Old: 100
API: 300

This is completed

I have renamed the table entirely on s5 dewiki

Old weight 300

@Ladsgroup could you review this request? Thanks!
@cmelo a quick glance shows that you are not having any single PK on any of the tables, is that correct? Or should we assume cew_id cewa_id cewuba_id would be PK for each table?
If you could provide the full "show create table" for each table, that would be clearer to review.

This was pushed to the repo

In T365568#9857942, @Marostegui wrote:

Done this for now:

cumin2024@db1185.eqiad.wmnet[dewiki]> rename table flaggedpage_pending to T365568_flaggedpage_pending;
Query OK, 0 rows affected (0.003 sec)

cumin2024@db1185.eqiad.wmnet[dewiki]>

Thank you so much Scott.
Arnaud, good idea! Let's review things on that other task.

@cmelo is this task meant to be for DBAS? There is no mention to the storage you want to use, is this request for a MariaDB database in our misc sections? If so please follow: https://wikitech.wikimedia.org/wiki/MariaDB#Database_creation_template

This looks good to me - thank you.
@ABran-WMF as a person who's not dealt with parsercache much (yet) - can you check https://wikitech.wikimedia.org/wiki/MariaDB/troubleshooting and see if with that documentation you'd be able to swap a dead parsercache host or you are missing some bits.

I will get this done on Tuesday next week

This is done

Thank you Amir - this is great to have! Thanks for working on it!

This was a test

DNS merged

@brouberol do you use replication?

show slave status\G

Done

Yes, I think we can. Let's do it in a controlled way just in case. Let's drop all codfw and we can check in a few hours the event_log and make sure events are still being killed, if so, we can drop eqiad.

In T365098#9861924, @Ladsgroup wrote:

So far looks okay but we need to set candidate masters in puppet as comments. None has it.

The old master is being repooled now automatically. Closing this.

In T366552#9858819, @Marostegui wrote:

Leaving this ticket opened as a reminder to move the backup source to the new master

Yes, the ops database is where the events are installed and that is not being touched.

Yes, I can see events being killed on hosts which don't have ops user.

cumin2024@db1163.eqiad.wmnet[ops]> select stamp from event_log order by stamp desc limit 1;
+---------------------+
| stamp               |
+---------------------+
| 2024-06-04 04:54:33 |
+---------------------+
1 row in set (0.001 sec)

@brouberol first you need to stop mariadb, then remove wmf-mariadb104 and then install wmf-mariadb106 via puppet (or however it is handling from your end).
Once done, start mariadb again and don't forget to run: mysql_upgrade

@Ladsgroup can you give me a lost of hosts without that user?

If I remember correctly the ops user was migrated to root (I haven't checked but I recall something like that). The database is still ops but I believe the definer was migrated to root. Otherwise, if the user is missing in so many hosts, that's definitely something we need to look at.

Leaving this ticket opened as a reminder to move the backup source to the new master

Topology is moved but db2141:3311 remains unchanged as the backups are running. Will move it once it is finished and in sync

All done

Old weight 400

Done

Every production database has been migrated to 10.6 - pending only some backup sources tracked at https://phabricator.wikimedia.org/T360751

Done this for now:

cumin2024@db1185.eqiad.wmnet[dewiki]> rename table flaggedpage_pending to T365568_flaggedpage_pending;
Query OK, 0 rows affected (0.003 sec)

In T365356#9850706, @Scott_French wrote:

Copying over from T362786#9806963, spots to update include:

1. (definitely) https://wikitech.wikimedia.org/wiki/MariaDB/troubleshooting - update instructions for depooling a pc instance
2. (definitely) https://wikitech.wikimedia.org/wiki/Dbctl - clarify "add a section" instructions to include the pc section case
3. (maybe) https://wikitech.wikimedia.org/wiki/Parser_cache - note that the configuration uses etcd as a source of truth in production

Let's give it two more days just in case.

@Ladsgroup is this good to go?

Host back in sync with the master, I am going remove downtime and start repooling.