[multicast] fix cache invalidation to work across Nexus instances #9742
+235
−82
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The previous implementation used an AtomicBool to signal cache invalidation from sled_expunge()/sled_upsert() to the multicast background task. This only affected the Nexus that received the request, while other Nexus instances would never know to invalidate their caches. As per discussion, this pattern of out-of-band communication with background tasks is something we aught to avoid: if a task accepts input to determine what to do, it must separately handle cases where that input is unavailable (after restart, or when another Nexus instance receives the request). This bifurcates the code between the common case and rarely-exercised fallback paths. This PR replaces the AtomicBool approach with inventory collection ID tracking. The reconciler now checks if the inventory collection ID has changed (database-driven), and invalidates caches when it detects a new collection. This works across all Nexus instances since the DB is the source of truth. Now, the cache refresh always uses SledFilter::InService, which excludes expunged sleds. This provides a safety net even if/when cache invalidation is delayed, as the next cache refresh will exclude any expunged sleds. This update also includes a new test and test extensions around cache invalidation involving the multicast background task.
jgallagher
reviewed
Jan 29, 2026
jgallagher
approved these changes
Jan 29, 2026
Contributor
jgallagher
left a comment
jgallagher
left a comment
There was a problem hiding this comment.
LGTM, just a couple tiny nits. Thanks for the quick turnaround here and putting up with all my questions.
zeeshanlakhani
force-pushed
the
zl/mcast-task-cache-handle-fix
branch
from
1291dbb to
4d8a579
Compare
…nnel - Track sled baseboard->sp_slot mappings instead of collection IDs - Only invalidate when sled physical locations change (rare), not on every inventory collection (frequent) - Use existing inventory watch channel pattern shared by background tasks (not DB) - Update tests to activate inventory_loader before reconciler - Update TODO with DDM information - fix nits
zeeshanlakhani
force-pushed
the
zl/mcast-task-cache-handle-fix
branch
from
bc05306 to
6885e4d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The previous implementation used an AtomicBool to signal cache invalidation from sled_expunge()/sled_upsert() to the multicast background task. This only affected the Nexus that received the request, while other Nexus instances would never know to invalidate their caches.
As per discussion, this pattern of out-of-band communication with background tasks is something we aught to avoid: if a task accepts input to determine what to do, it must separately handle cases where that input is unavailable (after restart, or when another Nexus instance receives the request). This bifurcates the code between the common case and rarely-exercised fallback paths.
This PR replaces the AtomicBool approach with inventory collection ID tracking. The reconciler now checks if the inventory collection ID has changed (database-driven), and invalidates caches when it detects a new collection. This works across all Nexus instances since the DB is the source of truth.
Now, the cache refresh always uses SledFilter::InService, which excludes expunged sleds. This provides a safety net even if/when cache invalidation is delayed, as the next cache refresh will exclude any expunged sleds.
This update also includes a new test and test extensions around cache invalidation involving the multicast background task.