Skip to contentSkip to conversation

state tracking and missing statedump end

DISCUSSION Alexandre Montplaisir 2 participants 2 messages May 15, 2015 2 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

Hi Mathieu,

Thanks for the info, good to know!

However, we (in Trace Compass) don't actually make use of the "statedump_end" event. Should we?

For now, we only use the "lttng_statedump_process_state" events in our state model. The information contained in these events is applied to the state of the involved process. This state becomes valid starting at the timestamp of each statedump event.

What type of information does the "lttng_statedump_end" event implies?

Cheers,
Alex

Quoted 24 lines · Mathieu Desnoyers

Hi Alexandre, Hi Matthew,

You will certainly be interested in this commit in lttng-modules:

commit 84c7055e17f9bf7628fe1004db732d932ed2a77b

Author: Mathieu Desnoyers <[email protected]>
Date:   Thu May 14 18:29:58 2015 -0400

Fix: missing statedump end event when block device listing available

Incorrect handling of lttng_enumerate_block_devices() return value of 0 (which means "ok"), causing do_lttng_statedump() to return early.

To causes missing "lttng_statedump_end" in traces where block device enumeration is available (CONFIG_KALLSYMS=y).

Thanks,

Mathieu

A message that is not in the archive (sent off-list or not yet received): [email protected]

#3 lore raw

----- Original Message -----

Quoted 6 lines

Hi Mathieu,

Thanks for the info, good to know!

However, we (in Trace Compass) don't actually make use of the "statedump_end" event. Should we?

Yes, you should.

Quoted 7 lines

For now, we only use the "lttng_statedump_process_state" events in our state model. The information contained in these events is applied to the state of the involved process. This state becomes valid starting at the timestamp of each statedump event.

What type of information does the "lttng_statedump_end" event implies?

There is some info about the state of processes active at trace start, or interrupt handlers active at trace start, which could be impractical to fetch by the statedump events per se. Rather, we use a rendez-vous which implies that a known kernel thread has been scheduled on each CPU before we issue the statedump end event. Having this quiescent state allows to fixup some unknown state that might still be unknown at statedump end.

I recommend you look at lttv state.c fix_process().

Thanks,

Mathieu

Signature · 3 lines
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com