菜单
Automated health checks for dial9 Tokio runtime traces. Detects long polls, task leaks, scheduling delays, blocking calls, queue buildup, worker imbalance, CPU contention, and span anomalies. Use when you want a quick automated assessment of trace health.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Parse and load dial9 Tokio runtime trace files. Covers the ParsedTrace schema, event types, field definitions, parse options, time filtering, symbol resolution, and timestamp conversion. Use when loading traces or understanding the trace data model.
JavaScript analysis toolkit for parsing and analyzing dial9 Tokio runtime traces. Always start trace diagnosis with analyzeTraces() from analyze.js, then use parseTrace() and lower-level helpers only to confirm assumptions or drill into raw events.
Analysis pipeline API for dial9 traces. Covers analyzeTraces() aggregation, buildWorkerSpans, attachCpuSamples, scheduling delays, flamegraphs, span data, and the full return schema. Use when analyzing parsed traces or building custom analysis pipelines.
Diagnostic recipes for common questions about dial9 Tokio runtime traces. Covers finding long polls, task leaks, worker utilization, blocking calls, wake chains, span analysis, task dumps, time-window debugging, and estimating allocation totals from sampled `Alloc` events. Use when answering specific diagnostic questions about trace data.
Zoom into a narrow time window of a dial9 trace to see every worker and OS thread at one moment. Use after an aggregate pass (`analyze.js`, `red_flag_scan.js`) flags a timestamp — a long poll, a queue spike, a latency outlier. Use when the user says "zoom in", "what was happening at +6953ms", or "show me the window around that poll".
Root-cause why a poll was long, not just where it was. Use after `red_flag_scan.js` or `analyze.js` flags a long poll and you need to explain it and recommend a fix — including the off-CPU case with no scheduling events captured. Use when the user says "why is this poll long", "what is it blocked on", "is something holding a lock", or "off-CPU but no sched events".
| name | dial9-red-flags |
| description | Automated health checks for dial9 Tokio runtime traces. Detects long polls, task leaks, scheduling delays, blocking calls, queue buildup, worker imbalance, CPU contention, and span anomalies. Use when you want a quick automated assessment of trace health. |
Run scripts/red_flag_scan.js against any trace to surface common Tokio runtime problems.
node scripts/red_flag_scan.js <trace.bin or directory>
Each finding has a severity: critical, warning, or info.
A single .poll() call took too long. This blocks the worker from processing other tasks. The fixed >10ms warning / >50ms critical cutoffs here are a coarse default, not a universal truth — "long" is really relative to this runtime's own poll distribution. In a service whose p99 poll is 500µs, a 1ms poll is a severe tail outlier these cutoffs miss entirely; in a batch job whose p99 is 40ms, a 20ms poll is normal. Calibrate against pollDurationByLoc (p50/p99 per spawn location) before trusting an absolute threshold. Look at poll.cpuSamples and poll.schedSamples for stack traces. To root-cause why a flagged poll was long — especially an off-CPU one with no scheduling stacks — use the dial9-diagnose-long-poll skill (which thresholds on p99 by default), and dial9-zoom-window to inspect the surrounding instant.
Active task count grows without bound. Tasks are spawned but never complete. Check taskSpawnLocs for spawn locations of unterminated tasks.
Time between Waker::wake() and the task being polled exceeds 5ms. All workers are busy. Fix: shorter polls, more workers, or yield points.
Scheduling samples (source=1) reveal blocking system calls (file I/O, DNS resolution, mutex contention) on the async runtime. These should use spawn_blocking or a dedicated thread.
Global injection queue exceeds 100 (warning) or 1000 (critical). The runtime cannot keep up with incoming work.
Poll counts differ by more than 3x across workers. Work-stealing may not be distributing evenly, or one worker is stuck on long polls.
Workers are active but spending less than 50% of wall time on CPU. The kernel is descheduling them due to CPU contention.
Worker unpark takes more than 1ms of kernel scheduling wait. Indicates CPU contention at the OS level.
A single poll contains more than 20 span enter/exit pairs. Usually a tight loop without yielding.
A span whose duration exceeds 10x the P50 for its name. Flags individual slow operations.
Spans with enter but no exit. Small counts are normal at segment boundaries. Large counts may indicate task cancellation or a bug in span instrumentation.