Add M10.9a load measurement, governor and overload shedding
Foundation / checks (deploy) (push) Successful in 3m20s
Foundation / checks (ecs) (push) Successful in 1m59s
Foundation / checks (client) (push) Failing after 10m0s
Foundation / checks (network) (push) Successful in 13m9s
Foundation / checks (tsan) (push) Successful in 8m1s
Foundation / checks (checks) (push) Successful in 32m2s
Foundation / checks (postgres) (push) Failing after 26m39s
Foundation / postgres-tools (push) Skipped
Foundation / server-image (push) Skipped
Foundation / image (push) Skipped
Foundation / checks (deploy) (push) Successful in 3m20s
Foundation / checks (ecs) (push) Successful in 1m59s
Foundation / checks (client) (push) Failing after 10m0s
Foundation / checks (network) (push) Successful in 13m9s
Foundation / checks (tsan) (push) Successful in 8m1s
Foundation / checks (checks) (push) Successful in 32m2s
Foundation / checks (postgres) (push) Failing after 26m39s
Foundation / postgres-tools (push) Skipped
Foundation / server-image (push) Skipped
Foundation / image (push) Skipped
ADR-0097 (M10.9 load and abuse qualification; only evolution-dev reference-worker runs qualify). evo-server measures each simulated tick's work (host step plus the M10 bridge, with the session layer's share kept separately) and feeds a portable LoadGovernor: grant logins are refused at a 2 s p95 of 15 ms or any tick debt, replication is shed in up to three levels at 20 ms or debt >= 2 (radius cap 7/7/5/3, far updates x1/x2/x2/x4, no fresh baselines, own markers unaffected), and both relax after 5 s of calm. Handoff tickets are still admitted. /metrics gains evo_m10_* gauges when the listener is enabled. Adds the missing ADR index entries for ADR-0094 to ADR-0096. Refreshes the dev session window. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -4904,3 +4904,47 @@ Not done:
|
||||
- deployment of protocol v4 and schemas v12-v14 (needs a rebuilt migrations image; not done
|
||||
from routine development);
|
||||
- a GUI run across two regions on evolution-dev (only one dev region exists).
|
||||
|
||||
## 2026-10-07 — M10.9a load measurement and governor (ADR-0097)
|
||||
|
||||
Implemented:
|
||||
- evo-server measures each simulated tick's work: the host step plus the M10 bridge
|
||||
(session poll, replication, action carry). Waits are excluded, and the session layer's
|
||||
share is kept separately.
|
||||
- `LoadGovernor` (`include/evo/server/load_governor.hpp`, portable): a window of 40 ticks,
|
||||
refuse at p95 ≥ 15 ms or any tick debt, shed one level per second at p95 ≥ 20 ms or
|
||||
debt ≥ 2, and relax one step per 5 s of calm (p95 < 10 ms, no debt).
|
||||
- `RegionSessions::set_load`: while refusing, grant logins are closed with `capacity`
|
||||
before signature work; handoff tickets are still admitted.
|
||||
- `ReplicationServer::set_shed`: levels 0-3 cap the radius (7/7/5/3) and the near radius
|
||||
(3/3/3/2) and stretch far updates (x1/x2/x2/x4). Stored interests are kept, and own
|
||||
markers are unaffected.
|
||||
- `/metrics`: `evo_m10_*` gauges (sessions, refusing, shed level, work and session-work
|
||||
percentiles, maximum) and a refused-logins counter, rendered only with the listener.
|
||||
evo-server prints `m10 load ...` on every change.
|
||||
- The ADR index gains the missing entries for ADR-0094 to ADR-0096.
|
||||
|
||||
Tests:
|
||||
- `ops-contracts`: the governor state machine (thresholds, step interval, hysteresis, calm
|
||||
restart, debt, policy validation) and the M10 metric families, including worst-case size.
|
||||
- `network`: replication shedding (caps, no fresh baseline, 16-tick far cadence at level
|
||||
3, recovery at level 0).
|
||||
- `quic-sessions`: grant logins refused under load (the grant is not consumed), a ticket
|
||||
login admitted, and logins accepted after relaxing.
|
||||
- Postgres `region_m10_session_tests`: a real evo-server's `/metrics` shows one session,
|
||||
measured work, not refusing, shed 0.
|
||||
|
||||
Executed:
|
||||
- **Linux GCC 14:** `postgres-check.sh` with QUIC, all suites; presets `quic` (ops,
|
||||
network, QUIC sessions) 3/3. The local reference (Release, one session) measured work
|
||||
p95 7.1 ms and p99 8.8 ms, with the session layer at p99 4.5 ms. This is a development
|
||||
reference only, flagged for 10.9b.
|
||||
- **Linux Clang 19:** clang-tidy 19 on the governor, ops, replication, sessions, evo-server
|
||||
`main.cpp` with the bridge, and the tests (one finding fixed); clang-format 19.
|
||||
- **Windows MSVC 19.44:** presets `debug` 60/60 (ops-contracts included), `quic` 4/4 and
|
||||
`client-network` 14/14.
|
||||
|
||||
Not done:
|
||||
- qualification on the evolution-dev reference worker (10.9d; needs v4 and v12-v14
|
||||
deployed);
|
||||
- load generation (10.9b) and the abuse suite (10.9c).
|
||||
|
||||
Reference in New Issue
Block a user