You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 76cd429
Browse filesBrowse the repository at this point in the historyBrowse files
authored
fix(relay): close a REQ whose history read failed instead of sending EOSE (#8692)
## Summary
**Merge order:** this lands and deploys **before**
[#8688](#8688). That PR's buzz-acp
resume cursor counts on a failed history read never ending in EOSE.
Until #8688 ships, buzz-acp re-sends a REQ as soon as it gets an error
CLOSED. The relay's REQ rate limit bounds that loop, and buzz-acp
already parks on a rate-limited CLOSED. #8688 replaces this with a
per-subscription backoff.
When a historical query fails for a reason other than a statement
timeout, the relay logs the error and still sends `EOSE`. EOSE means
"you have every stored event" (`ARCHITECTURE.md` §5, NIP-01). A client
that resumes from its last-seen point therefore treats the failed read
as complete, advances past it, and never replays that history. This is
RPR #1: a failure must not become an authoritative empty result.
The history task now returns the CLOSED reason to send:
- **Statement timeout:** `["CLOSED", sub_id, "error: query timed out"]`,
unchanged.
- **Any other DB error:** `["CLOSED", sub_id, "error: database error"]`.
- **Cleanup:** CLOSED goes through the existing `close_if_owner` path,
so the subscription, its sub-registry filters and its topic refcount are
released, the same as on a timeout.
- **Docs:** `ARCHITECTURE.md` §5 now states this contract.
**Clients don't hot-loop on it.** Checked at `main` 7c165c3:
- **Desktop:** `relayClosedPolicy.ts` / `relayClosedRecovery.ts`
classify `error: database error` as retryable. Live subscriptions back
off from 1s to 30s and keep their `since`. History-mode requests reject.
- **Mobile:** `relay_closed_policy.dart` / `relay_session.dart` do the
same, with their own capped backoff.
- **buzz-acp:** today it re-REQs at once, bounded by the relay's REQ
rate limit. [#8688](#8688) adds a
per-subscription backoff that also restores the loss record.
### What got simpler
- **One constant for the CLOSED reason.** `"error: database error"` was
a string literal repeated across `req.rs` (5×) and `count.rs` (4×). It
is now one constant, `DATABASE_ERROR_CLOSED`.
- **One mapping function.** `history_read_closed_reason` maps a DB error
to its CLOSED reason in one place.
### Identified debt (not changed here)
- **NIP-50 search path.** After a non-timeout full-text-search or
hydrate error, the search path still sends `EOSE` with partial results.
It needs the same treatment, but it has a separate result-assembly path,
so it's left out of this PR.
### Related issue
Adds the regression wpfleger96 asked for in [this review on
#8653](#8653 (review)):
a history error followed by a reconnect still replays the history.
### Testing
New Postgres test
`api::bridge::postgres_tests::failed_history_read_closes_and_reconnect_replays_the_history`:
1. Lock the replica's `events` table past its 300ms `statement_timeout`
and send a history REQ. It must get exactly `["CLOSED","t","error: query
timed out"]`.
2. Break the `events` table on both the writer and replica pools and
send a REQ. It must get exactly `["CLOSED","h","error: database
error"]`.
3. After each CLOSED, the subscription map, the sub-registry filters and
the topic refcount must all be cleared.
4. Restore the table and send the same REQ on a new connection. It must
get the stored event, then `EOSE`.
Run on Blox at 95a8255:
- `cargo fmt --check`
- `cargo clippy -p buzz-relay --all-targets -D warnings`
- Postgres lane via `scripts/postgres-test-run.sh -p buzz-relay`
(499/499)
- The new test and `routed_read_cancel_maps_to_timeout_contract` 5×
- **Mutation checks:** each of these makes the new test fail at its
frame assertion.
- Restoring `EOSE` for a non-timeout error.
- Mapping every failure to `error: database error`.
- **`just test` integration half:** run by hand, because
`_ensure-services` can't start its Redis container on that workstation
while a native redis-server holds the port.
- Workspace `cargo test --test '*'`: every binary passes except two
`buzz-acp` `git_bootstrap` tests.
- Those two fail the same way on that workstation without this change,
so they come from the environment.
- `buzz-db` and `buzz-auth` integration tests pass.
End-to-end against a locally built relay at e5621f9, with buzz-acp from
#8688:
- **While `events` was missing:** every channel REQ got `CLOSED error:
database error`, and no channel EOSE was sent.
- **After the table was restored:** the next retry replayed the missed
mention, followed by EOSE.
The later commit only changes tests and docs.
buzz-review-completed
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Signed-off-by: Luke Tornquist <[email protected]>
Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>
Co-authored-by: Claude Opus 5.5 <[email protected]>
Copy file name to clipboardExpand all lines: ARCHITECTURE.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -378,7 +378,7 @@ This prevents a race where a non-member receives live fan-out events from a priv
378
378
379
379
### Historical Query (EOSE)
380
380
381
-
After registering, the REQ handler queries Postgres for stored events matching the filters (up to 500 per filter, hard cap). These are sent as `["EVENT", sub_id, event]` frames before `["EOSE", sub_id]`. New events arriving after EOSE are delivered via the fan-out path.
381
+
After registering, the REQ handler queries Postgres for stored events matching the filters (up to 500 per filter, hard cap). These are sent as `["EVENT", sub_id, event]` frames before `["EOSE", sub_id]`. New events arriving after EOSE are delivered via the fan-out path. For an ordinary (non-search) REQ, EOSE marks a successful bounded history query, so a failed history read never ends in EOSE: it retires the subscription with `["CLOSED", sub_id, "error: query timed out"]` for a statement timeout or `"error: database error"` for anything else.
382
382
383
383
**Client consumption invariant.** A client rebuilding channel state must
384
384
open its live subscription before (or overlapping) the finite history
0 commit comments