Metrics

Prisma Client metrics give you a detailed insight into how Prisma Client interacts with your database. You can use this insight to help diagnose performance issues with your application.

info

If you want an even more detailed insight into your Prisma Client's performance, at the level of individual operations, see Tracing.

About metrics

You can export metrics in JSON or Prometheus formats and view them in a console log, or integrate them into an external metrics system, such as StatsD or Prometheus. If you integrate them into an external metrics system, then you can view the metrics data over time. For example, you can use metrics to help diagnose how your application's number of idle and active connections changes.

Prisma Client provides the following metrics:

• Counters (always increase):

  • prisma_client_queries_total: The total number of Prisma Client queries executed.
  • prisma_datasource_queries_total: The total number of datasource queries executed (SQL queries in relational databases, and commands in MongoDB).
    • The value returned by prisma_datasource_queries_total can be greater than prisma_client_queries_total, because some Prisma Client operations create multiple queries.
  • prisma_pool_connections_closed_total: The total number of pool connections closed.
  • prisma_pool_connections_opened_total: The number of currently open pool connections.

• Gauges (can increase or decrease):

  • prisma_client_queries_active: The number of currently active Prisma Client queries.
  • prisma_client_queries_wait: The number of Prisma Client queries currently waiting for a connection because all connections are in use.
  • prisma_pool_connections_busy: The number of currently busy pool connections. These pool connections are currently executing a datasource query.
  • prisma_pool_connections_idle: The number of pool connections that are not currently being used. These pool connections are waiting for the next datasource query to run.
  • prisma_pool_connections_open: The number of pool connections open.

• Histograms (metrics data divided into a collection of values; we call each container in the collection a "bucket"):

  • prisma_client_queries_wait_histogram_ms: The time waiting for a pool connection for all Prisma Client queries in ms.
  • prisma_client_queries_duration_histogram_ms: The execution time for all executed Prisma Client queries in ms. This includes the time taken to execute all database queries, and to carry out all database engine activities, such as joining data and transforming data to the correct format.
  • prisma_datasource_queries_duration_histogram_ms: The execution time for all executed Datasource queries in ms.

You can add global labels to your metrics data to help you group and separate your metrics, for example by infrastructure region or server.

Prerequisites

To use Prisma Client metrics, you must do the following:

1. Install the appropriate dependencies.
2. Enable the metrics feature flag in your Prisma schema file.

Step 1. Install up-to-date Prisma ORM dependencies

Use version 3.15.0 or higher of the prisma and @prisma/client npm packages.

npm install prisma@latest --save-dev
npm install @prisma/client@latest --save

Step 2: Enable the feature flag in the Prisma schema file

In the generator block of your schema.prisma file, enable the metrics feature flag:

generator client {
  provider        = "prisma-client-js"
  previewFeatures = ["metrics"]
}

Retrieve metrics in JSON format

When you retrieve metrics in JSON format, you can use them in the format they are returned, or send them to StatSD to visualize how they change over time.

To retrieve metrics in JSON format, add the following lines to your application code:

const metrics = await prisma.$metrics.json()
console.log(metrics)

This returns metrics as follows:

{
  "counters": [
    {
      "key": "prisma_client_queries_total",
      "labels": {},
      "value": 0,
      "description": "Total number of Prisma Client queries executed"
    },
    {
      "key": "prisma_datasource_queries_total",
      "labels": {},
      "value": 0,
      "description": "Total number of Datasource Queries executed"
    },
    {
      "key": "prisma_pool_connections_closed_total",
      "labels": {},
      "value": 0,
      "description": "Total number of Pool Connections closed"
    },
    {
      "key": "prisma_pool_connections_opened_total",
      "labels": {},
      "value": 1,
      "description": "Total number of Pool Connections opened"
    }
    ...
  ],
  "gauges": [
    ...
  ],
  "histograms": [
    ...
  ]
}

