We've observed a correlation between our cache rate decreasing and a spike in 
cache ready busy failures (proxy.process.cache.read_busy.failure), though I've 
yet to determine the cause for this.

What I really don't understand is why restarting resolves the issue for about 
24 hours?  I know the hash of directories is stored in memory…is there anything 
else that would be rebuilt on a restart related to caching?

From: Peter Walsh <[email protected]<mailto:[email protected]>>
Reply-To: 
"[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Date: Fri, 25 Jan 2013 23:51:59 -0800
To: "[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Subject: Issue with cache hit ratio degradation over time, restart resolves 
temporarily

Hello all,
Lately we have been experiencing an odd issue with our cluster running 3.2.

After about 24 hours we'll see a 10-20% drop in cache hit ratio, which will 
hold steady for several hours before dropping again another 10-20% in what 
looks like stair step.  Eventually, it will drop to almost 0%.

Whats *really* odd is that restarting the cluster temporarily resolves the 
issue, bringing the cache hit ratio back to normal levels for another day or so.

We've added more cache space but that has shown no change in behavior. Memory, 
disk space, and CPU utilization is normal.

The only difference in usage I've identified, is our use cases in which we use 
the Cache API in our plugin have increased.

Any thoughts anyone? Thanks in advance.



Reply via email to