perf: reduce cacheline ping pong in workers availability#23321
Merged
Conversation
pepyakin
requested review from
Rjected,
gakonst,
mattsse,
mediocregopher,
shekhirin and
yongkangc
as code owners
Contributor
|
mediocregopher
approved these changes
Apr 1, 2026
github-merge-queue Bot
pushed a commit
to tempoxyz/tempo
that referenced
this pull request
Apr 3, 2026
Automated nightly update of reth dependencies from `paradigmxyz/reth` main branch. ## Upstream reth changes [`f8efc76...c82d435`](paradigmxyz/reth@f8efc76...c82d435) 🔗 Amp thread: https://ampcode.com/threads/T-019d517e-981b-75fd-b3cf-8ad2455f38f6 **RPC** - Remove `0x` prefix from `admin_peers` id and return keccak256 node ID in `admin_nodeInfo.id` to match go-ethereum format ([#23318](paradigmxyz/reth#23318), [#23319](paradigmxyz/reth#23319)) - Apply count filter only after `after` is consumed in pagination ([#23338](paradigmxyz/reth#23338)) - Pre-allocate vectors in `eth_feeHistory` ([#23334](paradigmxyz/reth#23334)) - Add `bal` RPC methods ([#23330](paradigmxyz/reth#23330)) **Engine** - Include backpressure in reported `persistence_wait`, make wait-time mimic CL slot time ([#23308](paradigmxyz/reth#23308)) **Perf** - Reduce cacheline ping-pong in workers availability ([#23321](paradigmxyz/reth#23321)) **Net** - Avoid itertools `Format` panic in tracing log ([#23331](paradigmxyz/reth#23331)) **DB** - Cap `storage_v2` unwind history by MDBX tip ([#23335](paradigmxyz/reth#23335)) **Bench** - Add warmup step for big blocks mode ([#23323](paradigmxyz/reth#23323)) - Make payload handling ethereum-only ([#23324](paradigmxyz/reth#23324)) - Use repository instead of GitHub Actions cache to track hourly runs ([#23306](paradigmxyz/reth#23306)) - Upload nightly regression results to ClickHouse ([#23344](paradigmxyz/reth#23344)) **Chore** - Migrate `allow(clippy::` to `expect(clippy::` ([#23340](paradigmxyz/reth#23340)) - Remove unused return value from `dispatch_with_chunking` ([#23341](paradigmxyz/reth#23341)) - Update nixpkgs to 25.11 and refresh flake inputs ([#23342](paradigmxyz/reth#23342)) ## Migrations 🔗 Amp thread: https://ampcode.com/threads/T-019d517e-baf0-727b-a57e-0488240af973 - **Reth dependency bump**: All `reth-*` git dependencies updated from rev `f8efc76` to `c82d435` - **New `GetBlockAccessList` trait impl**: `TempoEthApi` now implements `GetBlockAccessList` (imported from `reth_rpc_eth_api::helpers::bal`), likely a new required trait added upstream in the reth update [GitHub Workflow](https://github.com/tempoxyz/tempo/actions/runs/23933027282) --------- Co-authored-by: Arsenii Kulikov <klkvrr@gmail.com>
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
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.
Every time a core updates the availability counter it claims the cacheline invalidating it on other cores. Workers essentially ping pong this cacheline to each others. This showed up during perf c2c profiling.
The solution is to have each worker report its status on its own cacheline. When a non-worker core wants to know the status/counts, it will request the cachelines making them shared temporarily. This path should become a bit more expensive but AFAIU it's also a less frequent one.
At least that's the theory.
This is reopening of perf: reduce cacheline ping pong in workers availability #23287. This time we got a better evidence that it's actually improving things, see: