Mochitest runs hang when a browser session navigates the test frame more than 200 times within 10 seconds
Categories
(Testing :: Mochitest, defect)
Tracking
(Not tracked)
People
(Reporter: jstutte, Unassigned, NeedInfo)
References
Details
Attachments
(1 obsolete file)
Since bug 1922677, every non-system navigation counts against the navigation rate limit (dom.navigation.navigationRateLimit.count, 200 per browsing context in dom.navigation.navigationRateLimit.timespan, 10 s). When the limit is hit, nsDocShell::InternalLoad drops the navigation and returns NS_OK, and the only trace is a console error that does not reach the test log.
TestRunner navigates the test frame twice per test: to the test, then to iframe-between-tests.html. A browser session that runs about 100 short tests within 10 s therefore exceeds the limit, the iframe.src for the next test is dropped, and the run hangs until the harness times out.
This showed up for bug 2051892, which makes timers fire more precisely: the first 100 tests of a mochitest-webgl2-ext session on macOS aarch64 opt went from 14-24 s to about 9.5 s, and every run then hung at the 100th test (shaders-with-mis-matching-uniforms.html) with the test frame still on iframe-between-tests.html.
Bug 1922677 itself worked around the same hangs by setting dom.navigation.navigationRateLimit.count=0 in about 15 test manifests and a crashtest list, and bug 2051892 adds the WebGL conformance manifests. Each of these places carries a comment pointing here. That runs those suites with a different configuration than we ship, and any other suite of fast tests can hit the same hang as machines or chunking change.
Once the harness is fixed, the count=0 overrides should be removed again.
| Comment hidden (obsolete) |
| Comment hidden (obsolete) |
Comment 3•1 month ago
|
||
The severity field is not set for this bug.
:florian, could you have a look please?
For more information, please visit BugBot documentation.
Description
•