had set the remote webdriver url to be |
|
no idea if this is related or not. stack trace whilst it is "hanging" of the selenium server shows the following thread which appears to be suspect: (selenium 4.6.0) full stack trace : |
|
Seems to be fixed now, see jenkinsci/warnings-ng-plugin#1643. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
In https://github.com/jenkinsci/acceptance-test-harness/releases/tag/5527.vf5d62a_388c39 the containers have been adapted to match the Selenium 4 upgrade (#1077).
This update does not work for my plugins, it breaks all of my tests:
The text was updated successfully, but these errors were encountered: