Morgan Tocker

There is a possible race in acquiring locks the connection pool to threads +1: - The checksum acquires a LOCKS TABLES .. READ to start the RR connections for checksumming...

We had a failure scenario where the pod was terminated but with a grace period waiting for the migration to complete. The networking on the pod had been shutdown but...

The main goal of setting a target chunk time is to limit the maximum time row locks are held during chunk copies. Since chunk copies are performed in auto_commit the...

It is expected that inserting into a larger table can slow down. This is particularly true if the table has unique indexes or is using Aurora (can't use insert buffer)....

Usually the checksum phase takes about 10-20% of the time that the copier phase does. However, on some very IO bound cases I have noticed it take significantly longer. Much...

Algorithmically we can support resuming from a source (old term: master) node failure. But we don't test it and thus don't advertise it. We should be able to do this....

For spirit to survive aurora cutover, it should enable this option: https://github.com/go-sql-driver/mysql#rejectreadonly Otherwise, it can end up with a scenario where it reconnects too quickly on failover and keeps using...

Currently we auto-select the `rds-ca-2019-root.pem` certificate if the hostname matches an RDS hostname. There's 2 problems with this: 1. Users may select a newer RDS certificate. 2. Users (outside of...

See if we can remove the canal library. We have quite a few workarounds around it. It might be better to use go-mysql directly instead.

This is a regression from https://github.com/cashapp/spirit/pull/332 The binlog event it is waiting on looks to be a GTID event: ``` mysql> SHOW BINLOG EVENTS IN 'mysql-bin-changelog.002567' FROM 82282378 LIMIT 10;...