Expand to view the full output

{
  "counters": [
    {
      "key": "prisma_client_queries_total",
      "labels": {},
      "value": 2,
      "description": "Total number of Prisma Client queries executed"
    },
    {
      "key": "prisma_datasource_queries_total",
      "labels": {},
      "value": 5,
      "description": "Total number of Datasource Queries executed"
    },
    {
      "key": "prisma_pool_connections_open",
      "labels": {},
      "value": 1,
      "description": "Number of currently open Pool Connections"
    }
  ],
  "gauges": [
    {
      "key": "prisma_client_queries_active",
      "labels": {},
      "value": 0,
      "description": "Number of currently active Prisma Client queries"
    },
    {
      "key": "prisma_client_queries_wait",
      "labels": {},
      "value": 0,
      "description": "Number of Prisma Client queries currently waiting for a connection"
    },
    {
      "key": "prisma_pool_connections_busy",
      "labels": {},
      "value": 0,
      "description": "Number of currently busy Pool Connections (executing a datasource query)"
    },
    {
      "key": "prisma_pool_connections_idle",
      "labels": {},
      "value": 21,
      "description": "Number of currently unused Pool Connections (waiting for the next datasource query to run)"
    },
    {
      "key": "prisma_pool_connections_open",
      "labels": {},
      "value": 1,
      "description": "Number of currently open Pool Connections"
    }
  ],
  "histograms": [
    {
      "key": "prisma_client_queries_duration_histogram_ms",
      "labels": {},
      "value": {
        "buckets": [
          [0, 0],
          [1, 0],
          [5, 0],
          [10, 1],
          [50, 1],
          [100, 0],
          [500, 0],
          [1000, 0],
          [5000, 0],
          [50000, 0]
        ],
        "sum": 47.430541000000005,
        "count": 2
      },
      "description": "Histogram of the duration of all executed Prisma Client queries in ms"
    },
    {
      "key": "prisma_client_queries_wait_histogram_ms",
      "labels": {},
      "value": {
        "buckets": [
          [0, 0],
          [1, 3],
          [5, 0],
          [10, 0],
          [50, 0],
          [100, 0],
          [500, 0],
          [1000, 0],
          [5000, 0],
          [50000, 0]
        ],
        "sum": 0.0015830000000000002,
        "count": 3
      },
      "description": "Histogram of the wait time of all Prisma Client Queries in ms"
    },
    {
      "key": "prisma_datasource_queries_duration_histogram_ms",
      "labels": {},
      "value": {
        "buckets": [
          [0, 0],
          [1, 0],
          [5, 2],
          [10, 2],
          [50, 1],
          [100, 0],
          [500, 0],
          [1000, 0],
          [5000, 0],
          [50000, 0]
        ],
        "sum": 47.134498,
        "count": 5
      },
      "description": "Histogram of the duration of all executed Datasource Queries in ms"
    }
  ]
}

Histograms in JSON data

Each histogram "bucket" has two values. The first one is the upper bound of the bucket, and the second one is the count (the number of data values that fall into that bucket). In the following example, there are two instances of values between 11 and 20, and five instances of values between 21 and 30:

...
[20, 2],
[30, 5],
...

Use Prisma Client metrics with StatsD

You can send JSON-formatted metrics to StatsD to visualize your metrics data over time.

info

Note: You must provide counter metrics to StatsD as a series of values that are incremented or decremented from a previous retrieval of the metrics. However, Prisma Client's counter metrics return absolute values. Therefore, you must convert your counter metrics to a series of incremented and decremented values and send them to StatsD as gauge data. In the code example below, we convert counter metrics into incremented and decremented gauge data in diffHistograms.

In the following example, we send metrics to StatsD every 10 seconds. This timing aligns with the default 10s flush rate of StatsD.

import StatsD from 'hot-shots'
let statsd = new StatsD({
  port: 8125,
})

let diffMetrics = (metrics: any) => {
  return metrics.map((metric: any) => {
    let prev = 0
    let diffBuckets = metric.value.buckets.map((values: any) => {
      let [bucket, value] = values
      let diff = value - prev
      prev = value
      return [bucket, diff]
    })
    metric.value.buckets = diffBuckets
    return metric
  })
}

let previousHistograms: any = null
let statsdSender = async () => {
  let metrics = await prisma.$metrics.json()

  metrics.counters.forEach((counter: any) => {
    statsd.gauge('prisma.' + counter.key, counter.value, (...res) => {})
  })

  metrics.gauges.forEach((counter: any) => {
    statsd.gauge('prisma.' + counter.key, counter.value, (...res) => {})
  })

  if (previousHistograms === null) {
    previousHistograms = diffMetrics(metrics.histograms)
    return
  }

  let diffHistograms = diffMetrics(metrics.histograms)

  diffHistograms.forEach((diffHistograms: any, histogramIndex: any) => {
    diffHistograms.value.buckets.forEach((values: any, bucketIndex: any) => {
      let [bucket, count] = values
      let [_, prev] =
        previousHistograms[histogramIndex].value.buckets[bucketIndex]
      let change = count - prev

      for (let sendTimes = 0; sendTimes < change; sendTimes++) {
        statsd.timing('prisma.' + diffHistograms.key, bucket)
      }
    })
  })

  previousHistograms = diffHistograms
}

setInterval(async () => await statsdSender(), 10000)

