Background

The MetricsPlatform extension fetches instrument configs from an MPIC instance running on dse-k8s. During initial deployment at least, we should be actively monitoring the performance of this system.

At any time we will want to know:

• The number of MPIC API requests being made
• The number of MPIC API requests succeeding
• The number of MPIC API requests failing
• The number of MPIC API requests timing out
• The number of malformed API responses
• The median, p75, p95, p99 round-trip-time (RTT) of the API requests

AC

• The above are instrumented
• A Grafana dashboard is created

Requirements

• QA passed?
• Documentation
  • https://www.mediawiki.org/wiki/Extension:MetricsPlatform should be updated

Notes

1. Read https://www.mediawiki.org/wiki/Manual:Stats and https://prometheus.io/docs/practices/naming/.

• The number of MPIC API requests succeeding
• The number of MPIC API requests failing
• The number of MPIC API requests timing out

We should use the Stats library provided by MediaWiki Core to do this, e.g.

if ( $isSuccess ) {
  $label = 'success';
} else if ( $isTimeout ) {
  $label = 'timeout';
} else {
  $label = 'failure';
}

MediaWikiServices::getInstance()->getStatsFactory()
  ->withComponent( 'MetricsPlatform' )
  ->getCounter( 'mpic_api_request_total' )
  ->setLabel( $label )
  ->increment();

• * The median, p75, p95, p99 round-trip-time (RTT) of the API requests

$startTime = microtime( true );

// ...

MediaWikiServices::getInstance()->getStatsFactory()
  ->withComponent( 'MetricsPlatform' )
  ->getTiming( 'mpic_api_request_duration_seconds' )
  ->observe( ( microtime( true ) - $startTime ) * 1000 );