Bug #509
closed
The --read-timer option cause many drops of events when kernel and ust tracing are active
Added by Raphaël Beamonte over 11 years ago.
Updated over 11 years ago.
Description
The procedure to reproduce the problem is explained in the bug report #479.
The read-timer was introduced to be able to obtain an UST trace without having the application waiting during the "write" syscall. In our case, it effectively reduces the latency added by LTTng, but it drops many events.
Please describe your test conditions precisely:
- disk I/O throughput available,
- buffer size, for kernel, and for user-space,
- which events are enabled in the kernel, in user-space,
- what I/O throughput of trace is generated by your workload ?
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
- Status changed from New to Feedback
- Status changed from Feedback to Invalid
Following discussions, it appears that this is only reproduced when both kernel and ust tracers are doing high-throughput tracing, going beyond the available system I/O. It is expected that events get lost in this type of scenario.
Also available in: Atom
PDF