checkId: check.agent.task.failure.rate plugin: stellaops.doctor.agent severity: warn tags: [agent, task, failure, reliability]
Task Failure Rate
Surfaces a rising share of failing agent tasks so operators can investigate a systemic problem before it stalls promotions and deployments.
Reads IAgentTaskStore.GetExecutionStatsAsync for the current tenant and evaluates the percentage of failed tasks in the configured lookback window.
Default thresholds:
| Result | Failure rate |
|---|---|
| Warn | 5% |
| Fail | 20% |
The default lookback window is 60 minutes. Tune with Doctor:Agent:TaskFailureRate:LookbackMinutes, WarnPercent, and FailPercent.
Evidence collected: the lookback window start, total / succeeded / failed task counts, the computed failure-rate percentage, and the threshold that was applied. If no tasks ran in the window, the check returns Pass.
Why It Matters
A rising task failure rate is an early indicator of systemic problems: infrastructure issues, misconfigured environments, expired credentials, or agent software bugs. Catching a spike early allows operators to intervene before promotion and deployment work queues stall.
Common Causes
- Registry or artifact store unreachable
- Expired credentials used by tasks
- Agent software bug introduced by a recent update
- Target environment misconfigured
- Disk exhaustion on agent hosts
- OOM kills during resource-intensive tasks
Start by finding the common error across recent failures, then fix the underlying dependency or roll back the change that introduced it.
Docker Compose
# List recent failed tasks and look for a shared error signature
stella agent tasks --status failed --last 1h
# Inspect agent logs for the failing window
docker compose -f devops/compose/docker-compose.stella-ops.yml logs agent --tail 500
Bare Metal / systemd
# List recent failed tasks
stella agent tasks --status failed --last 1h
# Inspect agent logs for the failing window
journalctl -u stella-agent --since '1 hour ago'
Kubernetes / Helm
# List recent failed tasks
stella agent tasks --status failed --last 1h
# Inspect agent pod logs for the failing window
kubectl logs -l app.kubernetes.io/component=agent -n stellaops --tail=500
Verification
stella doctor run --check check.agent.task.failure.rate
Related Checks
check.agent.resource.utilization– disk or memory exhaustion causes task failurescheck.agent.task.backlog– retries of failed tasks inflate the queuecheck.agent.heartbeat.freshness– agents dropping mid-task surface as failurescheck.agent.version.consistency– a bad agent version often spikes the failure rate
