- Line of setup
- 1
- app.use(traxx.middleware())
- Releases
- 11
- Published on npm
- Alert channels
- 3
- Slack, Teams, Google Chat
- Write batches
- 3s
- Bulk inserts to MongoDB
01
The problem
Most Express apps start with no visibility. When a route gets slow or starts throwing 500s, the first to notice is a user. The usual fixes are heavy — an APM agent, a hosted vendor, a dashboard to configure — when a small team really wants one row per request that it can query.
Traxx is that row. It stores every request as its own MongoDB document, with no pre-aggregation, so any question — the slowest routes for one tenant, the errors since a deploy — is just a query. The one rule: tracking must never make the request slower.
02
How it works
- 01
Middleware
Starts a process.hrtime() timer and wraps res.json, res.send and next() to capture the response body and any error.
- 02
On finish
Once the response is flushed, it computes latency, resolves the route pattern (/shop/:id, not /shop/42) and builds the log.
- 03
BullMQ queue
The log is pushed to a Redis-backed queue. The user already has their response — nothing waits on the database.
- 04
Worker
Drains the queue into a buffer and sends an alert for every status code you configured.
- 05
Bulk insert
Every 3 seconds the buffer goes to MongoDB in a single insertMany.
03
Key decisions
Queue the write, don't await it
Writing to MongoDB inline adds a network round-trip to every request. A BullMQ queue takes it off the critical path and absorbs bursts.
Ruled outInline database writes inside the middleware
Raw rows, not aggregates
One document per request keeps every question answerable later. Aggregation becomes a query, not a schema decision made up front.
Ruled outPre-aggregated counters per route
Batch the inserts
An insertMany every few seconds turns thousands of tiny writes into a handful of bulk ones.
Ruled outOne insert per queued job
Custom fields per middleware
Pass { tenantId } and every log carries it — multi-tenant analytics without a schema change.
Pluggable alert channels
Slack, Teams and Google Chat sit behind one notifier, and the trigger is either a list of status codes or a { min, max } range.
Ship the types
The package ships its own index.d.ts, so TypeScript users need no @types install.
04
In the code
const express = require("express");
const Traxx = require("traxx");
const app = express();
const traxx = new Traxx({
mongoUri: process.env.MONGO_URI,
redisUri: process.env.REDIS_URI,
notifications: {
statusCodes: { min: 500, max: 599 },
channels: [{ type: "slack", options: { webhookUrl: process.env.SLACK_WEBHOOK_URL } }],
},
});
// Every request on every route is now tracked
app.use(traxx.middleware());
// Tag logs per tenant for multi-tenant analytics
app.use((req, res, next) => traxx.middleware({ tenantId: req.tenantId })(req, res, next));
app.listen(8080, async () => {
await traxx.init(); // connects Mongo + Redis, starts the worker
});05
What I'd do next
Close the crash window. The worker acknowledges a job as soon as it's buffered, so a crash before the next flush loses up to 3 seconds of logs. Acknowledging after the bulk write fixes it.
Redact by default. Request and response bodies are stored as-is. Before it sees personal data, it needs a redaction list for fields like password and token.
Cap the storage. Logs are kept forever today; a TTL index on timestamp would bound the collection.
Run the worker apart. The worker starts inside the app process. Under heavy traffic it should run as its own service.
06
Stack
- Runtime
- Node.js ≥ 16Express.js
- Queue
- BullMQRedis (ioredis)
- Storage
- MongoDBMongoose
- Alerts
- SlackMicrosoft TeamsGoogle Chat
- Distribution
- npmTypeScript declarations