Bug 175269 – WPT harness errors on leaks and iOS-sim EWS bots

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Alexey Proskuryakov]

A large number of tests are failing on leaks bot with something that looks like a harness error. See e.g. <https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&revision=220122&showExpectations=true&tests=imported%2Fw3c%2Fweb-platform-tests%2FWebCryptoAPI%2FgenerateKey%2Ffailures_RSASSA-PKCS1-v1_5.worker.html>.

Here are some of the failures it hits:

--------------------------------------
Harness Error (FAIL), message = Error in remote http://localhost:8800/WebCryptoAPI/generateKey/failures_RSASSA-PKCS1-v1_5.worker.js: NetworkError:  A network error occurred.
--------------------------------------
layer at (0,0) size 800x600
  RenderView at (0,0) size 800x600
layer at (0,0) size 800x8
  RenderBlock {HTML} at (0,0) size 800x8
    RenderBody {BODY} at (8,8) size 784x0
      RenderBlock {DIV} at (0,0) size 784x0
--------------------------------------
CONSOLE MESSAGE: line 1677: TypeError: null is not an object (evaluating 'this.message_target.removeEventListener')
CONSOLE MESSAGE: line 3: NetworkError:  A network error occurred.

FAIL Untitled NetworkError:  A network error occurred.

--------------------------------------
Harness Error (FAIL), message = Error in remote: [object Event]
--------------------------------------

[youenn fablet]

Are these errors all related to testing workers in WPT?

[Alexey Proskuryakov]

I don't think so, failures on many non-worker tests seem to be similar. See e.g. <https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r220343%20(2511)/imported/w3c/web-platform-tests/IndexedDB/idbobjectstore_add5-pretty-diff.html>

All results:

https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r220343%20(2511)/results.html

[youenn fablet]

Looking at some tests, it seems as if testharnessreport.js is not sent correctly to the browser (dump as render tree and not as text).
For some other tests, it is as if testharness.js  is not sent correctly (function test undefined).

Looking at https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r220343%20(2511)/wptwk_process_log.out.txt, there is no clear error from the WPT server.

Looking at one failing test (/XMLHttpRequest/abort-after-receive.htm) vs a successful test (/XMLHttpRequest/abort-after-send.htm), it seems as if the unsuccessful test is not making any HTTP request to get testharnessreport.js.

Any idea why retrieval of some HTTP resources would cause issues in that particular configuration?
How slow are these bots?

We could try to add more wpt logging.

[Alexey Proskuryakov]

> Any idea why retrieval of some HTTP resources would cause issues in that particular configuration?

No. If there no request in the log, it could be a WebKit cache bug of some sort, but I don't know how much we can trust WPT logging.

> How slow are these bots?

Not too slow, the tests run in just 1 hour.

[Alexey Proskuryakov]

This appears to be quite reproducible locally with a release build.

run-webkit-tests --no-build --exit-after-n-crashes-or-timeouts 50 --exit-after-n-failures 500 --release --leaks --dump-render-tree --no-retry-failures --no-sample-on-timeout imported/w3c/web-platform-tests

[Alexey Proskuryakov]

...and one needs to delete test results directory before running the tests for this to reproduce!

rm -rf WebKitBuild/Release/layout-test-results

[youenn fablet]

I am also able to reproduce the issue fairly consistently when delete the output folder.

I filed https://bugs.webkit.org/show_bug.cgi?id=175543 for that particular issue.

[Alexey Proskuryakov]

Hrmpf, this is still happening after the fix for bug 175543.

[youenn fablet]

Created attachment 318112 [details]
add logging

[youenn fablet]

(In reply to youenn fablet from comment #9)
> Created attachment 318112 [details]
> Patch

wpt log does not show anything abnormal.
It is as if testharnessreport.js is not executed correctly on DRT.
Adding some WPT logging to check whether responses have the expected size.

Do we have other bots than WK1 having this issue?

[Ryan Haddad]

(In reply to youenn fablet from comment #10)
> (In reply to youenn fablet from comment #9)
> > Created attachment 318112 [details]
> > Patch
> 
> wpt log does not show anything abnormal.
> It is as if testharnessreport.js is not executed correctly on DRT.
> Adding some WPT logging to check whether responses have the expected size.
> 
> Do we have other bots than WK1 having this issue?
I think I was seeing a similar issue with the ios-sim EWS bots last week. They seem to be working right now, but I will monitor them and update this bug if I find any examples.

[WebKit Commit Bot]

Comment on attachment 318112 [details]
add logging

Clearing flags on attachment: 318112

Committed r220754: <http://trac.webkit.org/changeset/220754>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/33902404>

[Alexey Proskuryakov]

Re-opening, as this was only logging.

