Build Logging

Each deploy build emits granular per-step logs to the lk-build-logs Axiom dataset. These run inside the Docker container (Node.js) and provide detailed visibility into the transpile → bundle → deploy pipeline.

This is separate from the wide event system which captures one deploy:complete event per build. Build logs capture many events — one per pipeline step, with timing and detail payloads.

Architecture

CONTAINER (Node.js)                           AXIOM
───────────────────                           ─────
createAxiomBuffer() ──buffer──▶ flush() ──▶ lk-build-logs
  │
  ├─ push({ phase: 'transpile', status: 'start' })
  ├─ push({ phase: 'transpile', message: 'compiled 42 sections' })
  ├─ push({ phase: 'transpile', status: 'done', durationMs: 12340 })
  ├─ push({ phase: 'bundle', status: 'start' })
  └─ ...

Event shape

Every build log event follows the AxiomEvent interface:

interface AxiomEvent {
  buildId: string           // UUID — primary key for queries
  slug?: string             // client slug
  _time: string             // ISO timestamp
  level: 'debug' | 'info' | 'warn' | 'error'
  phase?: string            // pipeline step: transpile, bundle, deploy, complete, failed
  message: string           // human-readable description
  step?: string             // sub-step within phase
  status?: string           // start, done, error
  durationMs?: number       // step duration
  detail?: Record<string, unknown>  // arbitrary payload
}

Buffered ingest

The createAxiomBuffer() factory from scripts/lib/axiom.ts batches events before sending to reduce HTTP overhead:

Parameter	Value
Max buffer size	20 events
Max buffer age	3 seconds
Urgent flush	On level: 'error', status: 'start', or status: 'done'

Urgent events (errors, phase start/done) flush the buffer immediately so the dashboard and tail CLI see progress without delay. All other events batch up to 20 or 3 seconds.

The final flush() is awaited in the pipeline’s finally block to ensure all events are sent before the container exits.

Token configuration

Env Var	Dataset	Scope
AXIOM_BUILD_API_TOKEN	lk-build-logs	API token for Axiom ingest/query
AXIOM_BUILD_DATASET_NAME	lk-build-logs	Dataset name (defaults to lk-build-logs)

The Worker passes the token to the container via buildContainerPayload(). The container creates its own AxiomBuffer instance with this token.

Set locally in .mise.local.toml:

[env]
AXIOM_BUILD_API_TOKEN = "xaat-..."
AXIOM_BUILD_DATASET_NAME = "lk-build-logs"

Querying build logs

Worker endpoint

GET /builds/:buildId/logs?limit=1000&startTime=...&endTime=...
Authorization: Bearer {CLOUDFLARE_API_TOKEN}

Returns { events: AxiomEvent[] }. Used by the dashboard build detail view.

CLI — pnpm deploy:logs

pnpm deploy:logs <buildId>            # fetch + pretty-print all logs
pnpm deploy:logs <buildId> --follow   # tail mode (live build)
pnpm deploy:logs <buildId> --json     # raw JSON output

The --follow mode uses tailAxiomLogs() which polls every 2 seconds and stops when it sees a terminal phase (complete or failed). Timeout: 15 minutes.

APL queries in Axiom UI

Find all failed builds in the last 24 hours:

['lk-build-logs']
| where level == "error"
| summarize count() by buildId
| sort by count_ desc

Find slow transpile phases:

['lk-build-logs']
| where phase == "transpile" and status == "done"
| where durationMs > 30000
| project buildId, slug, durationMs, _time
| sort by durationMs desc

Key files

File	Purpose
scripts/lib/axiom.ts	createAxiomBuffer(), queryAxiomLogs(), tailAxiomLogs()
scripts/deploy-logs.ts	CLI log viewer (pnpm deploy:logs)
packages/cf-client-deploy/src/worker.ts	/builds/:id/logs endpoint