Ops-dashboard/lib/parse-worker-log.ts
Janpeter Visser 7e049ebdef feat(worker-logs): add worker run-log viewer page
Nieuwe /worker-logs pagina: een tabel van de laatste N (10/25/50/100)
worker-runs met een inline detailpaneel dat de stream-json output van
Claude Code als leesbare timeline toont (system-init, assistant-tekst,
tool-calls/results, result-kaart).

- lib/parse-worker-log.ts: pure parser — summarizeRunLog (tabel) +
  parseRunLog (timeline), discriminated-union events, server-side
  truncatie van grote tool-results.
- lib/worker-logs.ts: server-only fs-toegang, leest uit WORKER_LOGS_DIR
  (read-only bind mount), naam-regex + pad-confinement, .gz support.
- app/api/worker-logs[/[name]]: GET-routes, auth-guarded, force-dynamic.
- app/worker-logs: server page + client view (tabel, N-selector,
  auto-refresh) + detail (timeline, auto-refresh tijdens in-progress run).

Vereist een read-only bind mount van /srv/scrum4me/worker-logs in de
ops-dashboard-container (docker-compose.yml + WORKER_LOGS_DIR in .env).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-15 01:58:03 +02:00

444 lines
14 KiB
TypeScript

// lib/parse-worker-log.ts
//
// Parser for Scrum4Me worker run-logs (/srv/scrum4me/worker-logs/idea/runs/*.log).
// Each file is produced by `tsx run-one-job.ts > run_log 2>&1` and is a mix of
// plain-text `[run-one-job]` annotation lines and Claude Code `stream-json`
// event lines (the worker spawns `claude --output-format stream-json --verbose`).
//
// Two entry points:
// summarizeRunLog(raw, fileName) — one cheap line scan, for the table.
// parseRunLog(raw, fileName) — full event timeline, for the detail panel.
//
// Pure module, no dependencies — mirrors lib/parse-docker.ts / lib/parse-systemd.ts.
export type RunStatus = 'idle' | 'running' | 'success' | 'error' | 'token-expired' | 'unknown'
export interface RunLogSummary {
fileName: string
runId: string
startedAt: string | null
status: RunStatus
jobId: string | null
model: string | null
permissionMode: string | null
durationMs: number | null
numTurns: number | null
totalCostUsd: number | null
exitCode: number | null
eventCount: number
inProgress: boolean
errorSummary: string | null
}
export type MetaTag =
| 'claim'
| 'auth'
| 'quota'
| 'no-job'
| 'claimed'
| 'worktree'
| 'config'
| 'payload'
| 'spawn'
| 'claude-done'
| 'cleanup'
| 'exit'
| 'error'
| 'token-expired'
| 'timeout'
| 'other'
export type LogEvent =
| { kind: 'meta'; ts: string | null; tag: MetaTag; text: string }
| {
kind: 'system-init'
ts: string | null
model: string
permissionMode: string
tools: string[]
mcpServers: string[]
sessionId: string
cwd: string
version: string
}
| { kind: 'assistant-text'; ts: string | null; text: string; truncated: boolean }
| { kind: 'thinking'; ts: string | null; text: string; truncated: boolean }
| { kind: 'tool-call'; ts: string | null; id: string; name: string; input: string; truncated: boolean }
| {
kind: 'tool-result'
ts: string | null
toolUseId: string
isError: boolean
body: string
truncated: boolean
fullLength: number
}
| { kind: 'rate-limit'; ts: string | null; status: string }
| {
kind: 'result'
ts: string | null
subtype: string
isError: boolean
durationMs: number | null
numTurns: number | null
totalCostUsd: number | null
resultText: string
resultTruncated: boolean
}
| { kind: 'raw'; ts: string | null; text: string }
export interface ParsedRunLog {
summary: RunLogSummary
events: LogEvent[]
inProgress: boolean
responseTruncated: boolean
}
// Per-item caps keep the detail payload bounded even for ~350 KB raw logs.
