Added Galera autoassemble procedure description for HA cluster.

Closes-Bug: #1250573

Change-Id: Ic93640cddc1d0fb8d55b153da1039288909f81a4
This commit is contained in:
Andrey Korolyov 2013-12-13 01:12:47 +04:00
parent 0fea35a524
commit 9fcc81a545

View File

@ -73,10 +73,32 @@ redeploying the environment. To do so, follow these steps:
HowTo: Enable/Disable Galera Cluster Autorebuild Mechanism
----------------------------------------------------------
By defaults Fuel reassembles Galera cluster automatically without need for any
user interaction.
By default Fuel reassembles Galera cluster automatically without need for any
user interaction. The OCF script looks for existense of `mysqlprimaryinit`
variable and tries to assemble the cluster in a following way:
To prevent `autorebuild feature` you shall do::
- script checks for variable above and for temporary config file created
explicitly on deployment stage - if first is present and second missing,
then script enters autorebuild phase, otherwise it is skipped;
- script checks for the status of the current node - if it is syncronized
with quorum, procedure stops, otherwise latest epoch takes place as a
corosync variable parameter for selected node;
- sleep for five minutes come in place allowing other nodes to join
corosync quorum and push their epochs too;
- for every node in quorum script does comparison of epochs and if there is
at least one node with higher epoch, skips the rest and starts standalone
service which will join newly formed cluster later; if there is no higher
epochs script checks for temporary `mysqlmaster` flag and neither sets it if
flag does not exists nor falls back to the standalong mode;
- mysql process launches with empty `gcomm://` string forming a new quorum
and `mysqlmaster` deletes immediately, other nodes joins in a very short time.
To prevent `autorebuild feature` you should do::
crm_attribute -t crm_config --name mysqlprimaryinit --delete
@ -84,6 +106,10 @@ To re-enable `autorebuild feature` you should do::
crm_attribute -t crm_config --name mysqlprimaryinit --update done
To try an automated reassemble without reboot if cluster is broken just issue::
crm resource restart clone_p_mysql
.. index:: HowTo: Troubleshoot Corosync/Pacemaker
How To Troubleshoot Corosync/Pacemaker