Repository navigation
Forward-port #912, #917 and #911 (0.6.1 IBD and drain fixes) to master - #921
Merged
Merged
Conversation
…912) * Report blocks the stored-tail drain connects as chain events (#900) The drain connects a stored block once its parent is the tip: a block that arrived before its parent, or the tail an IBD batch leaves behind. It connected through connect_stored_block, which emits nothing, so -blocknotify, block announcement, Electrum and Esplora subscribers, the streaming API and ZMQ hashblock never heard of that block. The drain now emits BlockConnected after the connect and after the mempool has dropped the block's transactions, as accept_block does. ChainState::emit_chain_event becomes pub(crate) for it; connect_stored_block still emits nothing, since the IBD connector uses it too and stays quiet. The drain keeps its own -stopatheight check: the watcher in main now hears of drained blocks, but only after the walk has moved on. The watcher uses send_if_modified, so when both ask for shutdown the second neither logs nor wakes shutdown waiters again. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * Report a drained block before releasing the accept lock (#900) The stored-tail drain emitted BlockConnected after connect_stored_block had released accept_lock. accept_block reports a connect while it still holds the lock, so a submitblock of the drained block's child could connect and report in between, and subscribers heard of the child before its parent. connect_stored_block_and_report now purges the mempool and emits under the lock, as accept_block does, and emit_chain_event is private again. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> (cherry picked from commit 6856ee9)
* p2p: run the manager loop's maintenance on time under load (#909) The manager loop's maintenance pass holds IBD stall detection: the release of a height stuck with one peer (15 s at the connect cursor), the drop of silent peers, and work for idle peers. Its cadences count passes ("every 4 ticks (2s)"), so it has to run every 500 ms. Under initial block download it did not. The fast drain added in #781 set drain_only, so every run of fast drains ended in a wait for the next tick, and the pass after that tick found the queue full again. With the manager storing each arriving block, an fsync apiece, the queue is full at every wake. Measured on a fresh mainnet sync over two minutes, the pass ran 3-4 times instead of ~240, the stale release never fired, and the block at the connect cursor sat with one slow peer for up to 53 s. DrainPacer decides after each drain. Maintenance is due on an interval tick, as before, or once 500 ms have passed since it last started, and then runs after the drain in progress however full the queue is. A full queue otherwise goes straight back to draining (#781's throughput), and does so after maintenance too. A drain_now wake is neither a tick nor late, so it drains and waits, as #776 requires. The interval uses MissedTickBehavior::Delay, so the ticks a busy stretch skipped do not fire as a burst of passes. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * docs(release-notes): cite only runs of the shipped #909 fix Six runs of a release build of this branch against nine of the 0.6.0 release binary, replacing figures that mixed in earlier versions of the fix and an instrumented build. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> (cherry picked from commit 89ce0e5)
…911) * IBD: warn that the connector is stuck only after a 10 s wait (#904) The connector logged "Connector stuck waiting for block data" at WARN the first moment it waited on any height. Blocks arrive out of order and the connector takes them in height order, so early in a healthy sync that was nearly one warning per block: 112 of the first 624 log lines on a fresh mainnet sync, 490 of 508 in a fresh Umbrel install. StuckWait tracks which height the connector waits on and since when. The warning fires once that wait reaches 10 s, then every 60 s while it lasts, and carries waited_secs. A new height, or a successful connect, starts over. In the #904 run the waits worth a warning had the block in flight 23-60 s on one peer; 97 of the 112 warnings were under 5 s. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * IBD: raise the stuck-wait warning to 20 s, above the stale-block release (#904) The scheduler takes a block at the connect cursor back from a slow peer after 15 s and asks another. A 10 s threshold warned during waits that release was about to end; 20 s warns only once it has not. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> (cherry picked from commit 98bca53)
This was referenced Oct 7, 2026
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
1 participant
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.
Brings three fixes merged into
release/0.6for 0.6.1 back tomaster, one commit each, cherry-picked with-x:6856ee98, fixes Blocks connected by the stored-tail drain emit no chain event: -blocknotify, subscriptions and block announcement miss them #900): the stored-tail drain reports each block it connects withBlockConnected, purging the mempool and emitting underaccept_lockasaccept_blockdoes.-blocknotify, block announcement, Electrum, Esplora SSE, streaming and ZMQ now hear of a block that reached the tip that way.89ce0e5a, fixes IBD: the block at the connect cursor waits up to 60 s on one slow peer; the manager loop's fast drain skips stall detection #909): the P2P manager loop runs its maintenance pass on time under a full event queue, so the 15 s release of a block stuck with one peer fires during initial sync again.98bca53e, fixes IBD logs "Connector stuck waiting for block data" at WARN for nearly every block early in sync #904):Connector stuck waiting for block datawarns only after a 20 s wait on one block, then once a minute, withwaited_secs.No
mastercommit since therelease/0.6fork touches the files these change, so all three apply unchanged. The code patch is identical to the one onrelease/0.6:git patch-idofe10c19e1..98bca53e(code paths) equals that of this branch's diff.Release notes
The three
CHANGELOG.mdbullets and theirdocs/release-notes/0.6.1-pre.mdsections stay onrelease/0.6. They describe 0.6.1 and will come tomasterwith the 0.6.1 cut, as #901 brought 0.6.0 and #910 left #899's. Nothing is added to0.7.0-pre.md.Merging
"Rebase and merge" keeps the three commits separate, each with its
cherry picked fromline.Verified
Ran locally on this branch:
cargo clippy --all-targets --all-features --locked -- -D warnings;cargo test --test e2e --no-run --locked --features e2e;nodelib suite (2132 passed);-blocknotify,-stopatheightand ping/pong/sync tests (11), on the first two commits.On
release/0.6, each PR passed CI and a Devin review before merging. Push CI passed on the tree after #917.🤖 Generated with Claude Code