Add Redis job

[philippthun]

• Redis will be used for rate limiting and (at a later point in time) also for metrics.
• The monit process will only be started if either the Puma webserver is used or cc.experimental.use_redis is set to true (defaults to false).
• If Redis is used, the CC config redis.socket is set; this switches the rate limit implementation from using in-memory counters to Redis.

See also cloudfoundry/cloud_controller_ng#3290.

• I have viewed signed and have submitted the Contributor License Agreement

• I have made this pull request to the develop branch

• I have run CF Acceptance Tests on bosh lite

[philippthun]

I've benchmarked the different rate limiting implementations (with the helper script from this PR] and got the following results:

• The time cost for the in-memory implementation is ~0.004ms per rate limit event¹.

• For the Redis-based implementation, the time cost is ~0.1ms per rate limit event.

When rate limiting 10K users (all kinds of rate limits enabled and applied), there are 30K keys stored in Redis and the process consumes ~10MB. For 100K users (definitely an extreme scenario!), there are 300K keys and ~44MB memory is used.

Footnotes

1. ExpiringRequestCounter:increment, ConcurrentRequestCounter:try_increment?, ConcurrentRequestCounter:decrement ↩

[philippthun]

With this change we set vm.overcommit_memory=1 on VM level. As there are memory thresholds (e.g. restart_if_above_mb) for CC that can/should be configured, the risk should be low...