Merge lp:~mariosplivalo/charms/trusty/percona-cluster/lp1366997-workaround into lp:~openstack-charmers-archive/charms/trusty/percona-cluster/next

[Mario Splivalo]

This MP adds another percona-cluster charm config option, lp1366997-workaround, which adds a specific configuration to mysql so that primary key bug is avoided.

[Edward Hope-Morley (hopem)]

Mario gonna retarget to /next. Please backport to stable once merged.

[Billy Olsen (billy-olsen)]

From my understanding this config flag option 'lp1366997-workaround' is a flag which will be short-lived and hopefully removed if the upstream fix gets backported. I think it'd be best if we can find a solution which avoids the need for future deprecation.

If we can, I think it my be a good idea to have a flexible option which allows other options to be defined in a configuration file in the conf.d directory. This allows us to address this use case and possible future use cases without having to do additional code change with the convenient side-effect of avoiding the option deprecation/removal.

At the very least I think the config option should be renamed to convey a better indicator of what is addressing (symptom even) rather than the lpbug-workaround.

One other possible option is to simply enable this workaround all the time within the template. I'm not entirely sure of what the side-effects are of enabling these options so that may be the least desirable due to inadvertent risk (which is why it is toggled in the first place).

review: Needs Fixing

[Billy Olsen (billy-olsen)]

After discussion, the config option name is acceptable although I'm still not a fan.

Otherwise, code looks fine. I'll approve and merge into the /next branches for now. You can work on a backport.

review: Approve

=== modified file 'config.yaml'
--- config.yaml	2015-02-06 09:48:02 +0000
+++ config.yaml	2015-02-11 20:17:19 +0000
@@ -61,11 +61,15 @@
       disabled and a non-temporary address must be configured/available on
       your network interface.
   innodb-file-per-table:
-     type: boolean
-     default: True
-     description: | 
-       Turns on innodb_file_per_table option, which will make MySQL put each
-       InnoDB table into separate .idb file. Existing InnoDB tables will remain
-       in ibdata1 file - full dump/import is needed to get rid of large
-       ibdata1 file
-     
\ No newline at end of file
+    type: boolean
+    default: True
+    description: |
+      Turns on innodb_file_per_table option, which will make MySQL put each
+      InnoDB table into separate .idb file. Existing InnoDB tables will remain
+      in ibdata1 file - full dump/import is needed to get rid of large
+      ibdata1 file
+  lp1366997-workaround:
+    type: boolean
+    default: False
+    description: Adds two config options (wsrep_drupal_282555_workaround and wsrep_retry_autocommit) as a workaround for Percona Primary Key bug (see lplp1366997).
+

=== modified file 'hooks/percona_hooks.py'
--- hooks/percona_hooks.py	2015-02-10 11:27:46 +0000
+++ hooks/percona_hooks.py	2015-02-11 20:17:19 +0000
@@ -112,9 +112,12 @@
         'cluster_hosts': ",".join(hosts),
         'sst_method': 'xtrabackup',
         'sst_password': mysql_password,
-        'innodb_file_per_table': config('innodb-file-per-table')
+        'innodb_file_per_table': config('innodb-file-per-table'),
+        'lp1366997_workaround': config('lp1366997-workaround')
     }
 
+    print config
+
     if config('prefer-ipv6'):
         # NOTE(hopem): this is a kludge to get percona working with ipv6.
         # See lp 1380747 for more info. This is intended as a stop gap until

=== modified file 'templates/my.cnf'
--- templates/my.cnf	2015-02-06 09:48:02 +0000
+++ templates/my.cnf	2015-02-11 20:17:19 +0000
@@ -33,14 +33,13 @@
 # This changes how InnoDB autoincrement locks are managed and is a requirement for Galera
 innodb_autoinc_lock_mode=2
 
+{% if innodb_file_per_table -%}
 # This enables storing InnoDB tables in separate .ibd files. Note that, however
-# existing InnoDB tables will remain in ibdata file(s) unles OPTIMIZE is run 
+# existing InnoDB tables will remain in ibdata file(s) unles OPTIMIZE is run
 # on them. Still, the ibdata1 file will NOT shrink - a full dump/import of the
 # data is needed in order to get rid of large ibdata file.
-
-{% if innodb_file_per_table %} 
 innodb_file_per_table = 1
-{% else %}
+{% else -%}
 innodb_file_per_table = 0
 {% endif %}
 
@@ -69,10 +68,17 @@
 # query cache is not supported with Active/Active configuration
 innodb_buffer_pool_size = {{ dataset_bytes }}
 
-{% if ipv6 %}
+{% if ipv6 -%}
 [sst]
 sockopt=,pf=ip6
 {% endif %}
 
+{% if lp1366997_workaround -%}
+# Adding workaround for bug:
+#    https://bugs.launchpad.net/charms/+source/percona-cluster/+bug/1366997
+# Note that this should only be required for percona 5.5
+wsrep_drupal_282555_workaround = ON
+wsrep_retry_autocommit = 100
+{% endif %}
 
 !includedir /etc/mysql/conf.d/