Retrieve metrics in Prometheus format

When you retrieve Prisma Client metrics in Prometheus format, you can use them in the format they are returned, or send them to the Prometheus metrics system to visualize how they change over time.

To retrieve metrics in Prometheus format, add the following lines to your application code:

const metrics = await prisma.$metrics.prometheus()
console.log(metrics)

This returns metrics as follows:

# HELP prisma_client_queries_total Total number of Prisma Client queries executed
# TYPE prisma_client_queries_total counter
prisma_client_queries_total 14

...
# HELP prisma_pool_connections_busy The number of active connections in use.
# TYPE prisma_pool_connections_busy gauge
prisma_pool_connections_busy 0

...
# HELP prisma_client_queries_wait_histogram_ms The wait time for a worker to get a connection.
# TYPE prisma_client_queries_wait_histogram_ms histogram
prisma_client_queries_wait_histogram_ms_bucket{le="0"} 0
prisma_client_queries_wait_histogram_ms_bucket{le="1"} 3

Expand to view the full output

# HELP query_total_operations
# TYPE query_total_operations counter
query_total_operations 2

# HELP prisma_datasource_queries_total
# TYPE prisma_datasource_queries_total counter
prisma_datasource_queries_total 28

# HELP prisma_pool_connections_closed_total Total number of Pool Connections closed
# TYPE prisma_pool_connections_closed_total counter
prisma_pool_connections_closed_total 0

# HELP prisma_pool_connections_opened_total Total number of Pool Connections opened
# TYPE prisma_pool_connections_opened_total counter
prisma_pool_connections_opened_total 0

# HELP prisma_client_queries_active Number of currently active Prisma Client queries
# TYPE prisma_client_queries_active gauge
prisma_client_queries_active 0

# HELP prisma_client_queries_wait Number of queries currently waiting for a connection
# TYPE prisma_client_queries_wait gauge
prisma_client_queries_wait 0

# HELP prisma_pool_connections_busy Number of currently busy Pool Connections (executing a datasource query)
# TYPE prisma_pool_connections_busy gauge
prisma_pool_connections_busy 0

# HELP prisma_pool_connections_idle Number of currently unused Pool Connections (waiting for the next pool query to run)
# TYPE prisma_pool_connections_idle gauge
prisma_pool_connections_idle 21

# HELP prisma_pool_connections_open Number of currently open Pool Connections
# TYPE prisma_pool_connections_open gauge
prisma_pool_connections_open 1

# HELP prisma_pool_connections_open Number of currently open Pool Connections (able to execute a datasource query)
# TYPE prisma_pool_connections_open gauge
prisma_pool_connections_open 0

# HELP prisma_client_queries_wait_histogram_ms The wait time for a worker to get a connection.
# TYPE prisma_client_queries_wait_histogram_ms histogram
prisma_client_queries_wait_histogram_ms_bucket{le="0"} 0
prisma_client_queries_wait_histogram_ms_bucket{le="1"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="5"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="10"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="50"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="100"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="500"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="1000"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="5000"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="50000"} 3
prisma_client_queries_wait_histogram_ms_bucket{le="+Inf"} 3
prisma_client_queries_wait_histogram_ms_sum 0.023208
prisma_client_queries_wait_histogram_ms_count 3

# HELP prisma_client_queries_duration_histogram_ms Histogram of the duration of all executed Prisma Client queries in ms
# TYPE prisma_client_queries_duration_histogram_ms histogram
prisma_client_queries_duration_histogram_ms_bucket{le="0"} 0
prisma_client_queries_duration_histogram_ms_bucket{le="1"} 1
prisma_client_queries_duration_histogram_ms_bucket{le="5"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="10"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="50"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="100"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="500"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="1000"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="5000"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="50000"} 2
prisma_client_queries_duration_histogram_ms_bucket{le="+Inf"} 2
prisma_client_queries_duration_histogram_ms_sum 3.197624
prisma_client_queries_duration_histogram_ms_count 2

