notes.txt

# Elephants on Automatc

Introduction: It's time for State
- we've automated and scaled the stateless applications
- now we have statefulset
- time to automate stateful apps
- still some work to be done

Bot Pattern
- scheduler problems
  - too much application-specific logic
  - telephone lag
- instead, add code to control stateful nodes
  from within
- maintains
  initialization
  group members
  failover
  monitoring
- mention ContainerPilot?

Helm
- deploy app with lots of parameters
- coordinate multi-container application with dependencies
- really just parameterized kubernetes files with registry

Oleksii

Patroni
- RAFT is the fabric that holds together your PostgreSQL containers
- External consistent key-value store instead of a built-in one:
  advantages:
   - PostgreSQL failure doesn't break the consensus
   - use well-tested RAFT implementations (i.e. Etcd or Consul)
  disadvantages:
   - one needs to operate extra services
Features:
- synchronous vs asynchronous replication (consistency vs availability)
- some parameters need to be kept in sync on all cluster nodes (max_connections)

CitusDB
- sharded postgres
- query node, plus shard nodes
- shards with duplication
- automatically distribute queries

giant robot elephant from The Nantes Project
elephant helm by Blank Williams
robot elephant by Robert Chew
robot with elephant by Mark F Stosiak
stubborn elephant by Kolja Wilcke @01k





A long strange journey
- yak-shaving
- wanted to automate HA postgres
- containers solved a number of issues
- statefulness gulf
- building the foundation
- got involved with PetSet

Containerize Vs. Not?
- TL;DR: this is an "in 6 months" solution
- why to containerize
  - easy installation
  - encapsulation of service
  - force good DBA habits
  - orchestration
  - easy testing
- benefits of orchestration
  - manage multiple Postgres instances
  - auto-vivification
  - managing connections and pools
  - HA/DR
  - integrate with rest of stack
  - self-service for developers
  - use GCE
- drawbacks
  - alphas
    - waiting for fixes, features
  - extra setup (unless you use GCE)
  - new ways of admin/config/monitoring

The Prototypes
- why they're only prototypes
- containers
  - autopilot pattern
  - container considerations
    - log to stdout/err
    - install bootstrap & pvs
      - required parts:
        - config
        - hba
        - initial users & pwds
  - patroni container
    - guided tour
  - citusdb
    - guided tour
- kubernetes config
  - explanation of statefulset parts
  - patroni
    - etcd requirement
    - demo of deployment
    - PV vs. emphemeral
  - citusdb
    - demo of deployment
    - using citusdb

Productionizing
- list of needed items:
  - PVs
  - Vault
  - sidecar containers
    - backup
    - pooling
    - etcd
    - db management
    - pgadmin
- features on the waitlist
  - union configmap
  - Vault plugin
  - federated StatefulSets
  - issues with PVs
  - affinity
- more containerizing
  - Postgres Pro MM-Sync
  - BDR?

Q?