Tick cellular births only for accounts holding a session
Foundation / checks (ecs) (push) Successful in 2m19s
Foundation / checks (deploy) (push) Successful in 3m56s
Foundation / checks (tsan) (push) Successful in 9m34s
Foundation / checks (client) (push) Successful in 16m45s
Foundation / checks (network) (push) Successful in 18m57s
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 28m43s
Foundation / checks (postgres) (push) Canceled after 28m44s

A second account's queued birth outlived its session. Every following
batch failed check_life_control with EVL02, which the host reads as a
lost region lease, so region 100/3 fenced, reacquired and failed again
every 20 s, refusing logins. Include, and tick for, only births whose
account holds a session; the rest stay queued until the account returns.
Record the incident in ADR-0123. Also refresh the dev session window (3 h).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
dingo
2026-10-13 16:22:31 -04:00
co-authored by Claude Opus 5.5
parent b735860bff
commit f3aecaf36a
4 changed files with 34 additions and 6 deletions
+14 -4
View File
@@ -1021,7 +1021,19 @@ int serve_cellular(Options& o) {
}
}
}
const bool birth_waiting = !host.queued_births().empty();
// A queued birth publishes only while its account holds a session here: the
// publisher checks current account control, and a departed account's birth would
// fail every batch with the lease-fence code, so the region would fence itself in a
// loop. The birth stays queued and resumes when the account returns.
std::vector<server::CellularTickIntents::Birth> held_births;
for (const auto& birth : host.queued_births()) {
network::AccountId account;
account.bytes = birth.account;
if (held_births.size() < biology::max_cellular_tick_lives &&
bridge->sessions().holder(account))
held_births.push_back(birth);
}
const bool birth_waiting = !held_births.empty();
if (status.ready && state && (player_present || birth_waiting)) {
const auto now = std::chrono::steady_clock::now();
if (!next_tick)
@@ -1039,9 +1051,7 @@ int serve_cellular(Options& o) {
if (!builder) {
bridge->dropped(actions, true);
} else {
const auto births = host.queued_births().first(
std::min(host.queued_births().size(), biology::max_cellular_tick_lives));
const auto appended = builder->append(tick, baseline, actions, births);
const auto appended = builder->append(tick, baseline, actions, held_births);
if (!appended) {
bridge->dropped(actions, true);
} else {
@@ -12,7 +12,7 @@ spec:
env:
- {name: REGION_SESSION, value: "dev"}
- {name: REGION_SESSION_LISTEN, value: "0.0.0.0:7400"}
- {name: REGION_SESSION_WINDOW, value: "dev-20261013T191802-66969:1791929882"}
- {name: REGION_SESSION_WINDOW, value: "dev-20261013T202231-51334:1791933751"}
- {name: REGION_M10_MAX_GRANT_LIFETIME, value: "14400"}
ports:
# QUIC (ADR-0090): UDP only; nothing listens on TCP 7400.
@@ -86,6 +86,16 @@ region in DEV. Bringing one up exposed four gaps:
client sends one every tick. The default probe admission (4 actions per
8-tick window) refused every other one. The cellular bridge allows two per
tick of window, which leaves headroom for arrival jitter.
- **A departed account's birth waits instead of fencing the region.** A
second account requested a birth and disconnected. Its queued birth kept
the host ticking, and the cellular publisher's `check_life_control` refused
the batch with SQLSTATE EVL02. That is the same code as a lost region lease,
so the host fenced itself, reacquired and failed again every 20 seconds,
refusing logins (region 3, 2026-10-13 20:18 UTC).
- The host now includes and ticks only births whose account holds a session.
- Other births stay queued until the account returns.
- Restarting the pod cleared the in-memory queue. The reservation is
durable, and the client's query re-queues it.
- **Grants for region 3 must be minted with `--region 3 --lifetime 300` or
less.** Longer probe sessions require a separate, reviewed change to the
cellular grant ceiling; this ADR does not make it.
+9 -1
View File
@@ -7186,4 +7186,12 @@ opening starved the cell. Only patch energy feeds the reserve, and founder
intake (1,148,000 pJ/s) emptied its 4M and 16M pJ patches in seconds. DEV moves
to region 100/3 with 200M and 50M pJ patches (ADR-0123). Checked locally:
g++-14 `-Werror` and clang-tidy-19 on `apps/evo-server/main.cpp` with
`EVO_SERVER_M10`, plus the deploy policy suite.
`EVO_SERVER_M10`, plus the deploy policy suite.
Region 3 then livelocked. A second account's birth was queued, the account
disconnected, and every following batch failed `check_life_control` with
EVL02, which the host reads as a lost region lease. It fenced, reacquired and
failed again every 20 s, refusing logins with `shutdown`. A restart of
`Deployment/evo-region` (permitted by the agent contract) cleared the
in-memory queue, and the region held `ready 1` at epoch 8 from 20:20:44 UTC.
`serve_cellular` now includes, and ticks for, only births whose account holds
a session. Checked locally: g++-14 `-Werror`, clang-tidy-19 and clang-format.