mercurius-apollo-tracing

Fastify plugin to be used with Mercurius to collect performance metrics from your Graphql resolvers and send them to apollo studio.

Install

npm i mercurius-apollo-tracing

Usage

plugin can be registered like this:

const fastify = require('fastify')
const mercuriusTracing = require('mercurius-apollo-tracing')
const app = fastify()

// you need this if you want to be able to add the server to apollo studio
// they ping your server directly from the browser
app.register(require('fastify-cors'))

app.register(mercurius, {
  schema,
  resolvers,
  graphiql: true
}) // must be done before registering mercuriusTracing

app.register(mercuriusTracing, {
  apiKey: 'your:Api:Key', // replace 'your:Api:Key' with the one from apollo studio
  graphRef: 'yourGraph@ref' // replace 'yourGraph@ref'' with the one from apollo studio
})

API

Manual flush

You can flush traces manually at any time by :

app.apolloTracingStore.flushTracing()

Options

• endpointUrl?: string
• graphRef: string
• apiKey: string
• sendReportsImmediately?: boolean default: false
  • useful for lambda-like environment where the whole process exits right after serving the GQL request
• reportIntervalMs?: number default: 10000
• maxUncompressedReportSize?: number default: 4194304 maximum size of the payload in bytes
  • apollo studio ingress endpoint might not be able to handle bigger payloads, so we recommend tweaking this option carefully
• checkReportSizeRequestCountInterval?: number default: 100
  • defines how often the size of the metrics payload is checked in number of GQL requests. Lower value means more frequent byte size checks are performed on the traces awaiting to be sent.

Lambda

If you are running in lambda-like environment, keep in mind to pass sendReportsImmediately: true flag to registration options to make sure the report is send before process exits.

Performance

Plugin hooks into each resolver in your schema, so performance will be negatively affected. Performance will be impacted more if you have many fast/small resolvers. If you have less resolvers and it takes more time to resolve them, perf difference is lower. We've measured it in the benchmark here and observed slow down of roughly 25 percent. This is still smaller penalty than compared to running with tracing on apollo-server-fastify. Also in real world API where resolvers actually do something the difference will be much smaller though. These were actual results:

server	Without tracing(Requests/s)	With tracing enabled(Requests/s)
apollo-server-fastify+graphql-jit	4162.8	1478.4
mercurius	9162	6866

Ran on Ubuntu 21.04, Node 16.7.0 and AMD Ryzen 5900x.

Persisted queries

Yes this plugin works fine with them.