[dbccpage]

Requested by Jeremy · project thread

Stacked on #49 (which is stacked on #48). Only the last commit is new here.

What

• Authority.publish(..., observed_at=None): optional freshness in authority ms. It moves doc["updated"] forward (never back) and refuses a future observation as invalid_freshness_observation before commit.
• Bootstrap passes the authority time read before CREATE_REPLICATION_SLOT: every commit before it is inside the exported snapshot.
• Apply passes the poll's view time only when the peek showed this transaction was the last admitted one through the frontier. peek now reads two rows and returns (payload, offset, exhausted). A backlogged apply still does not reset freshness (test_backlog_publication_does_not_reset_source_freshness unchanged).
• advance_idle passes it too.

Evidence (local PG 16.15 + wal2json + Redis 7)

New test_production_continuous_writes_keep_bounded_reads_fresh_past_max_lag: 35 s of writes at 4 tx/s with the idle path (advance_idle + caught_up) disabled, which models "a commit in every poll". Bounded hit rate in [30 s, 35 s]:

• parent branch: 0% (661/661 head_miss; the head is withdrawn with source_lag_limit)
• this PR: 100% (722/722)

Unit test test_publication_with_an_observation_advances_freshness. Full native suite (non-cluster, non-disruptive): 58 passed. test_production_read_contention_during_publication[bounded] failed once in 2 runs. It is the known drain-rate flake: about 25 tx/s drain against 20 writes/s, unchanged by this PR (phase profile 24.3 to 26.4 tx/s on both branches). The fix for that is the batching in #47.

Not in this PR

• WAL batching (the other half of RD-3) is Coalesce committed WAL transactions into one generation per poll #47.
• "Persist the worker's SignedMirror across polls": tests/production_native.py::bounded() already builds one SignedMirror per test, and the CLI does not use signed heads. I found no per-poll instantiation to fix.

---

Generated by Claude Code

[dbccpage]

Live Database Containers failed on one test only: test_production_read_contention_during_publication[bounded] ("native source did not catch up within 30 seconds"). Every other native test passed, including all the new ones in this stack.

This failure is not from this PR. It is the known drain-rate limit on main: the worker publishes one generation per WAL transaction, about 13–25 tx/s, while the test writes 20 tx/s. main has failed the same way (run 42984776547). The same job passed on #48 and #50. #47 fixes it (coalescing committed WAL transactions into one generation per poll). I have not copied that fix here, to keep these PRs small. I will re-run the job once, when the build jobs in this run finish.

---

Generated by Claude Code