monitoring-plugins
monitoring-plugins copied to clipboard
Strategy: Convert all Continous Counters to Absolute Values
In GitLab by @markuslf on Jul 23, 2021, 16:23
With Continous Counters (unit "c") we have major disadvantages with regard to software like Grafana:
- Assume rendering a panel based on data for the last two years which was gathered once per minute. This means at least 1.1M measurements for one perfdata-item, or 15.8M measurements for 15 perfdata-item, and so on.
- To speed up a panel, Grafana has to aggregate values (for example by using mean()). For c-values, the calculated values are obviously wrong.
- With c-values, you can't design some meaningful and performant tables showing min(), max(), first(), last() or mean() values. c-values are missing any furter unit of measurements. You don't know if it's byte, percentage or second.
- Efficiency at all: Why always calculate the difference between two c-values in Grafana & Co. if this could be done by the monitoring plugin only once while gathering the data? This saves resources on the monitoring server.
This leads to:
- Convert all continous counters to absolute values.
- Never use continous counters at all.
After that, in all Grafana Panels dealing with c-values:
- Add meaningful units.
- Remove non-negative difference calculations.
- Add tables with values again.
In GitLab by @markuslf on Jul 23, 2021, 16:24
Continous Counter:
- CPU
- Usage interrupts
- soft_interrupts
- Disk I/O
_busy_time _read_bytes _read_merged_count _read_time _write_bytes _write_merged_count _write_time
- fs-xfs-stats
- PHP-FPM Status
- accepted conn
- PHP Status
- php-opcache-opcache_statistics-hits
- php-opcache-opcache_statistics-misses
- Rocket.Chat
- rc_totalLivechat
- rc_totalLivechatMessages
- rc_totalLivechatVisitors
- rc_totalDirectMessages
- rc_totalMessagesfs-xfs-stats
- rc_totalPrivateGroupMessages
- rc_totalPrivateGroups
- rc_totalRooms
- rc_totalUsers
- rc_uploadsTotal
- rc_uploadsTotalSize
- Starface Account Stats:
- closed_connections
- opened_connections
In GitLab by @markuslf on Feb 16, 2022, 21:53
unassigned @markuslf