1296358 - Force a JS GC in between reftests

Status: RESOLVED DUPLICATE of bug 1292218

(Testing :: Reftest, defect)

Description

[Mason Chang [Inactive] [:mchang]]

With Skia content on windows, on some reftest suits, the tests would occasionally out of memory and not be able to allocate a surface to render content into. Locally when running tests, I would see spikes from ~200mb to 1 gig before the GC would kick in and bring the memory usage back down to 200mb. This was rather variable as sometimes the spike would be 1 gig, sometimes 500 mb, sometimes 800mb. Instead of having random reftests fail due to OOM, let's GC in between some number of reftests.

Comment 1

[Mason Chang [Inactive] [:mchang]]

Attachment: Run the JS GC every 10 reftests

Thanks!

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Please coordinate with those working on bug 1292218.

Comment 3

[Mason Chang [Inactive] [:mchang]]

Here's a try push with skia content on windows (so not quite there yet) showing the reftests not always failing - https://treeherder.mozilla.org/#/jobs?repo=try&revision=7b875a2b2aa9&selectedJob=25792231

Here's what it was before this patch - https://treeherder.mozilla.org/#/jobs?repo=try&revision=81ef9ee53cd0&selectedJob=25423084

Here is a try push with only this patch on all platforms (should be ready in a couple of hours hopefully) = https://treeherder.mozilla.org/#/jobs?repo=try&revision=ba74ad510d76

Comment 4

[Andrew McCreight [:mccr8]]

Does my patch work instead? I don't really like forcing a full GC all the time if we can avoid it, because then we aren't actually testing incremental collection in reftests.

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

Comment on attachment 8782535 [details] [diff] [review]
Run the JS GC every 10 reftests

Review of attachment 8782535 [details] [diff] [review]:
-----------------------------------------------------------------

while this patch looks great technically- i think there is concern over doing this vs other efforts, lets resolve those concerns first.

Comment 6

[Andrew McCreight [:mccr8]]

This is basically the same thing as bug 1292218

Comment 7

[Andrew McCreight [:mccr8]]

Let me know if bug 1292218 doesn't suffice to fix the OOMs you are seeing.