-
Notifications
You must be signed in to change notification settings - Fork 3.1k
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
Tasks going OOM on Taskcluster #12874
Comments
I would be interested to know if we see this on tasks run on wpt-docker-worker; we can possibly adjust the instance type there if OOM is an ongoing issue. |
Where does the memory limit come from? |
Presumably from the AWS instance type. |
According to the logs, we're on a c3.xlarge, which doesn't exist in the AWS documentation… Third party documentation suggests that it has 7.5GB RAM, which should be plenty? |
@gsnedders has this issue recurred or is it safe to close it? |
@mdittmer I've seen it somewhat infrequently, but given we don't really have any decent way to notice it at the moment (as it doesn't fail the jobs). |
To estimate how frequently this issue occurs, I downloaded the logs for the 10 most recent commits to
Searching for "Cannot allocate" turned up two results:
The sample size is pretty small, but bear in mind that we run WPT in 15 "chunks" per browser. That means it occurred in two of @gsnedders' 15 tasks and 0 of WPT's 150. This supports @jgraham's earlier theory that this is related to the workerType and therefore not influential for the typical use cases of pull request validation and results uploading. |
Annoying though, given we did want it to be possible to run on people's own forks. |
This should no longer happen at least in this repo after #14290 . However, unfortunately, OOM is still possible on forks. We do have special (larger) memory guarantees on wpt-docker-worker (#13989 (comment)). Unless we drastically reduce the memory footprint of WPT (especially manifest), it will continue to be possible to run OOM on the default worker whose memory might be less than 4G. |
e.g., https://tools.taskcluster.net/groups/f7409aGOTa6OJLwvyw7igw/tasks/N0HriXt3TH62lxQhwftRlg/details went OOM (in the job for #12380)
The suite ends with:
If we're having this happen several times on one commit, then it seems likely this will happen semi-regularly, which makes it hard to rely on for PRs or anything else.
The text was updated successfully, but these errors were encountered: