Re: problem w/ read caching..

[Kent Overstreet <koverstreet-hpIqsD4AKlfQT0dZR+AlfA@public.gmane.org>]

On Mon, Oct 01, 2012 at 10:26:43PM +0000, Brad Walker wrote:
> Kent Overstreet <koverstreet@...> writes:
> 
> 
> > I was just browsing around the code, and I bet I know what it is -
> > btree_insert_check_key() is failing because the btree node is full.
> > 
> > But, we should confirm this really is what's going on...  Can you apply
> > this patch and rerun to test my theory? See if the number of times the
> > printk fires lines up with the number of cache misses.
> > 
> 
> 
> I applied this change and I see a LOT of the messages.
> 
> And the rate seems to be increasing.

Sweet, we know what it is then.

So, like I mentioned this won't be an issue on any workload with mixed
read/writes, so if that's what your production workloads are then this
may not matter to you.

For warming up the cache, doing a few random writes (just enough that
you hit all the btree nodes - and there aren't many btree nodes, cat
internel/btree_nodes) will fix it.

A real fix for this shouldn't be too hard, but it's not exactly trivial
and it'll be a pain to test... not quite sure when I'll get to it, but
it would be good to have it fixed.