Skip to content

Fix remote fetch and pin cleanup, and tighten test isolation - #26

Merged
mkhazraee merged 5 commits into
mainfrom
moein/touch_ups
Sep 10, 2026
Merged

Fix remote fetch and pin cleanup, and tighten test isolation#26
mkhazraee merged 5 commits into
mainfrom
moein/touch_ups

Conversation

@mkhazraee

@mkhazraee mkhazraee commented Sep 10, 2026

Copy link
Copy Markdown
Collaborator

Fix resource cleanup when remote fetch completion races with timeout or framework pin release fails.

  • Release discarded fetch slots when a successful completion arrives after main-thread expiry, preserving the failed result.
  • Retry failed framework pin releases during polling. Remove descriptors from usable state when release starts, and defer overlapping acquisitions until release succeeds or their original deadline expires.
  • Apply the same release handling to late and invalid pin results.
  • Give Guard integration tests explicit ZMQ context teardown and remove enforced test ordering.
  • Make the shared fake NIXL agent reject mismatched descriptor counts and sizes.
  • Avoid unused _BlockRecord allocations and repeated prefix slicing.
  • Document deferred eviction-heap growth with a TODO.

Regression coverage includes fetch completion timing and slot reuse, pin-release failures, blocked reacquisition and expiry, and discarded pin results.

@mkhazraee mkhazraee changed the title Fix remote fetch timeout cleanup and tighten test isolation Fix remote fetch and pin cleanup, and tighten test isolation Sep 10, 2026
@mkhazraee
mkhazraee requested a review from aranadive September 10, 2026 03:10
@mkhazraee
mkhazraee force-pushed the moein/touch_ups branch 2 times, most recently from 0ff9eb0 to a9d2332 Compare September 10, 2026 05:46
Signed-off-by: Moein Khazraee <moein@nvidia.com>
Signed-off-by: Moein Khazraee <moein@nvidia.com>
Signed-off-by: Moein Khazraee <moein@nvidia.com>
…cleanup

Signed-off-by: Moein Khazraee <moein@nvidia.com>

@karya0 karya0 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

Signed-off-by: Moein <moein@nvidia.com>
@mkhazraee
mkhazraee merged commit 2addabb into main Sep 10, 2026
3 checks passed
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.

2 participants