MongoDB exporter timeouts with a lots of collections

User Impact

No metrics are collected for MongoDB.

Description

When a MongoDB instance is being added to PMM, even if the limit of 200 collections is active, the metrics collections is taking too much time.

Here is a debug output:

./mongodb_exporter --collector.collstats --collector.collstats-limit=20000 --collector.dbstats --collector.diagnosticdata --collector.indexstats --collector.replicasetstatus --collector.topmetrics --compatible-mode --discovering-mode --mongodb.global-conn-pool --mongodb.uri=mongodb://127.0.0.1:27018
level=info ts=2022-01-13T18:16:56.477Z caller=tls_config.go:195 msg="TLS is disabled." http2=false
Start: 2022-01-13 18:17:01.830445413 +0000 UTC m=+5.365602998Finish: 2022-01-13 18:17:01.90742765 +0000 UTC m=+5.442585225
Total collections count took 76.982227ms
Collstats collector took 20.625976294s
DBStats collector took 134.991127ms
DBStats collector took 679.660172ms
Top collector took 4.28177123s
Collstats collector took 23.33693319s
 

As shown in the example above, Collstats, DBStats and Top Collectors are taking more than 1 second (which is the default timeout in high resolution).

Also Collectors Collect method is being called twice because since the list of metrics is being made on the fly, on the Describe function is calling prometheus.DescribeByCollect which actually collects the metrics to create a list of metric names so, Collect is being called twice.

Even without Collstats, DBStats and Top are taking too much time, 4-5 Sec. So with smaller limit for Collstat - it would timeout anyway.