openark/orchestrator v2.1.0

GA release v2.1.0

on GitHub

Changes since last release: v2.0.3...v2.1.0

• Massive SQL related changes, supporting sqlite as backend (sqlite support is tested but sqllite as backend is not included in this release).
  • Backend schema: changes to index names. By running this release first time you will experience a once time rebuild of almost all backend tables
  • Queries: many queries simplified. Multi-table DMLs broken into parts. GROUP/GROUP_CONCAT simplified; some logic moved to app.
  • Otherwise operations are unchanged; many tests added
• Supporting MultiMatchBelowIndependently. This supposedly step-back from multi-server bucket optimization is made possible by binlog indexing, released a year ago, which alleviates the pain from re-scanning master binlogs. Right now, for backwards compatibility, this defaults false; but potentially in the future it will change to true. Very large installments, where number of replicas per server is within the many dozens or hundreds, will still benefit from the bucketing approach (MultiMatchBelowIndependently: false)
• Fix bug in identifying Percona Server GTID based replicaiton
• Support DiscoveryIgnoreReplicaHostnameFilters (thanks @sjmudd). With growing number of apps that pretend to be replicas (gh-ost, tungsten, others...) orchestrator falls into the trap of trying to discover a "server" where that "server" is really some app. This is an initial, simple approach to filter out such apps based on hostnames. We may iterate on that in the future.
• More failure analysis scenarios and detection
• More...
• Also thanks @samveen (working on build script init script)
• Thanks to the many collaborators who report issues and discuss potential fixes