structured-logging-axiom

Structured Logging & Axiom Infrastructure

Zero-dependency structured logging for Next.js with Axiom integration via Vercel Log Drains. No external SDK required.

Architecture

Code -> createLogger('namespace') -> console.info(JSON) -> Vercel stdout -> Log Drain -> Axiom

Logger Implementation

Create a centralized logger module that replaces all raw console.* calls:

// lib/logger.ts
export type LogLevel = 'debug' | 'info' | 'warn' | 'error'

export interface Logger {
  debug(message: string, context?: Record<string, unknown>): void
  info(message: string, context?: Record<string, unknown>): void
  warn(message: string, context?: Record<string, unknown>): void
  error(message: string, error?: Error, context?: Record<string, unknown>): void
  child(context: Record<string, unknown>): Logger
}

export function createLogger(namespace: string, parentContext?: Record<string, unknown>): Logger
export function generateCorrelationId(): string

Features

• Namespaced loggers: createLogger('analyze-stream') identifies the source module
• Child loggers: log.child({ correlationId }) bakes in context for all subsequent calls
• Log level filtering: LOG_LEVEL env var override without redeploying
• Dual output: JSON in production (Axiom-compatible), human-readable in development
• Reserved key protection: Context keys colliding with log fields get ctx_ prefix
• Correlation IDs: generateCorrelationId() for request tracing across log entries

Usage

import { createLogger, generateCorrelationId } from '@/lib/logger'

const log = createLogger('my-module')

// Basic usage
log.debug('Verbose detail', { rowIndex: 42 })
log.info('Operation started', { itemCount: 12450 })
log.warn('Threshold exceeded', { sizeMB: 48.2 })
log.error('Parse failed', error, { filename: 'data.csv' })

// Child logger — bakes in context for all subsequent calls
const correlationId = generateCorrelationId()
const reqLog = log.child({ correlationId, requestId: 'abc-123' })
reqLog.info('Processing started')    // auto-includes correlationId + requestId
reqLog.debug('Item parsed', { idx }) // merges with parent context

// Chained children accumulate context
const itemLog = reqLog.child({ itemId: 'item-001' })
itemLog.info('Item processed')  // has correlationId + requestId + itemId

Log Levels

Level	When to Use	Production Default
debug	Verbose internals (row parsing, intermediate state)	Suppressed
info	Significant milestones (analysis started, export complete)	Logged
warn	Recoverable issues (no result found, threshold exceeded)	Logged
error	Failures requiring attention (parse crash, API error)	Logged

Runtime Override

Set LOG_LEVEL environment variable in Vercel to temporarily change log verbosity without redeploying:

LOG_LEVEL=debug  -> all levels logged
LOG_LEVEL=error  -> only errors logged

Validate the env var at startup — invalid values fall back to the environment default.

Dual Output Format

Development (human-readable):

[2026-02-14T01:30:00.000Z] INFO  [my-module] Processing started {"itemCount":12450}

Production (JSON for Axiom):

{"timestamp":"2026-02-14T01:30:00.000Z","level":"info","namespace":"my-module","message":"Processing started","itemCount":12450,"correlationId":"a1b2c3d4-k5m9n2"}

Format is selected at module load time based on NODE_ENV. Context fields are placed at the top level of JSON (not nested) for direct Axiom column indexing.

Reserved Key Protection

If context contains keys that collide with log entry fields (timestamp, level, namespace, message, error, stack), they are auto-prefixed with ctx_:

log.info('test', { message: 'user input' })
// JSON output: {"timestamp":"...","level":"info","message":"test","ctx_message":"user input"}

Axiom Integration — Two Approaches

Option A: Vercel Log Drains (Recommended)

Zero dependencies. Structured JSON goes to stdout, Vercel captures it, Log Drain webhook forwards to Axiom.

logger.ts -> console.info(JSON) -> Vercel stdout -> Log Drain webhook -> Axiom dataset

Pros: No npm packages, no code changes, no API keys in env vars, works with any logger that outputs to stdout Cons: Only captures server-side logs (SSR/API routes), slight delay (~seconds), no client-side browser logs

Setup:

1. Create Axiom account at axiom.co (free tier: 500MB/day, 30-day retention)
2. Create a dataset (e.g., my-app-logs)
3. Vercel Dashboard -> Settings -> Integrations -> Axiom -> Install
4. Connect your project and select Log Drains
5. Deploy — logs start flowing automatically

Option B: Axiom API Direct (For Client-Side or Custom Pipelines)

