I am using PhantomJS(2.0) to render a screenshot on Ubuntu 14.04. I have a pretty standard script to render a page (https://gist.github.com/Jaciones/b078326a0a6d664a5f30).
Has always worked great.
But now, with a particular website url, it will fail about 1 in 4 times. On the failing ones, it will hang for 60 seconds, and then exit (with no STDOUT), but the STDERR will be "Terminated". What does this mean? Is this PhantomJS crashing, or is this just a "timeout" internally trying to load a webpage? I have code in the in my rendering script to output when resources fail, but STDOUT is empty in the failing case (in the non failing case, everything works as expected).
Related
I am having an issue in Github Actions.
I am using Cypress to test the frontend of my app, and everything is working perfectly in the Cypress app.
But when I am pushing everything on master in GitHub and run the test via Github Actions, every now and then, the same flaky test is failing with he following error.
Timed out retrying after 5000ms: Expected to find element "xxx" but never found it.
This line is the problematic one:
cy.purposeElement("delete_user_dialog").should('be.visible')
The element we are talking about takes 0.2s to appear with a fade-in animation.
My guess is that the page was slow to respond, while Cypress was fast to act.
How can I avoid solve a flaky test like that?
I could use a cy.wait but it is not recommended and I do not want to increase the time of the test.
Plus, the result is as flaky.
It may be as simple as increasing the timeout, especially if the test works perfectly when running on the local machine.
cy.purposeElement("delete_user_dialog", {timeout:20_000}).should('be.visible')
To be 100% sure, do a burn test which is described here: Burning Tests with cypress-grep
Is there a way to tell the system to restart the test in case a specific rare system error comes up?
Basically sometimes we get strange errors related to elements being "obscured" or "stale", but which do not mean the site is not working etc. It has to do with the site's latency I believe like CSS not loading quickly enough etc.
For example is there a directive to tell the system that if an error like
[Facebook\WebDriver\Exception\ElementClickInterceptedException] Element
\<li id="nav_step0" class="nav-steps selected"> is not clickable at point (330,237)
because another element \<div id="ajaxloading_mask" class="mask"> obscures it
To simply relaunch the test again?
No, there is no way to relaunch failed test on specific error.
You can rerun all failed tests:
codecep run || codecept run -g failed
This command executes all tests, if any tests failed, it reruns only failed tests.
When using the APIs defined by Protractor & Jasmine (the default/supported runner for Protractor), the tests will always work okay on individual developer laptops. For some reason when the test runs on the Jenkins CI server, they will fail (despite being in the same docker containers on both hosts, and that was wildly frustrating.)
This error occurs: A Jasmine spec timed out. Resetting the WebDriver Control Flow.
This error also appears: Error: Timeout - Async callback was not invoked within timeout specified by jasmine.DEFAULT_TIMEOUT_INTERVAL.
Setting getPageTimeout & allScriptsTimeout to 30 seconds had no effect on this.
I tried changing jasmine.DEFAULT_TIMEOUT_INTERVAL to 60 seconds for all tests in this suite, once the first error appears then every test will wait the full 60 seconds and time out.
I've read and reread Protractor's page on timeouts but none of that seems relevant to this situation.
Even stranger still, it seems like some kind of buffer issue - at first the tests would always fail on a particular spec, and nothing about that spec looked wrong. While debugging I upgraded the selenium docker container from 2.53.1-beryllium to 3.4.0-einsteinium and the tests still failed but they failed a couple specs down - suggesting that maybe there was some optimization in the update and so it was able to get more done before it gave out.
I confirmed that by rearranging the order of the specs - the specs that had failed consistently before were now passing and a test that previously passed began to fail (but around the same time in the test duration as the other failures before the reorder.)
Environment:
protractor - 5.1.2
selenium/standalone-chrome-debug - 3.4.0-einsteinium
docker - 1.12.5
The solution ended up being simple - I first found it on a chrome bug report, and it turned out it was also listed right on the front page of the docker-selenium repo but the text wasn't clear as to what it was for when I'd read it the first time. (It says that selenium will crash without it, but the errors I was getting from Jasmine were just talking about timeouts, and that was quite misleading.)
Chrome apparently utilizes /dev/shm, and apparently that's fairly small in docker. There are workarounds for chrome and firefox linked from their README that explain how to resolve the issue.
I had a couple test suites fail after applying the fix but all the test suites have been running and passing for the last day, so I think that was actually the problem and that this solution works. Hope this helps!
It's not a problem and actually is a nice side affect, but it is confusing me.
When I run the test suite via the command line I see IE pop up and the test run.
When I run it with the exact same arguments from the Task Schedular though it doesn't display IE. The test seems to run correctly (I'm getting the expected TestResults.xml so it all looks OK.
Why's this happening though?
The command is:
"C:\Program Files (x86)\NUnit.org\nunit-console\nunit3-console.exe" "Path_to_test_assembly"
P.S. I'm using the .NET version of Selenium with the IE web driver.
Ok, it seems that the problem is with your access to remote machine. Your IE test are running as a background process on that machine, or running on wrong sessionID. It means that there could be more users/accounts, and your test is running on wrong one.
I'm not sure how exactly are you running this, but you could check your session ID's by typing qwinsta in command line on that machine.
If you want it to run properly you should pass this sessionID as a parameter when connecting to remote desktop, for example, if using psexec and your sessionID is 2 than you pass "-i 2" when starting it. It means that it will interact on user with sessionID 2 on that machine.
I get this error when running a number of tests in seleniums Bromine, The selenium RC version 1.0.2 outputs this:
WARN - GET /selenium-server/driver/?cmd=testComplete&1=&2=&sessionId=1274d41621c64fc08c1e7ea0a58f260b HTTP/1.0 java.lang.IllegalStateException: unexpected command json={command:"open",target:"/Library/Security/Login.aspx?ReturnUrl=%2fIndex.aspx",value:""} in place before new command selectWindow could be added at org.openqa.selenium.server.CommandQueue.doCommandWithoutWaitingForARe
sponse(CommandQueue.java:121)
Any ideas
Recently I had to track this problem on our testing environment and it appears, that the reason was Firefox crashing. If this error follows two 'Command timed out' exceptions, then your browser probably crashed or hanged.
Upon inspection of code of Selenium RC I realized that 'Unexpected command' error appears when there is an overflow in command queue. This can be caused by lack of responses from the browser, so if browser crashes, you end up receiving this error.
Check your dmesg logs (or some other logs, if not under Linux) to see, if there is anything suspicious. In my case there were entries like this:
plugin-containe[30867]: segfault at 0 ip 00007f07a6ff503d sp 00007f079d593260 error 4 in libxul.so[7f07a6265000+146f000]
libxul.so and plugin-container are modules of Firefox. Upgrading it to newer version helped in my case.
I'm also seeing this IllegalStateException, much too frequently. Its occurrence seems random, as if something in Selenium isn't synchronized properly. I have seen it several times in connection with a TestNG Listener that calls selenium to do a screen capture, but again, it's unpredictable.