[youenn fablet]

Looking at https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r220760%20(2656)/wptwk_process_log.out.txt and imported/w3c/web-platform-tests/user-timing/test_user_timing_measure_navigation_timing.html	test that fails for instance.

We got:

DEBUG:web-platform-tests:200 GET /resources/testharness.js (http://localhost:8800/user-timing/test_user_timing_measure_navigation_timing.html) 0
DEBUG:web-platform-tests:200 GET /user-timing/resources/webperftestharness.js (http://localhost:8800/user-timing/test_user_timing_measure_navigation_timing.html) 0
DEBUG:web-platform-tests:Wrote content of size 4880 for /user-timing/resources/webperftestharness.js
DEBUG:web-platform-tests:Wrote content of size 100163 for /resources/testharness.js
DEBUG:web-platform-tests:Got exception when writing response Traceback (most recent call last):
  File "/Volumes/Data/slave/sierra-leaks/build/LayoutTests/imported/w3c/web-platform-tests/tools/wptserve/wptserve/response.py", line 446, in write
    self._wfile.write(buffer)
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/socket.py", line 328, in write
    self.flush()
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/socket.py", line 307, in flush
    self._sock.sendall(view[write_offset:write_offset+buffer_size])
error: [Errno 32] Broken pipe

[Ryan Haddad]

Created attachment 318207 [details]
Results from ios-sim EWS

Attached an example that I think exhibits this issue on ios-sim EWS from bug https://bugs.webkit.org/show_bug.cgi?id=173457

[Alexey Proskuryakov]

> error: [Errno 32] Broken pipe

What process is this pipe connected to?

[youenn fablet]

(In reply to Alexey Proskuryakov from comment #18)
> > error: [Errno 32] Broken pipe
> 
> What process is this pipe connected to?

DumpRenderTree I would guess.

[Alexey Proskuryakov]

DumpRenderTree uses TCP to talk to the server, not a pipe. So the pipe has to be between processes in the WPT harness. A SIGPIPE means that the other process closes the pipe too early.

[youenn fablet]

Cannot SIGPIPE be sent to WPT if the socket WPT uses to write is not connected anymore, for instance if DRT decided to close the TCP connection?

[Alexey Proskuryakov]

It could technically, but I'm not sure if that's likely. WebKit won't close the connection until it has the response, so the scenario is not very plausible.

[youenn fablet]

(In reply to Alexey Proskuryakov from comment #22)
> It could technically, but I'm not sure if that's likely. WebKit won't close
> the connection until it has the response, so the scenario is not very
> plausible.

Do you have any other idea?
I was planning to add some more logging to check precisely which resources were failing to be sent and add timestamps so that we can check if somehow the WPT server is hanging.

[Robin Morisset]

Could this bug also explain the failing test result for iOS-sim on https://bugs.webkit.org/show_bug.cgi?id=175610 ?

[youenn fablet]

(In reply to Robin Morisset from comment #24)
> Could this bug also explain the failing test result for iOS-sim on
> https://bugs.webkit.org/show_bug.cgi?id=175610 ?

would need to look at the wpt log but it seems so.

[youenn fablet]

Since it only happens on some bots, could it be something similar to bug 164651?

[Alexey Proskuryakov]

It's similar in that the WPT server isn't working well in both cases, but I don't see a reason to believe that the root causes are the same.

[Ryan Haddad]

This continues to affect ios-sim EWS bots. They often don't finish tests or produce false positives on patches. We need to fix this ASAP.

[Ryan Haddad]

Youenn, have you had a chance to look in to adding additional logging?

[youenn fablet]

Created attachment 318992 [details]
Adding more logs

[Alexey Proskuryakov]

Comment on attachment 318992 [details]
Adding more logs

rs=me

[WebKit Commit Bot]

Comment on attachment 318992 [details]
Adding more logs

Clearing flags on attachment: 318992

Committed r221141: <http://trac.webkit.org/changeset/221141>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Ryan Haddad]

Reopening since the patch only added logging.

[youenn fablet]

I looked at some logs like https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r221152%20(2804)/wptwk_process_log.out.txt where tests do not run well and https://build.webkit.org/results/Apple%20Sierra%20(Leaks)/r221155%20(2805)/wptwk_process_log.out.txt which runs well.

When everything is fine, the server got requests and returns responses steadily.
When the run is wrong, there are periods like 10 seconds where no requests or responses are logged. 
The server continues to log "Broken pipe" errors but that is all.
I will skip a test that uses message2.py to reduce the number of these "Broken pipe" errors.