Send logs directly to Axiom's ingest API. Useful for browser-side logs or when you need guaranteed delivery.

logger.ts -> axiom.ingest([events]) -> HTTPS POST -> Axiom dataset

Pros: Works client-side (browser), guaranteed delivery, batch control, can send from Edge/Workers Cons: Requires @axiomhq/js package (~8KB), needs AXIOM_TOKEN + AXIOM_DATASET env vars

Setup:

npm install @axiomhq/js

// lib/axiom-transport.ts
import { Axiom } from '@axiomhq/js'

const axiom = new Axiom({ token: process.env.AXIOM_TOKEN! })

export async function flushToAxiom(events: Record<string, unknown>[]) {
  axiom.ingest(process.env.AXIOM_DATASET!, events)
  await axiom.flush()
}

Add a transport layer to the logger that buffers JSON entries and flushes in batches (e.g., every 5s or 50 events).

When to consider Option B:

• You need client-side (browser) error logs in Axiom (Log Drains only capture server stdout)
• You need guaranteed delivery (Log Drains are best-effort)
• You're deploying outside Vercel (e.g., Cloudflare Workers, self-hosted)
• You want custom batching, sampling, or routing logic

Comparison

Aspect	Log Drains (A)	Direct API (B)
Dependencies	None	@axiomhq/js
Server-side logs	Yes	Yes
Client-side logs	No	Yes
Env vars needed	None (Vercel handles auth)	AXIOM_TOKEN, AXIOM_DATASET
Delivery guarantee	Best-effort	Guaranteed (with flush)
Latency	~seconds	~immediate on flush
Code changes	Zero	Transport layer needed
Works outside Vercel	No	Yes

Useful Axiom Queries

// All errors in the last hour
| where level == "error" | order by _time desc

// Trace a single request
| where correlationId == "a1b2c3d4-k5m9n2" | order by _time asc

// Slow operations by namespace
| where namespace == "reverse-geocode" | summarize avg(durationMs) by bin(_time, 5m)

// Error rate by namespace
| where level == "error" | summarize count() by namespace | order by count_ desc

ESLint Guard

Enforce no-console: 'error' in ESLint flat config to prevent regression after migration:

// eslint.config.mjs
import nextConfig from "eslint-config-next";

const eslintConfig = [
  ...nextConfig,
  { rules: { "no-console": "error" } },
  {
    files: ["**/*.test.ts", "**/*.test.tsx", "**/*.spec.ts", "**/*.spec.tsx", "e2e/**"],
    rules: { "no-console": "off" },
  },
];

export default eslintConfig;

Only the logger module itself has eslint-disable-next-line exceptions.

Migration Patterns

When converting existing console.* calls to the structured logger:

From	To
console.log('[API] File type:', fileType)	log.debug('File type detected', { fileType })
console.error('Failed:', error)	log.error('Failed', error instanceof Error ? error : undefined)
console.warn('No results for', address)	log.warn('No results found', { address })
console.log(\Parsed ${count} rows`)`	log.info('Parsing complete', { count })
console.error(error.message); console.error(error.stack)	log.error('Operation failed', error) (logger handles stack automatically)

Rules

• Strip prefix tags ([API], [FilePreview], [SSE]) — the namespace handles identification
• console.log -> log.debug (verbose) or log.info (milestones)
• console.error -> log.error(message, errorObj?, context?) — pass Error object as 2nd arg
• console.warn -> log.warn(message, context?)
• Replace string interpolation with context objects: template literals -> { key: value }
• Combined console.error calls (message + stack) become a single log.error (logger extracts stack automatically)

Pitfalls

removeConsole in next.config.ts

Next.js compiler.removeConsole strips console.* calls from production bundles. This silently kills logger output since the logger calls console.debug/console.info/console.warn/console.error internally.

Fix: Disable removeConsole. The logger handles level filtering internally.

// next.config.ts
compiler: {
  // Console stripping disabled — structured logger handles level filtering
  // internally and emits JSON for Axiom log drain ingestion.
},

next-axiom is NOT a Drop-in

The next-axiom package does NOT auto-capture console.* calls. It provides its own Logger API that must be used explicitly. If you already have createLogger, adding next-axiom is redundant. Vercel Log Drains achieves the same result with zero dependencies.

Module-Level Evaluation

IS_PRODUCTION and the format function (JSON vs human) should be evaluated at module load time. This avoids a runtime check on every log call but means test environments always use the human formatter unless you mock NODE_ENV before import.