Notes from Containers & Databases session at ContainerDays Austin

containers_and_dbs.txt

containers & databases
======================

databases
---------

* postgres
* mysql & mariadb
* oracle
* redis
* cassandra & riak
* mongodb
* hbase & object stores
* sqlite3
* elastic search
* kafka

in containers?
--------------

* percona + galera, elastic search in swarm
* mysql + docker
* elastic search, postgres, mongo, docker + rancher


problems with dbs in containers
-------------------------------

* getting database to run in foregrond PID 1
* problems with logging (no stdout on mysql)
* storage affinity to nodes
  (want data to be local)
* failover, failback & services
* shard + node affinity
* vendor support
* hardware performance
* difficulty of migrating containers (killing nodes)
* database configuration
* lifecycle management
* container identity
* double-writing storage
* backup & restore

solutions
---------

PID1: mysql doesn't want to run in foreground,
  ran under runit with mysqld-safe
  not an issue for postgres

logging: only logs to files (mysql)
  give it different filehandles, symlinks for mysql.err
  logging to a volume?  yes, for big logs, log digest, splunk

storage & database containers:
  * manually deploying to nodes
  * swarm affinity rules
  * kubernetes petset: local, persistent, ephemeral
  * flocker
backup & restore:
  database tools
identity and sharding
  * PetSet
  * flynn& patroni
  * confd + etcd to config files
database configuration
  * confd
  * patroni etc.
lifecycle management
  * coordinate via etcd
  * rolling update from containers
  * pg_upgrade container
performance
  * not really different
  * except where it forces you onto network storage
  * minor latency due to network overlays

benefits
--------

* show that you can run legacy applications in a container
* deliver software better, including the database
* HA gets easier: STONITH, auto-vivication, encapsulation
* manage whole app stack in one place

other notes
-----------

look at the
autopilot pattern