Page MenuHomePhabricator

Test MariaDB 10.6 on Bullseye
Closed, ResolvedPublic

Description

MariaDB 10.6 has a number of features that can be very useful for us (https://mariadb.com/kb/en/changes-improvements-in-mariadb-106/) like:

  • Ignored Indexes
  • Native sys DB support
  • binlog_expire_logs_seconds
  • CREATE, ALTER, DROP and RENAME are crash safe

A bunch of options have been removed, so we'd need to double check if we are still using them: (investigation at: T301879#7759265)

innodb_adaptive_max_sleep_delay
innodb_background_scrub_data_check_interval
innodb_background_scrub_data_compressed
innodb_background_scrub_data_interval
innodb_background_scrub_data_uncompressed
innodb_buffer_pool_instances
innodb_commit_concurrency
innodb_concurrency_tickets
innodb_file_format
innodb_large_prefix
innodb_lock_schedule_algorithm
innodb_log_checksums
innodb_log_compressed_pages
innodb_log_files_in_group
innodb_log_optimize_ddl
innodb_page_cleaners
innodb_replication_delay
innodb_scrub_log
innodb_scrub_log_speed
innodb_sync_array_size
innodb_thread_concurrency
innodb_thread_sleep_delay
innodb_undo_logs

Details

SubjectRepoBranchLines +/-
operations/puppetproduction+0 -19
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+1 -0
operations/puppetproduction+0 -4
operations/puppetproduction+0 -4
operations/puppetproduction+0 -1
operations/softwaremaster+12 -0
operations/puppetproduction+1 -0
operations/puppetproduction+0 -3
operations/puppetproduction+0 -1
operations/puppetproduction+0 -6
operations/puppetproduction+0 -40
operations/puppetproduction+0 -16
operations/puppetproduction+1 -0
operations/puppetproduction+0 -4
operations/puppetproduction+0 -4
operations/puppetproduction+3 -2
Show related patches Customize query in gerrit

Related Objects

StatusSubtypeAssignedTask
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
OpenNone
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
ResolvedMarostegui
Resolvedjcrespo
OpenMarostegui

Event Timeline

There are a very large number of changes, so older changes are hidden. Show Older Changes

Mentioned in SAL (#wikimedia-operations) [2022-04-05T04:34:26Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Pool db1132 for testing T301879', diff saved to https://phabricator.wikimedia.org/P24087 and previous config saved to /var/cache/conftool/dbconfig/20220405-043426-marostegui.json

Mentioned in SAL (#wikimedia-operations) [2022-04-05T05:52:57Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Pool db1132 for testing T301879', diff saved to https://phabricator.wikimedia.org/P24093 and previous config saved to /var/cache/conftool/dbconfig/20220405-055256-marostegui.json

Mentioned in SAL (#wikimedia-operations) [2022-04-05T06:01:25Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Pool db1132 into API for testing T301879', diff saved to https://phabricator.wikimedia.org/P24094 and previous config saved to /var/cache/conftool/dbconfig/20220405-060124-marostegui.json

Just pooled db1132 into API group with very small weight to start capturing API queries.

Mentioned in SAL (#wikimedia-operations) [2022-04-05T06:36:48Z] <marostegui@cumin1001> dbctl commit (dc=all): 'More weight to db1132 T301879', diff saved to https://phabricator.wikimedia.org/P24095 and previous config saved to /var/cache/conftool/dbconfig/20220405-063648-marostegui.json

Change 778206 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/software@master] control-mariadb-10.6-bullseye: Add to repo

https://gerrit.wikimedia.org/r/778206

Pushed the above control file to the repo, so it is kept somewhere.