const TOOL_RESULT_CAP = 8 * 1024
const TEXT_CAP = 16 * 1024
const TOOL_INPUT_CAP = 4 * 1024
const RESPONSE_CAP = 1_500_000
const META_RE = /^(\S+)\s+\[run-one-job\]\s+(.*)$/
function cap(s: string, max: number): { text: string; truncated: boolean } {
if (s.length <= max) return { text: s, truncated: false }
return { text: s.slice(0, max), truncated: true }
}
/** Strip the `.log` / `.log.gz` suffix — the run id is the timestamp filename. */
export function runIdFromFileName(fileName: string): string {
return fileName.replace(/\.log(\.gz)?$/, '')
}
/** run-agent.sh names each file `$(date -u +%Y%m%dT%H%M%SZ).log`, so the name is the start time. */
function startedAtFromRunId(runId: string): string | null {
const m = runId.match(/^(\d{4})(\d{2})(\d{2})T(\d{2})(\d{2})(\d{2})Z$/)
if (!m) return null
return `${m[1]}-${m[2]}-${m[3]}T${m[4]}:${m[5]}:${m[6]}Z`
}
function classifyMeta(msg: string): MetaTag {
if (msg.startsWith('claim attempt')) return 'claim'
if (msg.startsWith('auth ok')) return 'auth'
if (msg.startsWith('quota probe')) return 'quota'
if (msg.startsWith('no job claimed')) return 'no-job'
if (msg.startsWith('claimed job_id=')) return 'claimed'
if (msg.startsWith('worktree path=')) return 'worktree'
if (msg.startsWith('config ')) return 'config'
if (msg.startsWith('payload written')) return 'payload'
if (msg.startsWith('spawn claude')) return 'spawn'
if (msg.startsWith('claude done')) return 'claude-done'
if (msg.startsWith('cleanup')) return 'cleanup'
if (msg.startsWith('exit code=')) return 'exit'
if (msg.startsWith('ERROR')) return 'error'
if (msg.startsWith('TOKEN_EXPIRED detected')) return 'token-expired'
if (msg.startsWith('claim timeout')) return 'timeout'
return 'other'
}
/** Cheap single-pass summary for the table — at most one JSON.parse (the result line). */
export function summarizeRunLog(raw: string, fileName: string): RunLogSummary {
const runId = runIdFromFileName(fileName)
const lines = raw.split('\n')
let jobId: string | null = null
let model: string | null = null
let permissionMode: string | null = null
let claudeExit: number | null = null
let runExit: number | null = null
let durationMs: number | null = null
let numTurns: number | null = null
let totalCostUsd: number | null = null
let eventCount = 0
let hasResult = false
let resultIsError = false
let resultSubtype: string | null = null
let tokenExpired = false
let hasErrorLine = false
let firstErrorMsg: string | null = null
for (const line of lines) {
if (!line) continue
const m = line.match(META_RE)
if (m) {
const msg = m[2]
if (msg.startsWith('claimed job_id=')) {
jobId = msg.slice('claimed job_id='.length).trim() || jobId
} else if (msg.startsWith('config ')) {
model = /\bmodel=(\S+)/.exec(msg)?.[1] ?? model
permissionMode = /\bpermission_mode=(\S+)/.exec(msg)?.[1] ?? permissionMode
} else if (msg.startsWith('claude done')) {
const e = /\bexit_code=(-?\d+)/.exec(msg)
if (e) claudeExit = Number(e[1])
const d = /\bduration_ms=(\d+)/.exec(msg)
if (d) durationMs = Number(d[1])
} else if (msg.startsWith('exit code=')) {
const e = /exit code=(-?\d+)/.exec(msg)
if (e) runExit = Number(e[1])
} else if (msg.startsWith('TOKEN_EXPIRED detected')) {
tokenExpired = true
} else if (msg.startsWith('ERROR')) {
hasErrorLine = true
if (!firstErrorMsg) firstErrorMsg = msg.replace(/^ERROR\s*/, '').slice(0, 300)
}
continue
}
const trimmed = line.trimStart()
if (trimmed.startsWith('{')) {
eventCount++
