Skip to content

Latest commit

 

History

History
42 lines (24 loc) · 2.15 KB

what-if.md

File metadata and controls

42 lines (24 loc) · 2.15 KB

What if?

Technical questions and answers. This document will be updated as we go

What if I'm using Statement Based Replication?

You can still migrate tables with gh-ost. We do that. What you will need is a replica configured with:

  • log_bin
  • log_slave_updates
  • binlog_format=ROW

Thus, the replica will transform the master's SBR binlogs into RBR binlogs. gh-ost is happy to read the binary logs from the replica. Read more

What if gh-ost crashes halfway through, or I kill it?

Unlike trigger-based solutions, there's nothing urgent to clean up in the event gh-ost bails out or gets killed. There are the two tables creates by gh-ost:

  • The ghost table: _yourtablename_gho
  • The changelog table: _yourtablename_ghc

You may instruct gh-ost to drop these tables upon startup; or better yet, you drop them.

What if the cut-over (table switch) is unable to proceed due to locks/timeout?

There is a lock_wait_timeout explicitly associated with the cut-over operation. If your table suddenly suffers from a long running query, the cut-over (involving LOCK and RENAME statements) may be unable to proceed. There's a finite number of retries, and if none of these succeeds, gh-ost bails out.

What if the migration is causing a high load on my master?

This is where gh-ost shines. There is no need to kill it as you may be used to with other tools. You can reconfigure gh-ost on the fly to be nicer.

You're always able to actively begin throttling. Just touch the throttle-file or echo throttle into gh-ost. Otherwise, reconfigure your max-load, the nice-ratio, the throttle-query to gain better thresholds that would suit your needs.

What if my replicas don't use binary logs?

If the master is running Row Based Replication (RBR) - point gh-ost to the master, and specify --allow-on-master. See cheatsheets

If the master is running Statement Based Replication (SBR) - you have no alternative but to reconfigure a replica with:

  • log_bin
  • log_slave_updates
  • binlog_format=ROW