Changes

This release brings some changes that you should be aware of when upgrading.

Timeouts

The exporter will now try to return within Prometheus timeout, even if all collectors have not yet finished. A new metric wmi_exporter_collector_timeout{collector="collector-name"} allows seeing when this happens. The exporter does this by reading the scrape timeout automatically sent by Prometheus in the X-Prometheus-Scrape-Timeout-Seconds header, and subtracting a "safety margin" (default 0.5 seconds), then aborting when that time has passed. If your system is heavily loaded, you may need to increase this margin, and can do so with the new --scrape.timeout-margin flag.

Perflib

This release contains the first use of Perflib in the cpu collector to give a lower-overhead way of fetching data than WMI. If this works out well, we expect to convert a majority of the existing collectors to Perflib, which should alleviate a lot of the issues around WMI we've seen.

Enhancements

• More detailed CPU metrics on Windows 2008R2 and higher (#334, thanks @wizardmatas!)
• ThermalZone collector for reading system temperature (#359, thanks @StewartThomson!)
• SQL collector now exposes error counters (#370, thanks @callvirtual!)
• SQL collector now exposes transaction metrics (#372, thanks @callvirtual!)
• Convert cpu collector to perflib (#335)
• Timeout scrapes (#340)

Bugfixes

• MSMQ collector was logging erronously (#327, thanks @marek-vrana!)
• CPU usage metrics should be counters (#331, thanks @nbellowe!)
• MSSQL metric dbreplica_transaction_delay_seconds was reported in microseconds instead of seconds (#351, thanks @sqlkabouter!)