[MDEV-14534] Tokudb stopped working after automatic update Created: 2017-11-29  Updated: 2017-11-29  Resolved: 2017-11-29

Status: Closed
Project: MariaDB Server
Component/s: Packaging, Storage Engine - TokuDB
Affects Version/s: 10.2.11
Fix Version/s: 10.2.11

Type: Bug Priority: Blocker
Reporter: Philip orleans Assignee: Unassigned
Resolution: Fixed Votes: 0
Labels: None
Environment:

linux Centos 7

Issue Links:
Relates
relates to MDEV-14524 TokuDB is unable to be built on Linux Closed

 Description   

I lost 8 servers last night after an atomatic update. Tokudb stopped working.
If the plugin needed an update, then it should have been updated prior to Mariadb. This is a major event.

2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for STORAGE ENGINE plugin TokuDB not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_trx not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_lock_waits not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_locks not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_file_map not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_fractal_tree_info not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_fractal_tree_block_map not supported by this version of the server)
 
2017-11-29  9:00:20 139940195203200 [ERROR] Can't open shared library 'ha_tokudb.so' (errno: 8, API version for INFORMATION SCHEMA plugin TokuDB_background_job_status not supported by this version of the server)



 Comments   
Comment by Elena Stepanova [ 2017-11-29 ]

philip_38, I assume you are using 10.2, although the affect version says 10.3.2.Yes, we are already aware, sorry about that. TokuDB 10.2.11 package is coming soon, expectedly later today.

Comment by Philip orleans [ 2017-11-29 ]

How to I tell yum to go back to the previous version of Mariadb so I can back in business?
There is no way to do it.
The command is
yum downgrade MariaDB-* and it fails.

Comment by Sergei Golubchik [ 2017-11-29 ]

We're fixing it, missing tokudb packages are already built, just need some time to get propagated to mirrors.

Comment by Philip orleans [ 2017-11-29 ]

Could you send me via email to my gmail the new plugin, if you have it?
My entire company is down

Comment by Philip orleans [ 2017-11-29 ]

Installed the new RPM and it fails to load.
I have jemalloc installed, what do I do?

2017-11-29 9:40:01 140385635121280 [ERROR] TokuDB is not initialized because jemalloc is not loaded
2017-11-29 9:40:01 140385635121280 [ERROR] Plugin 'TokuDB' init function returned error.
2017-11-29 9:40:01 140385635121280 [ERROR] Plugin 'TokuDB' registration as a STORAGE ENGINE failed.

Comment by Sergei Golubchik [ 2017-11-29 ]

You've probably picked up a wrong tokudb rpm.
The one from centos74 builder is built without jemalloc, on centos73 — with, on centos7 — with.

So if you use tokudb binary that was not built with jemalloc, dynamic linker will not automatically load it for you. You either need to LD_PRELOAD it (mysqld_safe has a malloc-lib option for it) or disable startup jemalloc check in tokudb. Or use a different tokudb rpm.

Comment by Philip orleans [ 2017-11-29 ]

can you indicate what changes exactly needs to be applied to mysqld_safe?
Everybody will have the same issue
I think this is another bug. If Tokudb was built without jemalloc, and before this version it was built with jemalloc, then we need to automate the changes so nodboy goes down unexpectedly.
Business cannot afford to go down, for no reason.

Comment by Elena Stepanova [ 2017-11-29 ]

Most likely for previous installations/upgrades you were using centos7-amd64 repo. That's the one which our repo configuration tool offers for any CentOS 7, regardless subversions:
https://downloads.mariadb.org/mariadb/repositories/#mirror=netinch&distro=CentOS&distro_release=centos7-amd64--centos7&version=10.2

So, if that's what you probably have in your yum repo configuration.

But now, you had to go straight to downloads, so you've chosen centos74 – logical choice, if that's what you have, but not the same as you had before.
Just use whatever flavor you were previously using (you can double-check it in your repo config).

Comment by Philip orleans [ 2017-11-29 ]

You guessed it right in all aspects.
So kindly make sure that the next person does no suffer from the same issue, namely, that jemalloc is installed but it needs to be preloaded as well.
I would love to use jemalloc.

Comment by Sergei Golubchik [ 2017-11-29 ]

tokudb is already in the repository, so upgrading normally should work for the next person with no jemalloc issues.
Very sorry for this, but thanks for the early alert!

Generated at Thu Feb 08 08:14:19 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.