Change 778206 merged by jenkins-bot:

[operations/software@master] control-mariadb-10.6-bullseye: Add to repo

https://gerrit.wikimedia.org/r/778206

I am feeling pretty comfortable with the 10.6 tests. I will probably leave the host running with small weight on s1 after the holidays (as I am out tomorrow and friday)

Change 784355 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db1132: Enable notifications

https://gerrit.wikimedia.org/r/784355

Mentioned in SAL (#wikimedia-operations) [2022-04-20T04:51:09Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Pool db1132 into s1 T301879', diff saved to https://phabricator.wikimedia.org/P25523 and previous config saved to /var/cache/conftool/dbconfig/20220420-045108-marostegui.json

Change 784355 merged by Marostegui:

[operations/puppet@production] db1132: Enable notifications

https://gerrit.wikimedia.org/r/784355

I have pooled db1132 into s1 (main and api) with small weight:

--- eqiad/groupLoadsBySection/s1 live
+++ eqiad/groupLoadsBySection/s1 generated
@@ -1,6 +1,7 @@
 {
     "api": {
         "db1119": 100,
+        "db1132": 10,
         "db1134": 100,
         "db1135": 100,
         "db1164": 100,
--- eqiad/sectionLoads/s1 live
+++ eqiad/sectionLoads/s1 generated
@@ -7,6 +7,7 @@
         "db1105:3311": 50,
         "db1106": 50,
         "db1119": 200,
+        "db1132": 50,
         "db1134": 200,
         "db1135": 200,
         "db1163": 325,

Going to leave it serving until Friday.

Mentioned in SAL (#wikimedia-operations) [2022-04-21T05:09:31Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Depool db1132 T301879', diff saved to https://phabricator.wikimedia.org/P25874 and previous config saved to /var/cache/conftool/dbconfig/20220421-050931-marostegui.json

Mentioned in SAL (#wikimedia-operations) [2022-04-25T06:38:24Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Repool db1132 into s1 T301879', diff saved to https://phabricator.wikimedia.org/P26432 and previous config saved to /var/cache/conftool/dbconfig/20220425-063823-marostegui.json

Just repooled db1132 into s1, going to leave it serving during the whole week.

This is for parsercache https://gerrit.wikimedia.org/r/c/operations/puppet/+/785798/ but I will do that for production too. That if was there for the time we had 10.1 and 10.4 but now that 10.1 is gone, we can simply leave it disabled.

Change 786670 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] production.my.cnf.erb: Leave rowid disabled

https://gerrit.wikimedia.org/r/786670

Change 786670 merged by Marostegui:

[operations/puppet@production] production.my.cnf.erb: Leave rowid disabled

https://gerrit.wikimedia.org/r/786670

So far db1132 has been performing very well on s1, I am going to increase its weight now on both main and API, and leave it serving for the rest of the week. If all goes fine, I might migrate a s8 host to 10.6 next week, as the queries there might have other issues than a normal wiki.

Mentioned in SAL (#wikimedia-operations) [2022-04-27T08:17:27Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Increase db1132 weight T301879', diff saved to https://phabricator.wikimedia.org/P26703 and previous config saved to /var/cache/conftool/dbconfig/20220427-081727-marostegui.json

Change 787119 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] core-percona.my.cnf.erb: Update optimizer options

https://gerrit.wikimedia.org/r/787119

Change 787119 merged by Marostegui:

[operations/puppet@production] core-percona.my.cnf.erb: Update optimizer options

https://gerrit.wikimedia.org/r/787119

Mentioned in SAL (#wikimedia-operations) [2022-04-28T07:22:01Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Increase db1132 weight T301879', diff saved to https://phabricator.wikimedia.org/P26860 and previous config saved to /var/cache/conftool/dbconfig/20220428-072200-marostegui.json

As everything looks good with db1132, I have given it normal weight (like any other host serving API+MAIN) which is 200 on main and 100 on API. I want to explore if this host, for any reason would report more slow queries than a 10.4 one

Mentioned in SAL (#wikimedia-operations) [2022-04-29T06:30:19Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Depool db1132 T301879', diff saved to https://phabricator.wikimedia.org/P27001 and previous config saved to /var/cache/conftool/dbconfig/20220429-063019-marostegui.json

Depooled db1132 for the long weekend.

