Changes since last release: https://github.com/github/orchestrator/compare/v2.1.0...v2.1.2

Notable:

• structured, audited recovery steps:
  • each recovery has audited steps, identified by UID
  • UID presented in post failover hooks
  • underlying table: topology_recovery_steps
  • presented in web interface
• recovery speed improvements:
  • faster, concurrent execution of postponed functions
  • yet more optimization in prefering DC-local operations
  • incremental sleep following START SLAVE reduces wasted sleep time
  • more postponed functions for unurgent operations
  • recognizing hosts with discovery latencies, pushing them to postponed recovery
• graceful-master-takeover attempts setting replication credentials on demoted master (doable if master_info_repository=TABLE)
• FailMasterPromotionIfSQLThreadNotUpToDate config (bool, default false), when true and master fails while all replicas are lagging, heal topology but do not proceed with post-failover hooks, and in fact consider the failover as failed.
• Many metrics added by @sjmudd
• Fixes and improvements to GTID master recoveries
• binlog_row_image collected and made visible
• TravisCI now tests all PRs
• more...