Re: occasional "Warning: lost track of 460840 pulled message(s)"

[Uwe Kleine-K. <uwe...@kl...>]

On 11/9/21 12:45, Oswald Buddenhagen wrote:
> On Mon, Nov 08, 2021 at 10:53:00PM +0100, Uwe Kleine-König wrote:
>> A strange thing is that a run with -V works (unless it does not 
>> because the tunnel breaks), while with -D it reproducible fails :-\
>>
>> A typical ending then is:
>>
> my guess is that the server just disconnects, because things took too 
> long, though it seems slightly weird that enabling debugging would make 
> a difference.
> 
> we already know that the process actually consumes lots of cpu time.
> make a debug build of mbsync, and do some profiling ... though i'm not 
> sure how to do that best - the go-to recommentation would be "use 
> valgrind --tool=callgrind", but that would take *ages*, which might 
> distort the result (though it might not matter). "perf" (which is 
> kernel-based, afaik) might work better, but i never played with it.

My approach would be

	strace -f -tt -e decode-fds=path,socket,dev,pidfd -b execve

That should at least make it obvious, why mbsync terminates, doesn't it?

Best regards
Uwe