metrics

Module provides the interface described in @platform/metrics-types. On server the interface is implemented with public package prom-client that provides metrics on url /metrics Prometheus format.

More details about metrics type, parameters and how to use it see in docs to prom-client.

Explanation

Monitoring outgoing requests

To monitor the state of the outgoing requests (like number of requests, number of error, time execution) the module monkey-patches request and get methods of the standard modules http and https. To make it work just add metrics module to the app.

Next labels are added to metrics:

• http method
• http response code
• service name

Make service names showed in metrics instead of hostnames

Name of the service calculates by comparing request urls with values in MetricsServicesRegistry. Initially the register is bootstrapped with the inverted content of env variables. For example if some url from env is a substring of the request url, then the name of the env become the service name. If several envs matches this logic then the env with the longest url is used. If none of envs matches then domain used as service name label.

It is possible to give a hint to module about the service name in case url is dynamic. To do that:

• use token METRICS_SERVICES_REGISTRY_TOKEN;
• call metricsServicesRegistry.register("Part of the url or the whole url", "Name of service")

To explicitly associate the service name label with request, you should provide the x-tramvai-service-name header. This header will be consumed by the metrics module and then removed.

Event Loop Lag

This module has their own implementation of Event Loop Lag metric - nodejs_eventloop_setinterval_lag_seconds histogram, this metric implemented with setTimeout.

Client metrics

Module implements feature to collect metrics from the clients and share it with Prometheus by sending metrics from the client to server papi-route.

Metrics module can help in implementing this functionality in common cases. To create metric register provider for the token REGISTER_INSTANT_METRIC_TOKEN. Your provider should return list of two entities - first is a slug of papi-route and second is an instance of Counter. E.g.:

import { provide } from '@tramvai/core';

provide({
  provide: REGISTER_INSTANT_METRIC_TOKEN,
  scope: Scope.SINGLETON,
  multi: true,
  deps: {
    metrics: METRICS_MODULE_TOKEN,
  },
  useFactory({ metrics }) {
    return ['page-load', new Counter({ name: 'client_page_load_total', help: 'Client page load' })];
  },
});

After that to increment metric client_page_load_total you can call papi-route /metrics/page-load.

instantMetricsReporter

In practice it become clear that besides metric collection it often needed to collect logs with details. This can be implemented with instantMetricsReporter. When calling logger module will check that any metric with the slug equal to the event of the log is exist. If so module will send request to the corresponding papi-route.

Next way you can log event and increment server metric:

import { provide } from '@tramvai/core';
provide({
  provide: commandLineListTokens.init,
  multi: true,
  deps: {
    logger: LOGGER_TOKEN,
  },
  useFactory({ logger }) {
    return () => {
      window.on('load', () => {
        logger.info({ event: 'page-load' });
      })
    };
  },
}),

Metrics list

Application metrics

• http_requests_total counter - application response count by status code
• http_requests_execution_time histogram - application response time
• command_line_runner_execution_time histogram - (measure application lifecycle)[03-features/06-app-lifecycle.md]

Outgoing requests

• http_sent_requests_total counter - request count from application to external APIs
• http_sent_requests_duration histogram - request time from application to external APIs
• http_sent_requests_errors counter - request from application to external APIs errors count
• dns_resolve_duration histogram - DNS resolve time
• tcp_connect_duration histogram - TCP connect time
• tls_handshake_duration histogram - TLS handshake time

Node.js metrics

• nodejs_eventloop_lag_p90_seconds gauge - event loop lag in 90 percentile from prom-client
• nodejs_eventloop_setinterval_lag_seconds histogram - event loop lag from custom setTimeout measurement
• nodejs_heap_space_size_used_bytes gauge - used memory size from prom-client
• nodejs_gc_duration_seconds histogram - GC duration from prom-client
• nodejs_active_handles gauge - total number of active handles from prom-client
• nodejs_active_requests gauge - total number of active requests from prom-client

How to

Usage Example

import { createToken } from '@tinkoff/dippy';
import { Module, provide } from '@tramvai/core';
import { Counter, Metrics, METRICS_MODULE_TOKEN } from '@tramvai/tokens-metrics';

interface SomeModuleOptions {
  metrics: Metrics;
}

class SomeModule {
  private metricActionCounter: Counter;

  constructor(options: SomeModuleOptions) {
    this.metricActionCounter = options.metrics.counter({
      name: 'some_actions_total',
      help: 'Total count of some actions',
    });
  }

  public action(): void {
    this.metricActionCounter.inc();

    // Do some meaningful action
  }
}

export const SOME_MODULE = createToken<SomeModule>('someModule');

@Module({
  providers: [
    provide({
      provide: SOME_MODULE,
      useFactory: (deps) => new SomeModule(deps),
      deps: {
        metrics: METRICS_MODULE_TOKEN,
      },
    }),
  ],
})
export class SomeModuleContainer {}

Use metrics to profile performance in browser

To measure length of the events you must use method startTimer of classes Gauge, Histogram, Summary. In dev-mode these classes are patched and methods to work with timers will use PerformanceApi.

Example without additional fields:

const metric = metrics.gauge({
  name: 'request_measure',
  help: 'Request duration measure',
});

const endTimer = metric.startTimer();

fetch(url).then(() => {
  endTimer();

  // output the result - performance.getEntriesByName('request_measure');
});

Example with adding dynamic fields:

const metric = metrics.gauge({
  name: 'request_measure',
  help: 'Request duration measure',
});

const endTimer = metric.startTimer({ method: 'GET' });

fetch(url).then(() => {
  endTimer({ status: 200 });

  // output the result - performance.getEntriesByName('request_measure{method="GET",status="200"}');
});

Use custom port for metrics endpoint

It can be done with token UTILITY_SERVER_PORT_TOKEN as it works for any utility path:

Debug

The module uses loggers with the next ids: metrics:perf, metrics:papi

Exported tokens

link