dtrace discuss - Oct 2009 - Solution to problem : dynamic variable drops with non-empty dirty list

I read some of the previous posts on this forum, which discussed this particular
problem. I tried to use the solution that those posts suggested, but I am still
having a problem with my dtrace application running on a heavily loaded system.
I have the following lines in my D script,

#pragma D option aggsize=8m
#pragma D option bufsize=16m
#pragma D option dynvarsize=16m
#pragma D option aggrate=0
#pragma D option stackframes=MAX_STACK
#pragma D option cleanrate=50Hz

but still I am getting the following messages,

     Drop: 17458 dynamic variable drops
     Drop: 3426 dynamic variable drops with non-empty dirty list

and the amount data that my D script is reporting, is gradually dropping.

This means that the tuning that I am doing using the above #pragmas is not
sufficient to address this particular problem. Is this a known problem ? Is
there a workaround/fix to address this problem ?   Though my D script keeps
running,  it becomes useless after some time on such a heavily loaded system.
-- 
This message posted from opensolaris.org

On Thu, 2009-10-08 at 15:47 -0700, Krishnendu Sadhukhan
wrote:
> I read some of the previous posts on this forum, which discussed this
particular problem. I tried to use the solution that those posts suggested, but
I am still having a problem with my dtrace application running on a heavily
loaded system. I have the following lines in my D script,
> 
> #pragma D option aggsize=8m
> #pragma D option bufsize=16m
> #pragma D option dynvarsize=16m
> #pragma D option aggrate=0
> #pragma D option stackframes=MAX_STACK
> #pragma D option cleanrate=50Hz
> 
> but still I am getting the following messages,
> 
>      Drop: 17458 dynamic variable drops
>      Drop: 3426 dynamic variable drops with non-empty dirty list
> 
> and the amount data that my D script is reporting, is gradually dropping.
> 
> This means that the tuning that I am doing using the above #pragmas is not
sufficient to address this particular problem. Is this a known problem ? Is
there a workaround/fix to address this problem ?   Though my D script keeps
running,  it becomes useless after some time on such a heavily loaded system.

I had the same problem when tracing 48 loaded HD''s on a thumper and
what
helped was to use a ringbuffer.  You''ll want to have some trigger then
to dump the buffer or use ctrl-c.

#pragma D option bufpolicy=ring

This solution does not eliminate the problem - it just delays the occurrence of
the problem. I still see it happening, at a much later time though.
-- 
This message posted from opensolaris.org

Hey Krish,

You''re likely leaking dynamic variables. This can happen for example  
if you have self-> variables or associative arrays whose contents you  
never explicitly set to 0 to "free".

Adam

On Oct 8, 2009, at 3:47 PM, Krishnendu Sadhukhan wrote:
> I read some of the previous posts on this forum, which discussed  
> this particular problem. I tried to use the solution that those  
> posts suggested, but I am still having a problem with my dtrace  
> application running on a heavily loaded system. I have the following  
> lines in my D script,
>
> #pragma D option aggsize=8m
> #pragma D option bufsize=16m
> #pragma D option dynvarsize=16m
> #pragma D option aggrate=0
> #pragma D option stackframes=MAX_STACK
> #pragma D option cleanrate=50Hz
>
> but still I am getting the following messages,
>
>     Drop: 17458 dynamic variable drops
>     Drop: 3426 dynamic variable drops with non-empty dirty list
>
> and the amount data that my D script is reporting, is gradually  
> dropping.
>
> This means that the tuning that I am doing using the above #pragmas  
> is not sufficient to address this particular problem. Is this a  
> known problem ? Is there a workaround/fix to address this  
> problem ?   Though my D script keeps running,  it becomes useless  
> after some time on such a heavily loaded system.
> -- 
> This message posted from opensolaris.org
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org

--
Adam Leventhal, Fishworks                        http://blogs.sun.com/ahl