Re: Help with trace-cmd/ftrace recording process ID information
From: Will Hawkins
Date: Mon Jul 17 2017 - 18:31:20 EST
On Mon, Jul 17, 2017 at 4:17 PM, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> On Mon, 17 Jul 2017 16:06:37 -0400
> Will Hawkins <hawkinsw@xxxxxxxxxxxxx> wrote:
>
>
>> This seems to be the problem:
>>
>> On the "good" system, that file is up-to-date with cached PIDs and
>> comms. On the bad host, there are no cached entries from any of the
>> traces that I've run.
>>
>> Because these are running old kernels, there is no saved_cmdlines_size
>> knob to turn. Do you have any idea why the saved_cmdlines would not be
>> getting updated appropriately on the "bad" host? I know this is not
>> ideal, but I can try to reboot that host and see if something is
>
> Yeah, a reboot may work.
>
>> simply wedged. The system has been online for almost a year, so it's
>> possible that something has gone wrong.
>>
>> Any help you can offer would be great! Thank you, again, for your response!
>
> The recording of command lines only happens when tracing is done, and
> there were a few bugs with the older kernels that caused it to either
> stop and never start again, or to simply just miss a bunch of recording.
>
> It may be that it stopped and never started again, so you will only
> have a stale file.
This appears to have been the problem! I did a reboot and everything
is back to normal.
Is there a way to poke at the tracing infrastructure in the kernel to
get it to restart process recording? I would feel more comfortable
with a solution like that instead of rebooting, obviously. The 'ol
Windows "solution" makes me queazy :-)
Thanks again for your quick responses. I hope that I can repay you at
some point by contributing code to the great tools you've built!
Will
>
> -- Steve