Skip to content

Mount memory-backed /dev/shm emptyDir volume in PathwaysJobSet.add_user_workload. - #346

Open
copybara-service[bot] wants to merge 1 commit into
mainfrom
test_1000409655
Open

copybara-service[bot] wants to merge 1 commit into
mainfrom
test_1000409655

Conversation

@copybara-service

@copybara-service copybara-service Bot commented Oct 15, 2026 •

Copy link
Copy Markdown

Mount memory-backed /dev/shm emptyDir volume in PathwaysJobSet.add_user_workload.

Multiprocess Python data loaders such as Grain use shared memory (/dev/shm) for inter-process communication. Without an explicit memory-backed emptyDir volume mounted at /dev/shm on the pathways-head user workload container, containerd defaults /dev/shm to 64 MiB, causing Grain worker processes to crash with SIGBUS (exit code -7) when buffering batches across workers.

Update PathwaysJobSet.add_user_workload() to mount a memory-backed emptyDir volume (dshm, medium: Memory) at /dev/shm by default when /dev/shm is not already mounted.

@copybara-service
copybara-service Bot force-pushed the test_1000409655 branch 3 times, most recently from de7fd36 to de2c136 Compare October 16, 2026 20:16
@copybara-service copybara-service Bot changed the title Mount memory-backed /dev/shm emptyDir volume in PathwaysJobSet.add_user_workload Mount memory-backed /dev/shm emptyDir volume in PathwaysJobSet.add_user_workload. Oct 16, 2026
@copybara-service
copybara-service Bot force-pushed the test_1000409655 branch 3 times, most recently from ea6b764 to 847898e Compare October 19, 2026 20:24
…er_workload.

Multiprocess Python data loaders such as Grain use shared memory (`/dev/shm`) for inter-process communication. Without an explicit memory-backed `emptyDir` volume mounted at `/dev/shm` on the `pathways-head` user workload container, containerd defaults `/dev/shm` to 64 MiB, causing Grain worker processes to crash with `SIGBUS` (exit code `-7`) when buffering batches across workers.

Update `PathwaysJobSet.add_user_workload()` to mount a memory-backed `emptyDir` volume (`dshm`, `medium: Memory`) at `/dev/shm` by default when `/dev/shm` is not already mounted.

PiperOrigin-RevId: 1000409655

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant