[collectd] Single plugin runs bridging timestamps

[Paul Sadauskas]

I have a question about how collectd uses the timestamp when it
collects and records values.

I captured some output from the write_http plugin, filtered to show
two sequential runs of the memory plugin.

PUTVAL ubuntu.example/memory/memory-used interval=30 1330142682:70070272.000000
PUTVAL ubuntu.example/memory/memory-buffered interval=30
1330142683:55861248.000000
PUTVAL ubuntu.example/memory/memory-cached interval=30
1330142683:191954944.000000
PUTVAL ubuntu.example/memory/memory-free interval=30 1330142683:67403776.000000

PUTVAL ubuntu.example/memory/memory-used interval=30 1330142712:70057984.000000
PUTVAL ubuntu.example/memory/memory-buffered interval=30
1330142712:55881728.000000
PUTVAL ubuntu.example/memory/memory-cached interval=30
1330142712:191954944.000000
PUTVAL ubuntu.example/memory/memory-free interval=30 1330142712:67395584.000000

In the first block, the first timestamp ends in -682, while the others
end in -683. In the second block, the timestamps are all the same,
-712.

Why aren't the all always the same over a single run of the plugin? I
would expect the calls to memory_submit(), and consequently
plugin_dispatch_values(), to happen within microseconds of each other,
making those timestamps crossing from one second to the next extremely
unlikely. However, when running with an interval of 30 seconds, it
happens 4-5 times over the course of an hour, around 3% of the time. I
checked, and those

Additionally, it happens on my Ubuntu testing VM (collected 4.10), but
not when I run collectd (5.0.2) on my OSX host. I should investigate
it on a non-VM ubuntu box and/or compile 5.0.2 for Ubuntu 10.04, to
isolate if its the OS, virtualization, or the collectd version, but I
was hoping the list could also provide some enlightenment.

Thanks for your time,
Paul Sadauskas