prom-client: Memory leak

Hi!

I currently have an implementation of prom client running that seems to continuously grow the node heap until an eventual crash/service restart when collecting API metrics. Currently, I am collecting the default metrics and a couple custom ones that just count websocket connections. These all work fine. The problems arise from the API metrics I try to collect using a middleware that sits on our Hapi routes. This middleware listens for ‘onRequest’ and then ‘response’ for collecting a response time in milliseconds, while adding labels for method, path, statusCode, and usernames.

Here is what the middleware looks like (I do not believe the problem is with this):

/**
* This is the middleware that looks at each request as it comes through the server
* @param {Object} server This is the instantiated Hapi server that we wish to listen on.
* @param {Object} options This is an object that can contain additional data.
* @param {Function} next This is the callback function that allows additional requests to continue.
* @return {Function} Callback function.
*/
const middlewareWrapper = (server, options, next) => {
  server.ext('onRequest', (request, reply) => {
    request.startTime = process.hrtime()
    return reply.continue()
  })
  server.on('response', (request) => {
    logic.apiPerformance(request.response.statusCode, request.startTime, sanitizeRequestData(request))
  })
  return next()
}

The metrics themselves are used inside of that apiPerformance function which looks like this:

  /**
  * This function takes various information about a request, performs some basic calculations, and then sends that off to loggly.
  * @param {Integer} statusCode This is the HTTP statusCode that will be returned to the user with the response.
  * @param {Array} startTime The difference in time from the request recieved to the response given. https://nodejs.org/api/process.html#process_process_hrtime_time
  * @param {Object} responseData An object with sanitized response data.
  */
  apiPerformance: (statusCode, startTime, responseData) => {
    const diff = process.hrtime(startTime)
    const responseTime = ((diff[0] * 1e3) + (diff[1] * 1e-6))
    const time = new Date()
    logger.transports.loggly.log('info', 'Route Log', {
      responseCode: statusCode,
      route: responseData.route,
      host: responseData.host,
      port: responseData.port,
      method: responseData.method,
      userData: responseData.userData,
      responseTime: `${responseTime}ms`,
      params: responseData.params,
      timestamp: time.toISOString(),
      tags: ['events', 'api', 'performance']
    }, (error, response) => {
      if (error) console.log(error)
    })
    if (config.metrics.prometheus.enabled) {
      requestDurationSummary.labels(responseData.method, responseData.route, statusCode, responseData.userData.displayName).observe(responseTime)
      requestBucketsHistogram.labels(responseData.method, responseData.route, statusCode, responseData.userData.displayName).observe(responseTime)
    }
  }

The metrics are registered at the top of that file like so:

const requestDurationSummary = prometheus.createSummary('http_request_duration_milliseconds', 'request duration in milliseconds', ['method', 'path', 'status', 'user'])
const requestBucketsHistogram = prometheus.createHistogram('http_request_buckets_milliseconds', 'request duration buckets in milliseconds.', ['method', 'path', 'status', 'user'], [ 500, 2000 ])

Node: v6.1.0 Hapi: v16.4.3 prom-client: v9.1.1

We are including prom-client as a part of a custom package (its basically a common libs package) that we include with our projects. Inside of this package, we include prom-client and expose only the methods we use (collectDefaultMetrics, registerMetrics, createGauge, createSummary, createHistogram). Another thing to note is that all metrics except http_request_duration_milliseconds and http_request_buckets_milliseconds (the two metrics posted above) are registered inside of the application we are tracking metrics for. Those two are registered in the code I posted above, which is within the custom package itself.

Any other information I can provide please let me know. Any help or suggestions are greatly appreciated. Thanks!