# HELP prisma_datasource_queries_duration_histogram_ms Histogram of the duration of all executed Datasource Queries in ms
# TYPE prisma_datasource_queries_duration_histogram_ms histogram
prisma_datasource_queries_duration_histogram_ms_bucket{le="0"} 0
prisma_datasource_queries_duration_histogram_ms_bucket{le="1"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="5"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="10"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="50"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="100"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="500"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="1000"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="5000"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="50000"} 5
prisma_datasource_queries_duration_histogram_ms_bucket{le="+Inf"} 5
prisma_datasource_queries_duration_histogram_ms_sum 1.8407059999999997
prisma_datasource_queries_duration_histogram_ms_count 5

Metrics of type histogram expose three different class of values in the Prometheus format:

1. Multiple cumulative counters for observation buckets. These counters are suffixed with _bucket{le="<upper inclusive bound>"}. For example, prisma_datasource_queries_duration_histogram_ms has a counter exposed as prisma_datasource_queries_duration_histogram_ms_bucket{le="1"}

   When an observed value is less than or equal to the upper inclusive bound of a bucket, then Prisma Client metrics increments that bucket by 1. Suppose that you have buckets with the upper inclusive bounds 0, 1, 5, 10, and 50 respectively. If the observed value is 5 then Prisma Client metrics increments the third bucket onwards, because the value is greater than 0 and greater than 1, but less than or equal to 5, 10, and 50.

2. A single total sum for all observed values. This counter is suffixed with _sum. For example the total sum of prisma_datasource_queries_duration_histogram_ms is exposed as prisma_datasource_queries_duration_histogram_ms_sum.

3. The count of the number of events that have been observed. This counter is suffixed with _count. For example the total count of prisma_datasource_queries_duration_histogram_ms events is exposed as prisma_datasource_queries_duration_histogram_ms_count.

For more information, read the Prometheus documentation on metric types.

Use Prisma Client metrics with the Prometheus metrics system

In the majority of cases, Prometheus must scrape an endpoint to retrieve metrics. The following example shows how to send data with Express.js:

import { PrismaClient } from '@prisma/client'
import express, { Request, Response } from 'express'

const app = express()
const port = 4000
const prisma = new PrismaClient()

app.get('/metrics', async (_req, res: Response) => {
  const metrics = await prisma.$metrics.prometheus()
  res.end(metrics)
})

app.listen(port, () => {
  console.log(`Example app listening on port ${port}`)
})

The following example shows how to combine Prisma Client metrics with other Prometheus client libraries that are also served with a REST API endpoint in conjunction with Express.js:

import { PrismaClient } from '@prisma/client'
import express, { Request, Response } from 'express'
import prom from 'prom-client'

const app = express()
const port = 4000
const prisma = new PrismaClient()

const register = new prom.Registry()
prom.collectDefaultMetrics({ register })

app.get('/metrics', async (_req, res: Response) => {
  const prismaMetrics = await prisma.$metrics.prometheus()
  const appMetrics = await register.metrics()
  res.end(prismaMetrics + appMetrics)
})

app.listen(port, () => {
  console.log(`Example app listening on port ${port}`)
})

Global labels

You can add global labels to your metrics to help you group and separate your metrics. Each instance of Prisma Client adds these labels to the metrics that it generates. For example, you can group your metrics by infrastructure region, or by server, with a label like { server: us_server1', 'app_version': 'one' }.

Global labels work with JSON and Prometheus-formatted metrics.

For example, to add global labels to JSON-format metrics, add the following code to your application:

const metrics = prisma.$metrics.json({
  globalLabels: { server: 'us_server1', app_version: 'one' },
})
console.log(metrics)

This returns information in the following format:

{
  "counters": [
    {
      "key": "query_total_operations",
      "labels": { "server": "us_server1", "app_version": "one" },
      "value": 0,
      "description": "The total number of operations executed"
    },
    {
      "key": "prisma_datasource_queries_total",
      "labels": { "server": "us_server1", "app_version": "one" },
      "value": 0,
      "description": "The total number of queries executed"
    },
    ...
  ],
  "gauges": [
    ...
  ],
  "histograms": [
    ...
  ]
}