Repository navigation
GH-126491: Increase the threshold for the GC fast cycle test. - #126984
freakboy3742 wants to merge 2 commits into
Conversation
Sorry, something went wrong.
|
!buildbot iOS |
Sorry, something went wrong.
|
🤖 New build scheduled with the buildbot fleet by @freakboy3742 for commit af3c692 🤖 The command will test the builders whose names match following regular expression: iOS The builders matched are:
|
Sorry, something went wrong.
|
!buildbot Android |
Sorry, something went wrong.
|
🤖 New build scheduled with the buildbot fleet by @freakboy3742 for commit af3c692 🤖 The command will test the builders whose names match following regular expression: Android The builders matched are:
|
Sorry, something went wrong.
|
There's still a test_incremental_gc_handles_fast_cycle_creation failure on iOS/Android. (When this is passing, let's queue up a full buildbot session with the 🔨 test-with-buildbots or refleaks label as well.) |
Sorry, something went wrong.
... wat? Now they're both failing with the same count of 26250? I'll bump to 27k as originally suggested... but Murphy being who he is, I'm guessing that means they'll now both fail with a count 27300... 🤣 |
Sorry, something went wrong.
|
!builbot iOS |
Sorry, something went wrong.
|
!buildbot Android |
Sorry, something went wrong.
|
🤖 New build scheduled with the buildbot fleet by @freakboy3742 for commit 9d5a064 🤖 The command will test the builders whose names match following regular expression: Android The builders matched are:
|
Sorry, something went wrong.
|
!buildbot iOS |
Sorry, something went wrong.
|
🤖 New build scheduled with the buildbot fleet by @freakboy3742 for commit 9d5a064 🤖 The command will test the builders whose names match following regular expression: iOS The builders matched are:
|
Sorry, something went wrong.
|
Looks like Bevedere has obliged us with a full buildbot pass regardless... |
Sorry, something went wrong.
|
Ok... I was kidding when I predicted the test failure mode... What on earth is going on? |
Sorry, something went wrong.
|
I think I see what is going on now. My original assertion that increasing the loop count to 100k didn't change anything was based on having the assertion still set at 25k, and seeing the same 25,032 failure count. I see now that the GC count is increasing with more iterations. Checking now to see how high the threshold actually is... |
Sorry, something went wrong.
|
After further testing, it looks like the limit is a lot more than 30k... but the good news (for me) is that I can reproduce the problem on macOS. Closing this PR; I'll put notes back in the original ticket. |
Sorry, something went wrong.
#126502 Modified added an improvement to the behavior of the garbage collector, along with a test to verify that behavior. However, that test fails on iOS and Android (and quite possibly any test run that runs the test suite sequentially); the test imposes a limit of 25k objects, but iOS and Android both generate 25032 objects.
The 32 extra object count seems to be something baked in - if you increase the iteration count of the test to 100k, you get the same 25032 object count on iOS and Android.
Following this comment, this PR increases the threshold to 26k 27k.