Refresh MySQL config after a backup

[rafael]

Description

• This addresses the following issue: mysqlctld doesn't evaluate if my.cnf changed after taking a backup #2908
• It relies on already implemented method RefreshConfig().

Tests

• Tested locally.

[bbeaudreault]

This LGTM but we should probably also do on Restore?

[rafael]

Good call. Just dig into it and found that in that context it's calling ReinitConfig(), so I think it should be good.

However, now I'm not sure if I should call here ReinitConfig as well. When I was looking into this, both should work, but RefreshConfig seemed like a better fit. What do you think?

[bbeaudreault]

Ah I forgot. That's actually where I got the inspiration for RefreshConfig. Let me take a look.

[bbeaudreault]

I think we should stick with RefreshConfig. ReinitConfig exists so that it can additionally change the mysql serverId to avoid skipping transaction after a restore. In the case of taking a backup I think we want to keep the same serverId when it comes back up.

So we can leave Backup alone.

[bbeaudreault]

Test failures seem related to this change

[rafael]

@bbeaudreault - yeah looking into that right now.

[rafael]

@bbeaudreault - I think it should be good now. Fixed the test !

[bbeaudreault]

LGTM. cc @sougou

[sougou]

LGTM

[rafael]

While deploying this change to prod, we discovered an issue with this change. If you are using the following flag when starting mysqlctld: -mysqlctl_mycnf_template, you will have to pass it to the tablet as well. Otherwise you will run into this problem when running backups:

Unknown desc = TabletManager.Backup on xxx error: can't refresh mysqld config: Could not initConfig in xxx/my.cnfxxx: open /usr/local/config/mycnf/default.cnf: no such file or directory

I believe the reason is in this line. Function initConfig use by RefreshConfig() relies on that flag.

Besides documentation, I don't see any workaround for this problem.

@bbeaudreault, @sougou thoughts?

[sougou]

I feel like this needs to be fixed by a higher level design change. Until then, I'm ok with documenting it.

[bbeaudreault]

I also ran into this. I think possibly what should happen is mysqlctld should refresh the config before starting mysql, rather than vttablet refreshing the config before calling mysqlctld.

I havent had a chance to look at the code yet to see how hard that is. If it's too complicated or doesnt work as I think then documenting is fine for now.