Innodb_row_lock_waits values are incorrectly converted to milli units.

Bug #1311930 reported by markus_albe
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Monitoring Plugins
Invalid
Undecided
Unassigned

Bug Description

See attached graph, and the data below:

4/23/2014 13:21 0.0172222222
4/23/2014 13:22 0.065
4/23/2014 13:23 0.1509401709
4/23/2014 13:24 0.1624941725
4/23/2014 13:25 0.1256439394
4/23/2014 13:26 0.1256439394
4/23/2014 13:27 0.0847222222
4/23/2014 13:28 0.0473931624
4/23/2014 13:29 0.0375058275
4/23/2014 13:30 0.1301010101
4/23/2014 13:31 0.1494444444

I don't think a lock wait can be a fractional figure, and this seems just a wrong template that's converting the sampled values to milli units (diving by 1000).

Tags: i41494
Revision history for this message
markus_albe (markus-albe) wrote :
Revision history for this message
Roman Vynar (roman-vynar) wrote :

show global status like 'Innodb_row_lock_waits' is constantly increasing counter, thus it is set to DERIVE in the template.
This means the value that will be stored in RRD is calculated this way:
(LAST_VAL - CURR_VAL) / 300

It's difficult to figure out exactly from the information provided but the numbers you have listed seems to be coming from the graph itself averaging every minute as MySQL output should not be a fractional and RRD should store only every-5-minute values by default, e.g. 13:25, 13:30 etc.

For example, if on 13:25 we have Innodb_row_lock_waits = 2783639 and on 13:30 = 2783748, this means, that the following value gets stored in RRD:
(2783748 - 2783639) / 300 = 0.36333
Cacti will show 363.3m on the Daily (5 Minute Average) graph. However, the Hourly (1 Minute Average) graph may divide this by 5 additionally.

So the graph "Innodb Row Lock Waits" will be showing a fractional figure anyway, as it reflects the change in Innodb_row_lock_waits averaged (divided by) 300 seconds (it's the default cacti 5 min. poller interval). Of course, unless you have more than 300 row lock waits for the 5 min. and you will get 1 (1000m) on the graph.

Changed in percona-monitoring-plugins:
status: New → Incomplete
Changed in percona-monitoring-plugins:
status: Incomplete → Invalid
Revision history for this message
markus_albe (markus-albe) wrote :

I would say that the only possibl thing left to be done is documenting the type of aggregation for this -and potentially other similar- graphs.

Revision history for this message
Roman Vynar (roman-vynar) wrote :

Markus, actually that's the basics of RRD data source types which Cacti uses, see:
- http://oss.oetiker.ch/rrdtool/doc/rrdcreate.en.html
- http://docs.cacti.net/faq#what_does_u_m_p_etc_stand_for

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.