…tion-percent is a mutable config (#236)

…dent

…rints, removed unwanted reset of pointers

* Overload CPU reading metric

* fix

* fix

update insufficient replicas metrics to separate severity

* update overload-ignore-list to be use subnets as well

* get ip directly from conn

* add getIPV4 impl

* ntohl

* don't shift at the end of the loop

* update for more efficient impl and add ipv6 support

* array of arrays

* one combined array

* have to reassign output of sdscat

* include mask in operator<

* replace raw numbers with constants

* add log when we ignore client for loadshedding

…243)

* add overload-protect-tenacity to configure how many clients to evict

* need to divide tenacity config by 100 to get percentage

* TLS cname loadshedding ignore list

…connecting clients (#245)

#246)

* handle nodes in handshake state in statsd module cluster nodes metrics

* handle all possible outputs of cluster nodes

* Update src/modules/keydb_modstatsd/modmain.cpp

Co-authored-by: Sergey Kolosov <[email protected]>

* make error plural

Co-authored-by: Sergey Kolosov <[email protected]>

* build redisbloom in dockerfile

* add separate step for redisbloom

* end RUN properly

* can't verify with github?

* just needed ca-cert

* don't timeout replication when loading rdb

* fix bracket placement

https://jira.sc-corp.net/browse/CACHE-1543

Log hit rate metric separately because M3DB caps values at 360 million so dividing monotonically increasing `keyspace_hit/miss` values is meaningless with any significant amount of traffic. Multiplying by one million since M3DB only uses integer values.

We discovered that full sync loops were happening because after the replica loads the RDB file, the repl backlog which the primary sends over can sometimes exceed int32 max, causing an SSL error due to integer overflow:

```
Error writing to client: error:140D010F:SSL routines:SSL_write:bad length
```

After fixing that error by capping the  and testing in staging, we still observed errors because the primary attempts to send a massive (multi-GB) payload in one event loop, which on the staging instances takes over one minute causing the primary to fail health checks and restart. To address this we cap the writes sent per event loop to a configurable amount, similar to `NET_MAX_WRITES_PER_EVENT` in the normal case.

Tweak hit rate metric to get delta instead of aggregate

… in order

…too soon