[isdaniel]

Summary

Release 0.10.0. This PR only bumps the version (Cargo.toml, the pg-walstream-macros pin and the lockfiles) and renames the README's "Unreleased" section to "Upgrading from 0.9". Everything below is already on main since v0.9.0.

Breaking changes (details in README → "Upgrading from 0.9")

• ChangeEvent has a new public field xid: Option<u32> (Keep the subtransaction xid of streamed changes #149).
• ChangeEvent::encode / decode, deprecated since 0.9.0, are removed. Use serde_json (Remove the deprecated ChangeEvent binary codec #146).
• StreamingReplicationMessage::xid inside a streamed chunk is now the xid pgoutput wrote before the change, which is a subtransaction's xid after a savepoint (Keep the subtransaction xid of streamed changes #149).
• LogicalReplicationStream::new now runs IDENTIFY_SYSTEM (one more round trip, and it can fail with PermanentConnection) (Expose startup metadata from the stream's own connection and make startup non-blocking #141).
• send_feedback() before start() returns ReplicationError::Protocol. After start() and before the first message it sends an all-zero status update instead of returning early (Send a status update before the stream has received anything #147).
• A cancelled start(), slot_confirmed_flush_lsn() or timeline_history() now poisons the connection and later commands return PermanentConnection (Expose startup metadata from the stream's own connection and make startup non-blocking #141, Stop connect_async's host walk where libpq stops it #153).
• EventStream::next_event carries a reconnect across a dropped future. While one is parked, inner(), inner_mut() and current_lsn() panic until the next next_event() returns (Stop connect_async's host walk where libpq stops it #153).
• connection_timeout = Duration::ZERO now means no budget and retries as usual, instead of failing after the first attempt (Stop connect_async's host walk where libpq stops it #153).
• The libpq feature now also builds socket2.
• Dependency floors rose: tokio 1.54.1, tokio-util 0.7.20, smallvec 1.16.4, pq-sys 0.7.8, rustls 0.23.46, tokio-rustls 0.26.6, webpki-roots 1.0.11, aws-lc-rs 1.19.0; pg-walstream-macros needs syn 3.0.9 (chore: update dependencies to latest versions for improved stability #122, Update dependencies and improve async handling in libpq connection #124, Update dependencies: bump versions for tokio, smallvec, webpki-roots, and aws-lc-rs #151).

New

• Require an existing slot: with_existing_slot_required(true) turns a missing slot into a permanent error instead of creating it (Allow streams to require an existing replication slot #140, closes Option to refuse creating a missing replication slot #139).
• Startup metadata and non-blocking startup: server_identity(), slot_confirmed_flush_lsn() and timeline_history() are read on the stream's own connection. Both backends gain _async commands, so new, start, reconnects and snapshot() no longer block the executor. A cancelled snapshot() drops the slot it created (Expose startup metadata from the stream's own connection and make startup non-blocking #141).
• Snapshots of tables added to a running stream: AlignedSnapshot::capture / capture_aligned, EventType::commit_lsn() and next_event_before(C) on LogicalReplicationStream and EventStream (Align snapshots of tables added to a running stream and split the stream at them #142, closes Initial copy of a table added while the stream is already running #134).

Fixes

• A stream that is started but not yet read can now be kept alive with send_feedback() (Send a status update before the stream has received anything #147).
• A rolled-back savepoint's changes are dropped in the aligned merge, and streamed changes keep their subtransaction xid (Keep the subtransaction xid of streamed changes #149).
• connect_async resolves a host name only when its turn comes, so a slow later name no longer delays the first host (fix libpq lazy host resolution issue. #150).
• connect_async stops its host walk where libpq does, so a bad password is tried once, not once per address (Stop connect_async's host walk where libpq stops it #153).
• Reconnect: a timeout shorter than a reconnect no longer starves it, and cancelling during a stuck reconnect returns immediately (Stop connect_async's host walk where libpq stops it #153).
• AlignedSnapshot::capture now errors instead of returning an empty table when a requested table was truncated, rewritten or recreated after the snapshot, and refuses a side connection that landed on another cluster (Stop connect_async's host walk where libpq stops it #153).

Performance

• next_event future slimmed by keeping reconnect state out of it, undoing a size regression from the startup work (Keep reconnect state out of the next_event future #152).

CI / tooling

ClusterFuzzLite fuzzing (#120, #131), integration suites and the libpq backend counted in coverage (#144), Node 24 actions with the CodSpeed job pinned to ubuntu-24.04 (#132), lockfile bumps for cargo audit advisories (#135).

Related issues

Closes #139, #134 (already closed by their PRs; listed for the release notes). Related PRs: #140, #141, #142, #146, #147, #149, #150, #152, #153.

Type of change

• Bug fix
• New feature
• Performance improvement
• Refactor / internal cleanup
• Documentation
• CI / build (version bump / release)
• make before-git-push passes locally (check, build, format, audit, test, doc-check)
• Tests added/updated for the change (no code change; the bump adds no tests)
• Both backends build (cargo check --locked resolves every lockfile, root, examples, fuzz and load-tests)
• Docs / README updated if public API changed ("Upgrading from 0.9", install snippets now 0.10)

[codspeed]

Merging this PR will not alter performance

✅ 50 untouched benchmarks

---

_{Comparing rel/v0.10.0 (cc0b1a5) with main (f706d1f)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.44%. Comparing base (f706d1f) to head (cc0b1a5).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #154      +/-   ##
==========================================
+ Coverage   96.43%   96.44%   +0.01%     
==========================================
  Files          36       36              
  Lines       29290    29290              
==========================================
+ Hits        28246    28249       +3     
+ Misses       1044     1041       -3     

Flag	Coverage Δ
integration	64.35% <ø> (+<0.01%)	⬆️
unit	93.39% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.