PMM-2612: Scrape MySQL database only once and fix scrape counter.

[arvenil]

There are two issues with scrapping I would like to address:

• mysql_exporter_scrapes_total never increases

  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 2
  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 2
  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 2
  

  Add 'collect' URL parameter to filter enabled collectors #235 (comment)

• mysqld_exporter every request scrapes source database twice - this means double load for MySQL database itself. The scrape is done twice because Describe() also runs Collect(). Additionally, every request we register new collector so every request we call Describe() and Collect(). This is different than in other exporters where exporter is registered only once so Describe() is called only on first http request but not on subsequent. This was also true for this exporter up to this change https://github.com/prometheus/mysqld_exporter/pull/235/files#r198540277
  In our fork we fixed first issue already, but this resulted in:

  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 2
  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 4
  $ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
  mysql_exporter_scrapes_total 6
  

  Which clearly shows problem that database is now scrapped twice for every request.

This PR fixes both issues so it results in:

$ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
mysql_exporter_scrapes_total 1
$ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
mysql_exporter_scrapes_total 2
$ curl --silent localhost:9104/metrics | grep scrapes_total | grep -v '#'
mysql_exporter_scrapes_total 3

And internally database is indeed scrapped only once per http request.

[SuperQ]

Very nice, LGTM.

[juliusv]

Since UP is not an initialism (https://github.com/golang/go/wiki/CodeReviewComments#initialisms):

MySQLUP -> MySQLUp

[juliusv]

So we basically don't even try to register most of the metrics descs for this collector anymore? That's probably fine as it was best-effort anyway and we only have this one collector, but was surprised it wasn't explicitly mentioned.

[arvenil]

I've actually made a comment about it but looks like I forgot to press "Send" ;)

Yes. I think it doesn't make sense but your expertise here guys is highly required. To be honest I don't see when registering those descriptions could be useful, maybe if you write your own app, then you register all metrics first, they get cached, so you can just focus on sending metrics. But for standalone exporter where we explicitly say we can't determine all metrics I don't see the reason to do that. Any idea what the lack of registration of descriptions could impact? I didn't looked into implementation details so again, your knowledge here could be useful ;)

Also registering those metrics made sense as long as collector was registered once, on binary start, so first http request was calling Describe() but next request were using cached descriptions and Describe() was not called. Now we register collector every scrape, so Describe is called every http request anyway, cached descriptions are not re-used.

Initially I thought my idea was ugly hack, but I checked later and node_exporter does the same: https://github.com/prometheus/node_exporter/blob/master/collector/collector.go#L106-L110 They also in similar way register collector every http request https://github.com/prometheus/node_exporter/blob/master/node_exporter.go#L47

[juliusv]

Yeah, it's fine in this case.

[juliusv]

👍

[arvenil]

@SuperQ This closes also #245 and #269 ?