persist: move blob decoding to isolated runtime to avoid heartbeat starvation #34699
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
…starvation These tests demonstrate that CPU-bound decode work can block heartbeat tasks when run on the main tokio runtime, but not when run on the isolated runtime. The `decode_blocking_starves_heartbeat_task` test shows that synchronous CPU-bound work on a single-threaded runtime delays heartbeats by 800ms+. The `isolated_runtime_does_not_block_heartbeat` test shows that the same work on the isolated runtime keeps heartbeat delays under 12ms. This validates the approach of moving blob decoding to the isolated runtime to prevent persist reader lease expirations. Co-Authored-By: Claude Opus 4.5 <[email protected]>
This change moves blob decoding from the main tokio runtime to the isolated runtime, following the same pattern already used for encoding. Previously, decode_batch_part_blob was synchronous and ran on the calling runtime. When multiple large blob decodes occurred concurrently, they could saturate runtime workers and prevent heartbeat tasks from running, leading to persist reader lease expirations. The fix pre-decodes hollow batch parts in BatchFetcher::fetch_leased_part on the isolated runtime before creating FetchedBlob. This ensures the heavy CPU work happens on separate OS threads where the scheduler can context-switch, preserving liveness for heartbeat tasks. Key changes: - decode_batch_part_blob is now async and spawns work on isolated runtime - FetchedBlobBuf::Hollow stores the pre-decoded EncodedPart instead of raw bytes - EncodedPart now implements Clone to support FetchedBlob Clone - isolated_runtime is threaded through ReadHandle, Consolidator, etc. - Various T: Sync bounds added where required by the async decode Co-Authored-By: Claude Opus 4.5 <[email protected]>
antiguru
reviewed
Jan 14, 2026
bosconi
changed the title
bosconi
changed the title
bosconi
changed the title
Member
Author
|
@antiguru code comments have been expanded to explain the motivation more. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This change moves blob decoding from the main tokio runtime to the
isolated runtime, following the same pattern already used for encoding.
Previously, decode_batch_part_blob was synchronous and ran on the
calling runtime. When multiple large blob decodes occurred concurrently,
they could saturate runtime workers and prevent heartbeat tasks from
running, leading to persist reader lease expirations.
The fix pre-decodes hollow batch parts in BatchFetcher::fetch_leased_part
on the isolated runtime before creating FetchedBlob. This ensures the
heavy CPU work happens on separate OS threads where the scheduler can
context-switch, preserving liveness for heartbeat tasks.
Key changes: