New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Daily runs of Edge on Azure Pipelines #15174
Comments
https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5330 was the first full build, but it failed to get into wpt.fyi, seemingly because of #15172 fixes that and https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5461 is in progress with that fix. We'll see what the next error is :) |
https://staging.wpt.fyi/results/?sha=d11be4eae8 is the first sign of this maybe working. I uploaded wpt_report_10.json.gz from https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5461 at https://staging.wpt.fyi/api/results/upload and changed the browser name to "edge" to get anything to show up. @lukebjerring @Hexcles what to do about "edge_webdriver" as the product name? |
Where is that name originally coming from? You can set the fields in the upload form (see the api README.md on wpt.fyi), but the source should be corrected too. |
@lukebjerring we have two product names for edge in |
I'm not sure why the distinction propagates into the product name, but is this something that we should be handling in general? Do we have an exhaustive list of these cases?
|
If memory serves, @gsnedders found that two products names was easier, and also in favor of it could be said that it doesn't leave behind an argument for switching the executor when the transition is done. The only two current products with underscores are chrome_android and edge_webdriver. With the ExecutorCDP work expect we'll have chrome_cdp for a time, unless we use another method. Funny enough, edge_webdriver runs into a lot of trouble when running the webdriver/ tests, so I've exluded those tests from the setup. Adding the webdriver label would be a bit amusing, but ultimately harmless if it works. |
Job 3 in https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5461 failed:
I'm not sure where "Entering debug mode. Use h or ? for help." came from, but https://github.com/Microsoft/azure-pipelines-task-lib/blob/master/powershell/VstsTaskSdk/ToolFunctions.ps1 appears to be the code that makes noise after that. |
microsoft/azure-pipelines-task-lib#410 looks like it could be related. |
https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5536 is the next candidate for a full run. |
Yeah, vastly easier with how we currently handle everything; @jgraham landed the first bit of refactoring needed to actually have arguments for this (and I don't think a needless argument is really a big problem, FWIW).
It may be worth adding an arg for that, if we expect it to be long-lived, now the largest part of the refactoring is done. @foolip: with the above failures/retries, I presume this isn't quite ready to land yet? waiting for a fully successful run? |
https://staging.wpt.fyi/results/?run_id=6022134270787584 is wpt_report_13.json.gz from that build manually uploaded. As long as the build completes, it looks like this might work. |
https://wpt.fyi/results/?run_id=4849781323071488 / Looks like it's time to commit the config and get a run from master. |
PR up at #15196. @gsnedders for the "edge_webdriver" problem I had this fix: --- b/tools/wptrunner/wptrunner/wpttest.py
+++ a/tools/wptrunner/wptrunner/wpttest.py
@@ -85,7 +85,9 @@ class RunInfo(dict):
if rev:
self["revision"] = rev
- self["product"] = product
+ # To deal with edge_webdriver, chrome_android and similar, only use the
+ # first part of the product.
+ self["product"] = product.split("_")[0]
if debug is not None:
self["debug"] = debug
elif "debug" not in self: But I'm not sure if |
https://gist.github.com/foolip/3be27fa5b8c57ec1121eff4c32fdefb3 is a script for modifying the reports after the fact which I'll use to try uploading the next run. |
https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5754 is the first manually triggered daily run after the config landed in the repo. The edge_webdriver problem hasn't been resolved, but it should be possible to upload the results manually. |
As it turns out, runs have been getting into wpt.fyi and staging.wpt.fyi just fine, just as product edge_webdriver instead of edge. Here's one: https://wpt.fyi/results/?run_id=5693649719394304 |
I wonder if we should just handle underscore-separated things on the wpt.fyi side? I think it's highly unlikely we'll add an actual product with an underscore in it. Do we have some way to encode things like options in the |
I'd say we handle it in the /api/results/create, by mapping edge_* to edge and labelling it with the original name (e.g. edge_webdriver). That way we're not being as lossy with the info? |
Yeah, that would work. |
https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5754 didn't complete and is still "running" but will presumably time out. For some reason only 4 jobs were started and completed, then all remaining claim to be "Waiting for an available agent". But I triggered https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5765 which which uses the same agent pool and it finished. Smells like a bug in Azure Pipelines. @mustjab are you able to see what went wrong here? I'll start another build and see how that goes. |
https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5799 finished and I'll try changing the product to see if it then shows up by default. |
That worked, and Edge 44 now shows up on the frontpage of staging.wpt.fyi as it was an aligned run. I'll hold off doing the same for prod though, it's best to have web-platform-tests/results-collection#647 merged and comparing the results. |
I've renamed https://staging.wpt.fyi/results/?run_id=5193840969383936 back to edge_webdriver be in sync with prod and not cause later confusion. |
See web-platform-tests/wpt#15174 (comment) for the origins of this feature.
See web-platform-tests/wpt#15174 (comment) for the origins of this feature.
With more logs and a failed run https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=5991 there are more clues about what might be happening:
"WebDriver was not accessible within the timeout" suggests a problem with EdgeDriver. @thejohnjansen @mustjab do you know if it's possible to enable more logging for EdgeDriver to diagnose the problem? |
See web-platform-tests/wpt#15174 (comment) for the origins of this feature.
Can we see if we have any other |
@gsnedders if that could be a cause of this, we could move the cleanup which kills those processes to before running the tests. |
https://wpt.fyi/results/?label=azure&max-count=100&product=edge now has 3 runs. They're all bad, so here's a permalink that will keep working after I rename them: The runs didn't really succeed, for example job 11 in https://dev.azure.com/web-platform-tests/wpt/_build/results?buildId=8478 had only these results in wpt_report_11.json: {
"results": [
{
"duration": 0,
"message": null,
"status": "SKIP",
"subtests": [],
"test": "/streams/writable-streams/write.any.js"
},
{
"duration": 0,
"message": null,
"status": "SKIP",
"subtests": [],
"test": "/wasm/jsapi/instance/toString.any.js"
},
{
"duration": 0,
"message": null,
"status": "SKIP",
"subtests": [],
"test": "/streams/readable-streams/brand-checks.any.js"
}
],
"run_info": {
"bits": 64,
"browser_version": "44.17763.1.0",
"debug": false,
"has_sandbox": true,
"os": "win",
"os_version": "10.0",
"processor": "x86_64",
"product": "edge_webdriver",
"revision": "ce50d58dd4a10d34bd70242d3833e6940169a246",
"service_pack": "",
"verify": false,
"version": "10.0.17763",
"wasm": false
},
"time_end": 1551424024225,
"time_start": 1551423302738
} The logs show it was trouble from the start:
@mustjab do you have any tips to try debugging this? I think it's MicrosoftEdgeDriver which isn't responding. Perhaps it was left running from a previous job and is preventing any new runs? |
This is blocked on #16118. |
With #16920 merged this will probably soon be possible to resolve. We should have results every 6 hours if the setup is reliable. |
This was resolved and should be closed. There's a hickup tracked by #18397, but the initial work was done. |
This is like #14836, but for Edge on Windows 10.
@mustjab and @thejohnjansen have provided a self-hosted agent pool with 5 Window 10 VMs to make this possible.
The text was updated successfully, but these errors were encountered: