Add M10.4c evo-server QUIC listener and headless session client
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 1m46s
Foundation / checks (postgres) (push) Canceled after 0s
Foundation / checks (tsan) (push) Canceled after 0s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (checks) (push) Canceled after 3m44s
Foundation / checks (client) (push) Canceled after 3m49s
Foundation / checks (network) (push) Canceled after 1m59s

evo-server gains an opt-in M10 listener (--m10-*, QUIC builds only): RegionSessions
with the PostgreSQL control and action ledgers on their own connection under the
host's lease, carried actions submitted to the host and its acknowledgements and
drops turned into outcomes. evo-session-probe is a dev-only headless client. Real
process tests cover intentions, durable action outcomes, SIGKILL before and after
an action's commit, duplicate retries and the probe. The simulated watermark now
reaches every session every tick, so clients always know the server tick.
ADR-0092.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
dingo
2026-10-04 09:20:55 -04:00
co-authored by Claude Opus 5.5
parent 69f3ad7cf8
commit 0bf9901c7f
27 changed files with 1053 additions and 42 deletions
@@ -23,8 +23,9 @@ authoritative state yet. Committed batch traces record host inputs only, not req
state claim (position, energy): the message has no such fields and any extra bytes make
it invalid. Answers: an `IntentionReceipt` (admitted = scheduled for its tick, or a
refusal) and a cumulative `SimulatedThrough` watermark (every admitted intention up to a
sequence whose tick has been simulated in the epoch), repeated every 20 ticks so a lost
datagram is covered. Intentions are superseding: there is no per-intention durable record.
sequence whose tick has been simulated in the epoch), sent to every session on every
simulated tick (sequence 0 before its first), so a lost datagram is covered and clients
always know the server tick (until 10.5 snapshots carry it). Intentions are superseding: there is no per-intention durable record.
- *Actions* (stream): a client-chosen action id, unique per account and stable across
retries and reconnects, a kind and a tick. Answers: an `ActionReceipt` (scheduled, or a
refusal) and a final `ActionOutcome`: `committed_no_effect` with its commit sequence once
@@ -89,3 +90,30 @@ Schema v11 (`migrations/0011_region_actions.sql`, [region-actions-v1](../../sche
- Limits: action rows are retained (no pruning policy yet); clones (ADR-0067) do not copy
them; the two-argument publication of a commit that has an action block would still
return its receipt (the adapter never does this).
## M10.4c evo-server listener and headless client (2026-10-04)
- **evo-server** (`--m10-listen ADDRESS:PORT --m10-cert F --m10-key F --m10-issuer F
--m10-environment dev|stage|world0`, builds with EVO_BUILD_QUIC and PostgreSQL only;
opt-in, development and test, nothing deploys it): `RegionSessions` with
`PostgresControlLedger` and `PostgresActionLedger` on their own connection slot (reconnects
after store failures) under the host's current lease (`RegionHost::authority()`). After
each host step the sessions see the host's authority; actions scheduled up to 8 ticks
ahead go to the host (`submit_action`; a refusal is final `not_committed`); the host's
action acknowledgements and drops become outcomes. Draining closes sessions with
`shutdown`. The issuer public key is a 65-byte file (key id 1).
- **evo-session-probe** (dev only): generates a dev issuer key pair, signs a grant for a
given account and serial, logs in, sends intentions and probe actions, queries ids and
prints one machine-readable line per answer.
- **After a restart, `unknown` means "not on record"**: a new process cannot tell an id it
never saw from one admitted by a dead process. Both were never committed (the old epoch
is fenced and its commits are final), so retrying the same id is safe and commits once.
- **Found with real processes: clients need the server tick every tick.** A client that sent
only actions learned the server tick from receipts and from watermarks repeated every 20
ticks, so after waiting about a second for an outcome its next action named a past tick
(`outside_window`). `SimulatedThrough` is now sent to every established session on every
simulated tick (sequence 0 before its first intention): 24 bytes per tick on the datagram
channel, and the client's clock for the server tick until 10.5 snapshots carry it.
- Process evidence: actions survive SIGKILL after their commit (the outcome is answered from
the record by the next process, and a retry is a duplicate with the same outcome); an
action whose tick had not run at SIGKILL has no record, answers `unknown`, and commits
exactly once when retried.