steve-

great patch!  good contribution  (although Gap, Inc. might force you to 
change the name :P)

do you have any idea how the RRDs got corrupted?
-- 
matt

Today, Steven Wagner wrote forth saying...

> This patch was made against 2.5.2 - I haven't tested it on the latest CVS 
> version.
> 
> What you get:
> 
> *  gmetad starts passing a timestamp value down to the individual RRD 
> updating functions.
> *  If a timestamp isn't provided, it generates one using your platform's 
> POSIX-compliant time() function.
> 
> This patch may make gmetad unstable, which is why I'm posting it here - I'd 
> like to see what other people's experiences are with this patch.  I've 
> tested it on my Solaris 9 (yeah, I upgraded...) front-end and it's working 
> beautifully so far.
> 
> Last week I had all sorts of gmetad stability problems which turned out to 
> be related to having a number of truncated/corrupted RRD files.  gmetad 
> doesn't currently make allowances for this - if an RRD file exists, it 
> assumes it's working properly.  If it doesn't exist, it tries to create it. 
>   That's it.
> 
> I went through and deleted the truncated files by hand, but a more elegant 
> solution would clearly be to make gmetad a little more RRD-aware in the 
> long run.  I'm not sure exactly how this contributed to my gmetad crashing 
> every 40 minutes (perhaps there's some sort of memleak associated with an 
> RRD*() error condition?), but the problem popped up when I moved RRD 
> directories and accidentally left some truncated RRDs around, and 
> disappeared after I removed 'em.
> 
> So stay alert, trust no one, keep your laser handy, et cetera.
> 
> 


Reply via email to