[MCOL-2073] Document how to rebuild an inconsistent UM in a multi-UM environment Created: 2019-01-08  Updated: 2023-07-02  Resolved: 2023-07-02

Status: Closed
Project: MariaDB ColumnStore
Component/s: N/A
Affects Version/s: None
Fix Version/s: Icebox

Type: Task Priority: Major
Reporter: Geoff Montee (Inactive) Assignee: Unassigned
Resolution: Won't Do Votes: 2
Labels: None


 Description   

In a multi-UM environment, it is possible for one of the UMs to become inconsistent with the other. The most likely cause is replication issues. We should document how to rebuild a "bad" UM from a "good" UM. It is currently undocumented:

https://mariadb.com/kb/en/library/columnstore-multiple-user-module-guide/

One of the most common ways to rebuild a MariaDB replication slave is to use Mariabackup:

https://mariadb.com/kb/en/library/setting-up-a-replication-slave-with-mariabackup/

But as far as I know, Mariabackup does not support MariaDB ColumnStore at all at the moment.



 Comments   
Comment by Geoff Montee (Inactive) [ 2019-01-08 ]

Feedback from hill:

This procedure assumes using ssh and ssh is setup between pm1 to um2 and um1 to um2.
Also that Replication is enabled. Also does require download time with the stopsystem. Whole procedure takes about 10 minutes.

Here is the procedure running these commands from pm1:

  • run the following with mcsadmin
    stopsystem y
    removemodule um2 y
  • delete the columnstore directories from um2 server
    rm -rf /usr/local/mariadb
  • run the following with mcsadmin
    addmodule um2 'hostname'
    alterSystem-enableModule um2 y
    startsystem
Comment by John Dutchover [ 2019-01-08 ]

David,
I have performed this procedure (all commands as root user).

Comment by John Dutchover [ 2019-01-08 ]

MariaDB ColumnStore Replication Feature is enabled, however replication does not appear to be working. For starters, no replication user was created by using this procedure. Please advise.

Comment by David Hill (Inactive) [ 2019-01-08 ]

Try enabling again through the mcsadmin on pm1. Lets see if it reports passing

  1. mcadmin enablemysqlrep

You can check the logs on um1 and um2 to see if its successfully after its done.
So I assume ssh is setup between um1 and um2, as mentioned before

Comment by John Dutchover [ 2019-01-08 ]

As root, I needed to disable and re-enable (from PM1):

  1. mcsadmin disableMySQLReplication
  2. mcsadmin enableMySQLReplication

At this time it appears that master and slave are in sync. Will need to review with @Martin Adamec tomorrow.

Comment by Todd Stoffel (Inactive) [ 2023-07-02 ]

The "create date" on this ticket is pre-convergence with MariaDB server. If the issue still exists in a modern version of the engine/plugin please submit a new ticket.

Generated at Thu Feb 08 02:33:31 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.