[bitnami/mysql] Feat mysqldump master database

[michel-silva]

Description of the change

Implement mysql dump to cases of missing binary files

Benefits

The dump will be executed before replication start. If binary files is missing will not broken replication

Possible drawbacks

The master server must be used only this process

Applicable issues

#54028

[carrodher]

Thank you for initiating this pull request. We appreciate your effort. Just a friendly reminder that it's important to sign your commits. Adding your signature certifies that you either authored the patch or have the necessary rights to contribute to the changes. You can find detailed information on how to do this in the “Sign your work” section of our contributing guidelines.

In the same way, could you please check the markdown issue reported by the linter?

Feel free to reach out if you have any questions or need assistance with the signing process.

[michel-silva]

@carrodher Thank you for your fast reply.
I've resolved the lint messages and removed my commit that doesn't have the signature.

Could you check if it's all right? Let me know if I could make something to merge

[dgomezleon]

Thanks for contributing. I added some suggestions.

[michel-silva]

@dgomezleon Thank you for your review. I made the changes, could you check again?

[dgomezleon]

Thanks for applying the changes. I detected a couple of minor typos. Could you please take a look?

[michel-silva]

@dgomezleon Thank you for the new review. The changes have been applied.

[dgomezleon]

Thanks again. Could you please check my last question?

[dgomezleon]

I have one last question about this. This would execute RESET MASTER for every slave. Is this necessary?
Note that RESET MASTER is already executed in mysql_initialize() . Could you please confirm?

[michel-silva]

Hi @dgomezleon.
Sorry about my delay response.. I was on vacation and I returned in the last days.
Your question is very interesting and makes me think about how replication works. So let me explain my point of view.

I have a master setup and a slave was broken because of a crash on the slave server.. So, I make a reinstallation of the new slave server and when I up the MySQL slave server this error occurs and this error is because the Mysql master server is missing some binary files. To fix it, I could make a manual mysql dump on master and update the slave, but to do it I must login in recovery mode on MySQL slave because the mysql.user could not be transferred by binary files and we do not have any valid user to login. Analyzing this scenario, I make this approach to make an auto mysql dump on master and import to a slave database.

Analyzing your question about the mysql_initialize(), this is only exec on the first up of the container. In my scenario, the reset will not be executed and the mysql master binary files are not modified. To resolve this, I could manually get mysql dump on the master server, make the reset master and run the mysql dump in slave.

So, when the MySQL master server doesn't have all the binary files, for each slave that is added to the server the reset master will be necessary for the slave to start replicating at the point of the master server stop.

One important point that your question made me think about is: that is very important that all slave servers have the last data present in the master database, when we restart replication after reset master the slave has the same data and does not miss any data. So, I think is important to add one alert to make sure that all slaves have all data updated, before start the setup the new slave.

I don't know if it explains my point of view. But I'm open to having some discussion about this and making the better to the package.

[dgomezleon]

It makes sense to me. Thanks for the detailed explanation.

[github-actions]

This Pull Request has been automatically marked as "stale" because it has not had recent activity (for 15 days). It will be closed if no further activity occurs. Thank you for your contribution.

[mostafa-rz]

@michel-silva Thanks for your PR. I have this problem and your feature will help me a lot. I have two questions:

1. Why you are not using mysqldump --all-databases --master-data instead of searching for the tables? This is the way mysql suggests on their website. (Please look at 17.1.2.5.1 Creating a Data Snapshot Using mysqldump)
2. With your current solution the user databases on the master are not copied therefore the slave is not the same as the master. Is this intentional?

[mostafa-rz]

There is also another problem. If I increase the number of slaves, the previous slaves will lose the connection to the master because of the RESET MASTER flag in this line. It is also mentioned in mysql documentation .

RESET MASTER is not intended to be used while any replicas are running. The behavior of RESET MASTER when used while replicas are running is undefined (and thus unsupported), whereas PURGE BINARY LOGS may be safely used while replicas are running.

Step to reproduce:

1. Start with 1 slave docker compose -f docker-compose-replication.yml up
2. Increase number of slaves docker-compose -f docker-compose-replication.yml up --detach --scale mysql-master=1 --scale mysql-slave=3
3. Create a test database in master
4. The test database shown in slave-2 but not in slave-1

[michel-silva]

@mostafa-rz Sorry about my delay response. I will analyse your points and make some tests. Thank you for your valuation.

[michel-silva]

@mostafa-rz this PR fix the issue.
#55126