M10.9d: batch durable action lookups (one per poll, one per commit)
Foundation / checks (deploy) (push) Successful in 3m27s
Foundation / checks (ecs) (push) Successful in 1m57s
Foundation / checks (client) (push) Successful in 12m41s
Foundation / checks (network) (push) Successful in 13m33s
Foundation / checks (tsan) (push) Successful in 7m54s
Foundation / checks (checks) (push) Successful in 32m56s
Foundation / checks (postgres) (push) Successful in 46m10s
Foundation / postgres-tools (push) Successful in 18s
Foundation / image (push) Successful in 2m33s
Foundation / server-image (push) Successful in 4m24s

The fourth dev drill showed a synchronous PostgreSQL round trip per action on the
simulation thread: at admission (uniqueness) and again at commit (first-record rule).
About 12 lookups landed on each commit tick (commit ticks 58-70 ms, session p99 30-39 ms).

- ActionLedger::find_all; persistence::find_region_actions applies the schema v11
  find_region_action to many pairs in one statement (no migration).
- IntentAdmission::admit/verify: actions only the ledger can refuse wait for one verify
  per poll; receipts still go out in the same poll. committed() asks once per commit.
- Tests: deferred_actions (network_control), find_region_actions and find_all (Postgres).
- ADR-0097 fourth run, ADR-0092 amendment, VALIDATION; refresh the dev session window.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
dingo
2026-10-08 21:35:43 -04:00
co-authored by Claude Opus 5.5
parent 3ee22694cc
commit 07d12a4d82
13 changed files with 430 additions and 45 deletions
@@ -129,3 +129,20 @@ It is cumulative, so no admitted intention is ever missing from a later watermar
learns "simulated" up to 3 ticks (150 ms) later. Receipts are unchanged: one per intention.
Reason: on the reference worker, datagram sends dominated the session layer, and the
per-tick watermark was a quarter of them.
## Amendment (2026-10-09, M10.9d, ADR-0097): batched durable lookups
The durable checks are unchanged; only how many store round trips they cost changes.
- **Admission.** In evo-server, an action that passes every in-memory check waits for the
poll's single `verify`, which asks the ledger about all waiting actions at once and
sends their receipts in the same poll.
- While it waits, the action counts toward the unresolved limit.
- A repeat of it is a `duplicate`.
- It is never carried.
- A store failure, lost readiness or a new epoch before `verify` refuses it as
admission would (`not_ready`, `stale_epoch`).
- **Commit.** A commit's outcomes come from one request for all of its actions. The
first-record rule still decides each one.
- **API.** `IntentAdmission::action` keeps the immediate single lookup for callers that
want it.
- **Reason:** one round trip per action on the simulation thread dominated the session
layer on the reference worker (ADR-0097, fourth run).
@@ -387,3 +387,34 @@ the trust), then the drill ran at 21:00–21:05 UTC and was restored.
scope here.
- The drill summary now judges by budgets v2 and prints commit and other-tick maxima and
the poll and replicate averages.
## 10.9d fourth reference run (2026-10-09, image `754446fe…`)
Drill at 01:00–01:06 UTC, restored. It is the first run whose actions were all new (no
duplicates), so the first in which actions actually committed (about 12 per commit, one
commit about every 1.05 s).
- **Client side:** 24 sessions within every client budget (response p95 111 ms; 20 Hz;
no starvation; `duplicate_actions=0`). Outcome p95 about 1.2 s: the action waits for
the commit of the batch it rides. The governor refused logins beyond 24.
- **Server side, not within budgets v2:**
- commit ticks 58–70 ms (budget 50);
- session-layer p99 30–39 ms (1.5–1.6 ms per session), and 3–5.6 ms with a single
session;
- other ticks at most 12–19 ms; no tick debt.
- **Cause:** a synchronous PostgreSQL round trip per action on the simulation thread.
Admission looked up each new action's durable record (uniqueness across epochs,
processes and regions), and each commit looked up every committed action again, so the
first-record rule could hold. About 12 lookups land on each commit tick (≈ 35 ms on the
dev database), and each incoming action costs one round trip on whichever tick it
arrives. Locally a round trip is a fraction of a millisecond, which is why the profile
missed it. Drill 3 never reached the commit path.
- **Fix (no schema change):**
- `ActionLedger::find_all` answers many (account, action) pairs in one statement: the
schema v11 `find_region_action`, lease check included, applied to each pair through
`unnest … CROSS JOIN LATERAL`.
- `IntentAdmission::admit` holds an action that only the ledger can still refuse, and
one `verify` per poll confirms every waiting action with one request. Its receipt is
still sent in the same poll.
- `committed` asks once per commit.
- The store cost is now at most one round trip per poll plus one per commit, whatever
the number of sessions or actions.
- **Next:** a fifth run with this fix, then the declaration.
+24
View File
@@ -5254,3 +5254,27 @@ Executed:
Not done:
- shellcheck: not installed in this WSL.
- The fourth drill (needs the new image, a fresh bundle and the go-ahead).
## 2026-10-09 — M10.9d fourth evolution-dev drill, batched durable lookups
Fourth drill (image `754446fe…`, 01:00–01:06 UTC, restored Ready):
- 24 sessions within the client budgets;
- the server was not within budgets v2 (commit ticks 58–70 ms, session p99 30–39 ms).
The cause was one synchronous store round trip per action, at admission and again at
commit (ADR-0097, fourth run). Fix:
- `ActionLedger::find_all` and `find_region_actions` (one statement over the schema v11
function, no migration);
- `IntentAdmission::admit`/`verify`, with one verify per poll in `RegionSessions`;
- one request per commit.
Executed:
- WSL Release `quic`: build, plus the quic, network, ops-contracts and session-fuzz CTest
(6/6), passed.
- The 18 Postgres suites, passed. They include the new `find_region_actions` checks:
order, absent pairs, empty, fenced, and `PostgresActionLedger::find_all`.
- `deferred_actions` in `network_control`.
- clang-tidy 19 on the changed sources, clean after two fixes.
- Windows `debug` 60/60, `quic` 5/5, `client-network` 15/15.
Not done: the fifth drill (needs the new image, a fresh bundle and the go-ahead).