[ixpmanager] SFLOW Under Reporting?

[Ian Chilton]

Hi,

On 2023-06-27 10:49, Nick Hilliard (INEX) wrote:
> That sounds a lot better. In terms of approaches, I was considering a 
> straightforward
> fork, but probably the threads approach is simpler in this case.

Yes, it turns out to be a really simple change and seems to be working 
totally fine!


> change to the default timer interval from 60s to 300s

Is there some rational for that other than protecting it ever starting 
to take longer than 60s to do the flush?

It surely makes it less accurate and you risk overflowing something in 
the future by the counters getting too big as they are not flushed as 
often?


> also to make a copy of $matrix before passing to the thread, to ensure 
> that concurrency problems are ruled out.

When I was doing some testing with Python I had this - so it copied the 
variables to use before zeroing... but actually in this case I don' 
tthink it's a problem as without using threads::shared and marking a 
variable as shared, I think everything passed to threads are passed by 
value rather than reference anyway?

That certainly seems to be the case as passing process_rrd to a thread 
like this means that the matrix_init happens straight afterwards and 
it's not writing zeros to the RRD files.

Ian