Use memcache in local

[jml]

Updates the local deployment to use memcached. Also provides graphs for memcached.

Part of the fix for #442. Depends on weaveworks/scope#1602.

[tomwilkie]

• Can we have the memcache bits in the dashboard come before s3?
• Don't bother with server side QPS and hit ratio.
• Put latency on the same row as QPS pls.
• Can we get some mearue for memory pressure in memcache? I guess we'd see this as a lower hit rate, so not super important.

[tomwilkie]

Is there any limit to the size of these memcaches?

[jml]

All done. Added memory capacity graph to memcache, showing current bytes ÷ total bytes allowed.

Added explicit option to rc file for maximum memory size. Default is 64MB.

PTAL.

[tomwilkie]

Have run this locally, and dashboard is good.

However, I have a few concerns (which shouldn't block this PR):

• I still seem to be doing S3 gets. I was expecting this to mean we do basically no gets from S3?
• Similarly, why is the hit rate not 100%
• At least locally, the memcache get latency is quite high (3-400ms at 99th). I wasn't expecting this.

I suggest we get this merged and then investigate this some more in the dev environment.