Gaps in aggregated (day) charts, while hourly charts are OK

[Martin Loeschner]

I've set up my monit-clients to poll every 30 seconds. The charts for secons and hours are OK (gap-free) but the day chart shows large gaps of 30 minutes. Monit clients report according to plan, with 'Acceptable report skew' set to 1 cycles for localhost and 2 cycles for remote hosts. No downtimes reported.

I've attached a png with a screenshot. I'm using M/Monit 3.3-linux-x64 with Monit 5.6/5.10 and a MySQL-DB.

[Tildeslash]

Hello,

the problem could be caused by time synchronisation issue - the data are aggregated and if the time jumps, the aggregated data point may be skipped. We have observed time issues like this in the past with some ARM devices, which had problem with keeping correct time and have been periodically synchronised with timeserver, where the time jumped by several minutes regularly.

If the time is correct, we'll need M/Monit logs (error.log and mmonit.log) and the dump of statistics_aggregate_1m and statistics_aggregate_15m tables.

You can use [e-mail] for next communication (sending log files, dumps, etc.)

[Martin Loeschner]

Hello,

after checking ntp synchronisation I'll wait until tomorrow and get back to you then. Thanks.

[Martin Loeschner]

Hi,

the problem persist despite proper configured, working ntp. Log files and mysql-dumps sent. Thanks for investigation.

[Tildeslash]

Hello,

thanks for data, the analysis showed that the problem is caused by volatile system clock, more details sent by mail. When the time jumps, it skips the timestamp for which the data are missing and writes to different timestamp, which then contains 2x more data then usual, as it was passed twice.

The SQL statement which reveals the duplicates:

select count(*), collectedsec from statistics_aggregate_15m group by collectedsec;

[Tildeslash]

closing the issue - the problem is caused by system clock

[Tildeslash]

Removing component: M/Monit (automated comment)