#KairosDB (with embedded ScyllaDB)

Starting instructions

minikube

helm install helm/kairosdb
# endpoint will be a node port

aws

helm install -f helm/aws-dev.yaml ...
# endpoint is an internal AWS ELB

Features

• Minikube
• AWS environments
• Uses PVCs
• Goal is docker images don't need patching for config changes

Monitoring

• Can run nodetool on the actual nodes (as JMX is presently localhost only)
• various metrics in kairos itself (kairosdb.*) indicate health of TSDB

TODO

scylla

• get rid of the start-scylla script -- do config map and injection instead
  • also ensure that the main process (group) can catch a SIGTERM for gracefull shutdown
  • fix the background tasks method of start -- maybe two containers...
• Smarter sparse list for seed protocol (or just new a new provider)
• Evaluate inter-node compression settings
• Figure out why serviceName and statefull set name can't be different
• can we get scylla to listen on all ports (so localhost works for kairosdb)?
• there is some crosstalk when two running at once: WARN 2018-05-16 18:30:35,216 [shard 0] gossip - ClusterName mismatch from 172.17.0.7 braided-ibis-kairosdb!=winsome-mite-kairosdb

kairos

• consider flipping to MemoryQueueProcessor (or make this a help option)
• how to control memory usage (it may need jvm container hint)
• need to include https://github.com/kairosdb/kairos-healthcheck so we don't scan all metrics
• we have a bugfix for don't query TTL > 10year -- need to understand why we need it?
• inject my pod name as an init container instead of in wrapper.sh
• sleep delays in kairos are definitely not the best approach here
• need to change LoadBalancer to point only to our scylladb

benchmark

• create a read load