Re: [Xen-devel] About lost record of xentrace, how to avoid lost record?

[George Dunlap <George.Dunlap@xxxxxxxxxxxxx>]

On Wed, Nov 26, 2014 at 7:21 AM, hanyandong <hanyandong@xxxxxxxxx> wrote:
> hi all,
> I found xentrace will lost record if there are too many  event to trace.
> But every event is important to me, so I want to trace all of them, not lost
> one.
> what  could I do to achieve this goal ?
>
> If it need to modify the source code of xentrace, I will do it. But will
> anyone give me some instructions?

Fundamentally, you need to get the data from Xen out onto the disk.  A
few things to try:

* Make the in-Xen memory buffers larger, using the -S option.

This will help if the problem is just that your Xen traces are too
"bursty" or if dom0 doesn't get run often enough.  But if Xen is
simply generating traces faster than you can write to stable storage,
it's not going to work.

Note also that once you've run xentrace once, you can no longer resize
the buffers; you have to reboot your host.

* Reduce the volume of your traces.  You can use the "-e" option to
narrow down what kind of trace records you get -- make sure that you
only trace the records you want.  This will reduce the number of lost
records.

* Increase the speed of your disk.  SSD or RAID are of course options;
another would be to use a tmpfs or a ramdisk (although I haven't tried
this).

* If you only need data from a few seconds in the past, and you can
detect the condition you're trying to look at, you can use the "-M"
option.  This will copy trace records into a circular buffer in
xentrace's memory, and dump it to a file when xentrace exits.  This is
useful for example if you're trying to debug a crash, and you can tell
when the crash happens and send xentrace a SIGINT.

Good luck. :-)

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel