Experiments >

IdP: Analytic events

Experiment #23328th July, 2022by Joshua Nussbaum

What’s unique about the IdP I’m building is it’s focused on SaaS.

One of the challenges I saw in building my last SaaS project is knowing usage stats of each user. If a user isn’t using the system, I’d like to be notified as soon as possible so I can chat with them and find out what’s wrong.

Users that don’t use the system are bad for business. Zombie users are going to churn anyway, might as well be direct with them and learn what problems they are having early.

It’s a canary in the coal mine.

Plus, solving these kind of problems can surface important features that make the product stickier for everyone.

Idea

It would be cool if the IdP could track analytics about each member. Resource servers would send usage stats to the IdP via a REST endpoint. This would be in addition to to the IdP tracking the last sign-in time for each user (users that don’t sign in aren’t using the product and that’s important to know too.)

The IdP would use a ClickHouse database to store these kind of events.

Code

Let’s create a SvelteKit project with ClickHouse integration:

  1. Create a SvelteKit project
pnpm create svelte@next idp-analytics-experiment
  1. Create the ClickHouse database via Docker
# download clickhouse docker image
docker pull yandex/clickhouse-server

# start docker container, expose port 9000, 8123
docker run -d --name clickhouse -p 9000:9000 -p 8123:8123 yandex/clickhouse-server

# connect to clickhouse
docker exec -it clickhouse clickhouse-client

Inside clickhouse-client, run SQL to create the database and table:

CREATE DATABASE idp;

USE idp;

CREATE TABLE events (
  account_id UInt64,
  event_type String,
  inserted_at DateTime default now()
)
ENGINE MergeTree
ORDER BY inserted_at;
  1. Add the ClickHouse npm package
pnpm i -D clickhouse
  1. Create the data access abstraction
// src/lib/analytics.js
import { ClickHouse } from 'clickhouse'

const db = new ClickHouse({ database: 'idp' })

// insert rows into event table
export async function trackEvent(account_id, event_type) {
  const row = { account_id, event_type }

  return await db.insert('INSERT INTO events (account_id, event_type)', [row]).toPromise()
}
  1. Create a SvelteKit route
// src/routes/event.js
import { trackEvent } from '$lib/analytics'

export async function POST({ request }) {
  const { accountId, event } =  await request.json()

  try {
    await trackEvent(accountId, event)

    return {
      status: 200,
      body: { message: 'success' }
    }
  } catch {
    return {
      status: 500,
      body: { message: 'database error' }
    }
  }
}

Demo

Now we can hit the test server with events, and they should show up in ClickHouse:

curl localhost:3000/event 
  -H 'content-type: application/json' 
  --data '{"accountId": 1, "event": "search"}'

Notes

  • This should work with Prisma too, since ClickHouse supports the Postgres wire format. But DDL with prisma db push likely won’t work.
view all experiments

Stay tuned in

Learn how to add more experimentation to your workflow