Getting by without SUPER privilege about gh-ost HOT 10 CLOSED

looking into this

from gh-ost.

The first blocker I see right away is that SUPER is required for STOP SLAVE and START SLAVE.

Why do we need these two in the first place? Because gh-ost may be asked to switch binlog_format to ROW. In which case it issues:

• set global binlog_format='ROW'
• set session binlog_format='ROW'

However if the replica was running with STATEMENT, merely changing to ROW means nothing; the replication must be STOPped and STARTed for the setting to take effect. gh-ost does not want to assume it was set up correctly before running; therefore it always issues a STOP & START.

We could add a flag saying --trust-me-this-is-rbr to skip the above. I don't like it, but it's possible.

from gh-ost.

It should be possible to pre-test for RBR and not require SUPER.

from gh-ost.

@SuperQ how can you you confirm without doubt the replica is using RBR? Please read again my comment.

from gh-ost.

To elaborate: you see a replica. It says it has binlog_format=ROW. Fact: it may replicate via SBR, because maybe someone set global binlog_format=ROW without restarting replication.

It sounds far fetched, but in fact is a classic case in --switch-to-rbr

from gh-ost.

#156 will close this Issue, and will make it possible to run gh-ost without SUPER privilege.

from gh-ost.

I'm happy to merge #156 , and see thoughts on SBR/RBR in comment: #156 (comment)

from gh-ost.

SUPER is now not required for normal migration, see #156 and release https://github.com/github/gh-ost/releases/tag/v1.0.10

It is still required in --test-on-replica, see discussion on #162 and on #163

from gh-ost.

Hooks have been merged, and remove the last barrier for SUPER. I will write more about this on next release, which is expected early next week, at which time this bug will be closed.

from gh-ost.

all done

from gh-ost.