21

I've configured my Varnish cache to use a cache size of 512 MB, using:

varnishd ... -s file,/var/cache/varnish.cache,512M

Questions:

  • How do I monitor if 512 MB is too small, too big or okay?
  • How do I get the current usage size of the cache?
  • How do I see how often objects are removed from the cache due to the cache being full?
knorv
  • 1,789
  • 6
  • 19
  • 29

3 Answers3

19

I found the solution:

You can monitor how much of the maximum cache size (512 MB in this case) that Varnish has allocated by running varnishstat. Then look for the output lines "bytes allocated" and "bytes free".

The following relation holds:

Command line configuration of max size = [bytes allocated] + [bytes free]

Depending on the size of "bytes free" you can fine-tune your cache size to find a proper level.

knorv
  • 1,789
  • 6
  • 19
  • 29
16

Watch the n_lru_nuked counter. Each time it increases, Varnish throws something out of the cache because it is running low on storage.

Ideally, you wouldn't want the LRU counter to increase at all since kicking stuff out of cache usually means it must be re-fetched, but if your tail is long, LRU cannot be avoided.

greg0ire
  • 316
  • 1
  • 6
  • 26
perbu
  • 431
  • 3
  • 4
  • what does `if your tail is long` mean? – cherouvim Mar 22 '16 at 07:20
  • I think in this context it means if you have a lot of rarely-used items that end up in cache, taking up all the cache memory. It's a reference to this concept: https://en.wikipedia.org/wiki/Long_tail – cam8001 Jun 07 '16 at 12:09
1

Unless you expect all your data to fit in the cache, I'd look at the ratio of cache hits to cache misses. You get diminishing returns. Doubling the cache will probably halve your miss-rate, which is a big win when half your requests are uncached, not so much when 90% are.

eas
  • 268
  • 1
  • 2
  • 7