Mentioned in SAL (#wikimedia-operations) [2022-05-03T08:04:22Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Repool db1132 T301879', diff saved to https://phabricator.wikimedia.org/P27350 and previous config saved to /var/cache/conftool/dbconfig/20220503-080421-marostegui.json

Mentioned in SAL (#wikimedia-operations) [2022-05-04T14:24:51Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Depool db1132 T301879', diff saved to https://phabricator.wikimedia.org/P27468 and previous config saved to /var/cache/conftool/dbconfig/20220504-142449-marostegui.json

For the first time db1132 has been serving traffic during the weekend and everything looked good apart from the already reported query: T305427#7912741

Mentioned in SAL (#wikimedia-operations) [2022-05-11T05:13:07Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Depool db2146 T301879', diff saved to https://phabricator.wikimedia.org/P27778 and previous config saved to /var/cache/conftool/dbconfig/20220511-051307-marostegui.json

Change 790796 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db2146: Disable notifications

https://gerrit.wikimedia.org/r/790796

Change 790796 merged by Marostegui:

[operations/puppet@production] db2146: Disable notifications

https://gerrit.wikimedia.org/r/790796

Mentioned in SAL (#wikimedia-operations) [2022-05-11T05:17:04Z] <marostegui@cumin1001> dbctl commit (dc=all): 'Repool db2146 T301879', diff saved to https://phabricator.wikimedia.org/P27779 and previous config saved to /var/cache/conftool/dbconfig/20220511-051703-marostegui.json

I am repooling es1032 (10.6.8) - first ES RO host serving traffic with 10.6

After all the few changes on hosts during the last week, these are the hosts currently running live, with 10.6

Version: 10.6.7+deb11u1 4
db1124.eqiad.wmnet (host) (test host)
db1125.eqiad.wmnet (host) (test host)
db2122.codfw.wmnet (host) (s7)
db2146.codfw.wmnet (host) (s1)

Version: 10.6.8+deb11u1 7
db1111.eqiad.wmnet (host) (s8)
db1127.eqiad.wmnet (host) (s7)
db1132.eqiad.wmnet (host) (s1)
es1032.eqiad.wmnet (host) (es1)
es2022.codfw.wmnet (host) (es4)
es2030.codfw.wmnet (host) (es1)
pc2014.codfw.wmnet (host) (pc1)

Change 832588 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db1168: Install 10.6

https://gerrit.wikimedia.org/r/832588

Change 832588 merged by Marostegui:

[operations/puppet@production] db1168: Install 10.6

https://gerrit.wikimedia.org/r/832588

Installed 10.6 (with the fix) for T311106 on s6, so we currently have a replica on:

s1 (db1132)
s4 (db1143)
s6 (db1168)
s7 (db1127)
s8 (db1111)

Change 832607 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db2180: Install 10.6

https://gerrit.wikimedia.org/r/832607

Change 832607 merged by Marostegui:

[operations/puppet@production] db2180: Install 10.6

https://gerrit.wikimedia.org/r/832607

Installed 10.6 (with the fix) for T311106 on s6, so we currently have a replica on:

s1 (db1132)
s4 (db1143)
s6 (db1168)
s7 (db1127)
s8 (db1111)

Added the s6 codfw host too (db2180)

Change 837490 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] mariadb: Remove innodb_large_prefix flag.

https://gerrit.wikimedia.org/r/837490

Change 837491 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db1182: Upgrade from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837491

Change 837491 merged by Marostegui:

[operations/puppet@production] db1182: Upgrade from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837491

Installed 10.6.10 on db1182 and db2175 (s2).

Change 837497 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db1120: Migrate from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837497

Change 837497 merged by Marostegui:

[operations/puppet@production] db1120: Migrate from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837497

Change 837614 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db2178: Migrate from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837614

Change 837614 merged by Marostegui:

[operations/puppet@production] db2178: Migrate from 10.4 to 10.6

https://gerrit.wikimedia.org/r/837614

Installed 10.6.10 on db1200 and db2178 (s5)

Change 838074 had a related patch set uploaded (by Marostegui; author: Marostegui):

[operations/puppet@production] db2181: Install MariaDB 10.6

https://gerrit.wikimedia.org/r/838074

Change 838074 merged by Marostegui:

[operations/puppet@production] db2181: Install MariaDB 10.6

https://gerrit.wikimedia.org/r/838074

Change 837490 merged by Marostegui:

[operations/puppet@production] mariadb: Remove innodb_large_prefix flag.

https://gerrit.wikimedia.org/r/837490

I am going to consider this task done. The pending issue (adapt our tooling to work with 10.6) has its own task where it can be followed up T319383