[youenn fablet]

Created attachment 319033 [details]
Skipping imported/w3c/web-platform-tests/eventsource/event-data.html

[WebKit Commit Bot]

Comment on attachment 319033 [details]
Skipping imported/w3c/web-platform-tests/eventsource/event-data.html

Clearing flags on attachment: 319033

Committed r221170: <http://trac.webkit.org/changeset/221170>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Ryan Haddad]

Reopening because the issue is not resolved.

[youenn fablet]

On the EWS leak bot, I can see some TCP level errors like:


default	09:43:54.343176 -0700	DumpRenderTree	TIC TCP Conn Start [21:0x7fbbca7948d0]
default	09:43:54.351226 -0700	DumpRenderTree	TIC TCP Conn Event [21:0x7fbbca7948d0]: 1 Err(0)
default	09:43:54.351343 -0700	DumpRenderTree	TIC TCP Conn Connected [21:0x7fbbca7948d0]: Err(0)
default	09:43:54.354371 -0700	DumpRenderTree	TIC TCP Conn Event [21:0x7fbbca7948d0]: 2 Err(0)
error	09:43:54.354614 -0700	DumpRenderTree	TIC Read Error [21:0x7fbbca7948d0]: 1:57
default	09:43:54.355162 -0700	DumpRenderTree	TIC TCP Conn Cancel [21:0x7fbbca7948d0]
default	09:43:54.356212 -0700	DumpRenderTree	TIC TCP Conn Destroyed [21:0x7fbbca7948d0]
default	09:43:54.378052 -0700	DumpRenderTree	TIC TCP Conn Start [22:0x7fbbca79ad40]
default	09:43:54.381394 -0700	DumpRenderTree	TIC TCP Conn Start [23:0x7fbbca601770]
default	09:43:54.381732 -0700	DumpRenderTree	TIC TCP Conn Start [24:0x7fbbca1016d0]
default	09:43:54.397447 -0700	DumpRenderTree	TIC TCP Conn Event [22:0x7fbbca79ad40]: 1 Err(0)
default	09:43:54.397536 -0700	DumpRenderTree	TIC TCP Conn Connected [22:0x7fbbca79ad40]: Err(0)
default	09:43:54.399839 -0700	DumpRenderTree	TIC TCP Conn Event [24:0x7fbbca1016d0]: 1 Err(0)
default	09:43:54.399951 -0700	DumpRenderTree	TIC TCP Conn Connected [24:0x7fbbca1016d0]: Err(0)
default	09:43:54.404178 -0700	DumpRenderTree	TIC TCP Conn Event [24:0x7fbbca1016d0]: 2 Err(0)
default	09:43:54.404477 -0700	DumpRenderTree	TIC TCP Conn Event [22:0x7fbbca79ad40]: 2 Err(0)
error	09:43:54.405014 -0700	DumpRenderTree	TIC Read Error [22:0x7fbbca79ad40]: 1:57
error	09:43:54.405163 -0700	DumpRenderTree	TIC Read Error [24:0x7fbbca1016d0]: 1:57
default	09:43:54.405692 -0700	DumpRenderTree	TIC TCP Conn Cancel [22:0x7fbbca79ad40]
default	09:43:54.406162 -0700	DumpRenderTree	TIC TCP Conn Cancel [24:0x7fbbca1016d0]
default	09:43:54.406647 -0700	DumpRenderTree	TIC TCP Conn Destroyed [22:0x7fbbca79ad40]
default	09:43:54.406970 -0700	DumpRenderTree	TIC TCP Conn Destroyed [24:0x7fbbca1016d0]
default	09:44:20.479517 -0700	DumpRenderTree	TIC TCP Conn Event [23:0x7fbbca601770]: 3 Err(60)
error	09:44:20.479702 -0700	DumpRenderTree	TIC TCP Conn Failed [23:0x7fbbca601770]: 1:60 Err(60)
default	09:44:20.479859 -0700	DumpRenderTree	TIC TCP Conn Cancel [23:0x7fbbca601770]
default	09:44:20.480167 -0700	DumpRenderTree	TIC TCP Conn Destroyed [23:0x7fbbca601770]
error	09:44:20.481225 -0700	DumpRenderTree	HTTP load failed (error code: -1001 [1:60])
error	09:44:20.481645 -0700	DumpRenderTree	NSURLConnection finished with error - code -1001

It seems that in some cases, we got some ENOTCON/ETIMEDOUT errors leading to NSURLConnection failing with timeout.

[Alexey Proskuryakov]

Is there anything to undo here (remove logging, unskip tests)?

[youenn fablet]

We should unskip the event-data.html test.
The logging will be removed next time we resync WPT tests

[youenn fablet]

Created attachment 320912 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 320912 [details]
Patch

Clearing flags on attachment: 320912

Committed r222094: <http://trac.webkit.org/changeset/222094>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.