if (!hasResult && trimmed.startsWith('{"type":"result"')) {
try {
const obj = JSON.parse(trimmed)
hasResult = true
resultIsError = !!obj.is_error
resultSubtype = typeof obj.subtype === 'string' ? obj.subtype : null
if (typeof obj.num_turns === 'number') numTurns = obj.num_turns
if (typeof obj.total_cost_usd === 'number') totalCostUsd = obj.total_cost_usd
if (durationMs == null && typeof obj.duration_ms === 'number') durationMs = obj.duration_ms
} catch {
// malformed result line — ignore
}
}
}
}
const exitCode = claudeExit ?? runExit
const terminal = runExit != null || hasResult || hasErrorLine || tokenExpired
const inProgress = !terminal
let status: RunStatus
if (tokenExpired) {
status = 'token-expired'
} else if (jobId) {
if (inProgress) {
status = 'running'
} else if (
resultIsError ||
hasErrorLine ||
(claudeExit != null && claudeExit !== 0) ||
(runExit != null && runExit !== 0)
) {
status = 'error'
} else {
status = 'success'
}
} else {
// No job was claimed this iteration — the worker was idle / waiting.
status = 'idle'
}
let errorSummary: string | null = null
if (status === 'error' || status === 'token-expired') {
errorSummary =
firstErrorMsg ??
(tokenExpired ? 'TOKEN_EXPIRED detected in output' : null) ??
(resultIsError ? `result: ${resultSubtype ?? 'error'}` : null) ??
(exitCode != null ? `exit code ${exitCode}` : null)
}
return {
fileName,
runId,
startedAt: startedAtFromRunId(runId),
status,
jobId,
model,
permissionMode,
durationMs,
numTurns,
totalCostUsd,
exitCode,
eventCount,
inProgress,
errorSummary,
}
}
function normalizeContent(content: unknown): string {
if (typeof content === 'string') return content
if (Array.isArray(content)) {
return content
.map((b) => {
if (typeof b === 'string') return b
if (b && typeof b === 'object' && typeof (b as { text?: unknown }).text === 'string') {
return (b as { text: string }).text
}
return JSON.stringify(b)
})
.join('\n')
}
if (content == null) return ''
return JSON.stringify(content)
}
/* eslint-disable @typescript-eslint/no-explicit-any -- stream-json events are genuinely dynamic */
function pushJsonEvent(events: LogEvent[], obj: any): void {
const type = obj?.type
const ts: string | null = typeof obj?.timestamp === 'string' ? obj.timestamp : null
if (type === 'system') {
const mcp = Array.isArray(obj.mcp_servers)
? obj.mcp_servers.map((s: any) => (typeof s?.name === 'string' ? s.name : String(s)))
: []
events.push({
kind: 'system-init',
ts,
model: typeof obj.model === 'string' ? obj.model : '—',
permissionMode: typeof obj.permissionMode === 'string' ? obj.permissionMode : '—',
tools: Array.isArray(obj.tools) ? obj.tools.filter((t: unknown) => typeof t === 'string') : [],
mcpServers: mcp,
sessionId: typeof obj.session_id === 'string' ? obj.session_id : '',
cwd: typeof obj.cwd === 'string' ? obj.cwd : '',
version: typeof obj.claude_code_version === 'string' ? obj.claude_code_version : '',
})
return
}
if (type === 'rate_limit_event') {
events.push({
kind: 'rate-limit',
ts,
status: typeof obj.rate_limit_info?.status === 'string' ? obj.rate_limit_info.status : 'unknown',
})
return
}
if (type === 'assistant') {
const content = obj?.message?.content
if (Array.isArray(content)) {
for (const block of content) {
if (block?.type === 'text' && typeof block.text === 'string') {
const c = cap(block.text, TEXT_CAP)
events.push({ kind: 'assistant-text', ts, text: c.text, truncated: c.truncated })
} else if (block?.type === 'thinking' && typeof block.thinking === 'string') {
