Skip to content

delivery_router: redundant reliable-wake notify (channel full) logs at WARN, flooding healthy-engine logs #291

Description

@loning

Evidence

On a healthy trureturing host engine (substrate a8029a9, platform 8e82f4bc, uptime ~40min, deliveries flowing, DLQ drained), the supervise log accumulated 233×:

LEVEL=WARN dept=github-devloop-intake.admission error=no available capacity MSG=reliable wake notify failed

Analysis

crates/fkst-framework/src/supervise/delivery_router.rs:271notify_reliable does wake.try_send(()) on the bounded wake channel and logs every failure at WARN. When the channel is full the consumer already has a pending wake: the notify is redundant coalescing, the wake will happen anyway, and nothing is lost. Under event bursts (intake admission fan-in) this floods the log and poisons downstream monitor signals (our health snapshot counts warn/error lines; 620 of the instance's lines were dominated by this one benign path — a bad-raw-material generator for every log-reading harness).

Suggested fix

Distinguish TrySendError::Full (wake already pending → debug/trace, or drop silently) from TrySendError::Closed / registry errors (real faults → keep WARN). One-line match on the error variant.

Environment

  • host: trureturing legacy-layout engine, macOS
  • substrate: a8029a9 (dev), platform: 8e82f4bc8bcc

Metadata

Metadata

Assignees

Labels

fkst-class:standardservice-class-standard-display-onlyfkst-dev:enabledintake-approved-for-autonomous-developmentfkst-dev:thinkingconsensus-deliberation-in-progress

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions