Statistics dumps.

Erik Rozendaal erik at NLnetLabs.nl
Thu Jul 28 14:38:07 UTC 2005


Måns Nilsson wrote:
> However, through a combination of deficiencies in their collection system
> (outside my jurisdiction) and strange deviations in dump intervals, I'm
> getting inconsistent data and a very unhappy customer. The dump time is not
> at a steady 300 sec cycle, but deviates at strange intervals, and way too
> much at times. Given the non-real-time characteristics of Unices, I'm
> prepared to accept 10-20 second offsets, but not values ranging from 20% to
> 1400%. What is going on? 

NSD also dumps statistics when the process exits. So the shorter dump 
interval can be explained by database reloads (which causes NSD to 
restart itself).

The two big peaks in the green line are surprising. I'm not sure what 
could cause this. It looks like at some point the periodic SIGALRM that 
is used to dump statistics is lost, and dumping is not resumed until an 
NSD reload happens...

> A graphical representation of the data can be found at: 
> 
> <http://www.noc.kth.se/~mansaxel/wdel.png>
> 
> I notice two NSD processes running. Are they writing concurrently to the
> log? If so; is there any way to differentiate between them? 

Only the child process dumps statistics, since the parent process isn't 
answering any queries. If you run with multiple child servers (using the 
-N option) there is currently no way to distinguish the servers. Using 
syslog instead of a log file in this case.

Erik



More information about the nsd-users mailing list