Repository navigation
[ML] Reap per-deployment ml-child-ipc directories after pytorch_inference exits #3214
Description
Summary
The native controller now creates $TMPDIR/ml-child-ipc/<deployment-id> (mode 0700) before spawn. Nothing removes the directory when the deployment ends; FIFOs are unlinked by pytorch_inference, leaving empty directories behind.
Acceptance
- Define ownership (Elasticsearch process teardown vs controller) and remove the per-child directory when the deployment is finished.
- Document the lifecycle alongside
ensureChildIpcDirectory().
Related: #3188
Activity
[ML] Reap per-deployment ml-child-ipc directories when pytorch_infere…
[ML] Reap per-deployment ml-child-ipc directories when pytorch_infere…
[ML] Reap per-deployment ml-child-ipc directories when pytorch_infere…
[ML] Reap per-deployment ml-child-ipc directories when pytorch_infere…
[ML] Reap child IPC dirs and fail closed on degraded seccomp (#3217)