const c = cap(block.thinking, TEXT_CAP)
events.push({ kind: 'thinking', ts, text: c.text, truncated: c.truncated })
} else if (block?.type === 'tool_use') {
let inputStr: string
try {
inputStr = JSON.stringify(block.input, null, 2)
} catch {
inputStr = String(block.input)
}
const c = cap(inputStr, TOOL_INPUT_CAP)
events.push({
kind: 'tool-call',
ts,
id: typeof block.id === 'string' ? block.id : '',
name: typeof block.name === 'string' ? block.name : 'tool',
input: c.text,
truncated: c.truncated,
})
}
}
}
return
}
if (type === 'user') {
const content = obj?.message?.content
if (Array.isArray(content)) {
for (const block of content) {
if (block?.type === 'tool_result') {
const body = normalizeContent(block.content)
const c = cap(body, TOOL_RESULT_CAP)
events.push({
kind: 'tool-result',
ts,
toolUseId: typeof block.tool_use_id === 'string' ? block.tool_use_id : '',
isError: !!block.is_error,
body: c.text,
truncated: c.truncated,
fullLength: body.length,
})
}
}
}
return
}
if (type === 'result') {
const c = cap(typeof obj.result === 'string' ? obj.result : '', TEXT_CAP)
events.push({
kind: 'result',
ts,
subtype: typeof obj.subtype === 'string' ? obj.subtype : 'unknown',
isError: !!obj.is_error,
durationMs: typeof obj.duration_ms === 'number' ? obj.duration_ms : null,
numTurns: typeof obj.num_turns === 'number' ? obj.num_turns : null,
totalCostUsd: typeof obj.total_cost_usd === 'number' ? obj.total_cost_usd : null,
resultText: c.text,
resultTruncated: c.truncated,
})
return
}
// Unknown event type — keep a compact raw note so nothing is silently dropped.
events.push({ kind: 'raw', ts, text: cap(`${type ?? 'event'}: ${JSON.stringify(obj)}`, 2048).text })
}
/* eslint-enable @typescript-eslint/no-explicit-any */
function estimateSize(e: LogEvent): number {
switch (e.kind) {
case 'assistant-text':
case 'thinking':
case 'raw':
return e.text.length
case 'tool-call':
return e.input.length
case 'tool-result':
return e.body.length
case 'result':
return e.resultText.length
default:
return 64
}
}
/** Bound the whole payload — drop tool-result bodies oldest-first if still too large. */
function enforceResponseCap(events: LogEvent[]): boolean {
let total = 0
for (const e of events) total += estimateSize(e)
if (total <= RESPONSE_CAP) return false
for (const e of events) {
if (total <= RESPONSE_CAP) break
if (e.kind === 'tool-result' && e.body) {
total -= e.body.length
e.body = ''
e.truncated = true
}
}
return true
}
/** Full event timeline for the detail panel. */
export function parseRunLog(raw: string, fileName: string): ParsedRunLog {
const summary = summarizeRunLog(raw, fileName)
const events: LogEvent[] = []
for (const line of raw.split('\n')) {
if (!line.trim()) continue
const m = line.match(META_RE)
if (m) {
events.push({ kind: 'meta', ts: m[1], tag: classifyMeta(m[2]), text: m[2] })
continue
}
const trimmed = line.trimStart()
if (trimmed.startsWith('{')) {
try {
pushJsonEvent(events, JSON.parse(trimmed))
} catch {
// partial / malformed JSON line (e.g. a log read mid-write) — keep it raw
events.push({ kind: 'raw', ts: null, text: cap(line, TOOL_RESULT_CAP).text })
}
continue
}
// Non-JSON, non-meta noise (e.g. a bare `Warning: ...` from claude).
events.push({ kind: 'raw', ts: null, text: cap(line, TOOL_RESULT_CAP).text })
}
const responseTruncated = enforceResponseCap(events)
return { summary, events, inProgress: summary.inProgress, responseTruncated }
}