changes since 1.0.30: v1.0.30...v1.0.32

notable:

• support for --ask-pass: interactive password input
• fixed PRIMARY KEY/UNIQUE KEY check

Many additions since 1.0.28. The ultimate breakdown is v1.0.28...v1.0.30

Notable:

• Fixed cut-over stall: when cut-over happened to start executing even as lag or otherwise a throttling condition also emerged, a race condition would make the cut-over stall until throttling reason went away. The fix is that cut-over does not wait, and at worst, rolls back.
• fix to cut-over retries -- it didn't throttle properly in between and did not cleanup properly in between.
• cut-over timeout on "unknown" stalls. An example to "unknown" was the throttling problem described above. I don't know what other unknowns there are, but cut-over is now generically resilient to them.
• support for --skip-foreign-key-checks - can save inspection time, when one is certain no FKs actually exist
• Made column-rename detection heuristic more robust (it could wrongly infer a rename by some unfortunate comment or column-name)
• Reading and reporting replication lag before waiting on initial replication event -- this adds visibility in seemingly-hanging operations where the replica is lagging in the first place
• changelog table uses same engine as original table
• on startup, verifying replication works all all the way up to the master.

Many additions since 1.0.21. The ultimate breakdown is v1.0.21...v1.0.28

Notable:

• support for DATETIME subsecond resolution
• user-commanded throttling implies complete silence, not even heartbeat is written
• fixed log_slave_updates check on --test-on-replica and --migrate-on-replica
• cleanly closing streamer connection (connection that imposes as a replica) upon completion
• on --exact-rowcount, --concurrent-rowcount now defaults true
• added validation to confirm table structure on master & replica is identical (yes, the counter case happened)
• correct (though suboptimal) behavior on enum being part of the migration key

Many additions since 1.0.20. The ultimate breakdown is v1.0.20...v1.0.21

Notable:

• support for --critical-load-interval-millis -- a 2nd threshold sample to allows for graceful recovery from sudden and immediate spike.
• better support for non-utf charsets; help requested.
• support for dropping foreign keys. Watch out! This will drop your foreign keys. That's what it does.
• Better support for tungsten. Read the story

Notable:

• Support for latin1 (but for other character sets, please see #228). utf8 and utf8md4 supported out of the box.
• named cut-over: a configurable protection mechanism to require table name when commanding postpone on interactive commands. --force-named-cut-over is introduced.

Other minor enhancements.

Notable:

• Support hooks: external processes invoked by gh-ost at specific points in execution
• Sub-second lag resolution: if you require your replicas to lag within subsecond threshold, this is now supported
• RDS: this release should work with RDS, see discussion -- we will formalize documentation once behavior is tested more
• Tungsten: supply --tungsten --assume-master-host=the.master.com
• master-master improved support: explictly specify the master via --allow-master-master --assume-master-host=the.master.com. See cheatsheet
• Improved status report (ETA shows estimate even while throttling)

Notable:

• --concurrent-rowcount (#194): with --exact-rowcount, issue the count(*) concurrently even as row copy takes place (row count gets updated throughout the migration)
• Using STRICT_ALL_TABLES for all operations, as safety net ()
• Fixed support for 5.5 (#184)
• Relaxed config file parser (#198)

Notable:

• Last fixes to time-size related issue (gh-ost is now fully UTC)
• Fixed column rename DML apply failure