Default Admin Role is missing ACLs settings for Create DB Node and Dev Studio.
When viewing Global Jobs, the installation Progress window cannot be resized vertically.
DB Variables page does not load properly.
Find Most Advanced Node job sent with the wrong cluster id (0) causing it to fail.
Controller:
Postgres: postgres|postmaster executable names are both supported meaning that the postmaster process is now properly handled.
Javascript: s9s/host/disk_space_usage.js could not handle multiple partitions.
Javascript: s9s/mysql/schema/schema_check_*.js – prevent it to run if there are more than 1024 tables (configurable) to prevent I_S caused stalls.
Reading disk partition information failed as non root user.
Patch release: Nov 2nd, 2015
Build number:
clustercontrol-1.2.11-842
clustercontrol-cmonapi-1.2.11-135
clustercontrol-controller-1.2.11-998
Bugs fixed:
UI:
Change the favicon for ClusterControl to the one that is used on our site www.severalnines.com
MongoDB add node to replica set looks wrong.
Global Job Messages: Local cluster jobs are shown in the popup dialog.
Fix in Manage > Schema Users. Drop user even if user is empty (‘’@‘localhost’).
Add/Register Existing Galera Node: The Add Node button does not react/work if there is no configuration files in
the dropdown for the Add New DB Node form.
MongoDB add node to replica set dialog – text was cut.
Add/Register Existing Galera Node: The Add Node button does not react/work if there is no configuration files in the dropdown for the Add New DB Node form.
[PostgreSQL] Empty DB Performance graphs.
Installation progress window text disappears while scrolling back.
Controller:
Galera: Register_node job: registers node with wrong type.
Create DB Cluster: Checking OS is the same on all servers.
Create DB Cluster/Node, Add Node: Install cronie on Redhat/Centos.
Scheduled backups that are stored both on controller and on node (full and incremantals) fail to restore.
Increase size of Properties column in server_node table to contain 16384 characters. The following is needed on the cmon db: ALTER TABLE server_node MODIFY properties VARCHAR(16384) DEFAULT '';
CmonHostManager::pull(..): lets properly handle if JSon parse failes.
MongoDB: Check if there is a new member in the replica set and then reload the config.
MySQL: Bugfix for replication mysqldump backuping issues (appeared recently): lets exclude the temporary (name starts with #) DBs from backup.
Postgres: Add existing replication slave failed.
Character set on connection + cmon.tx_deadlock_log, change to use utf8mb4 to properly encode characters in Performance > Transaction Log preventing data from being shown. Do mysql -ucmon -p -h127.0.0.1 cmon < /usr/share/cmon/cmon_db.sql to recreate this table
Patch release: Oct 23rd, 2015
Build number:
clustercontrol-1.2.11-826
clustercontrol-cmonapi-1.2.11-131
clustercontrol-controller-1.2.11-985
Bugs fixed:
UI:
In Cluster jobs list, Delete and Restart buttons do not work.
Add Replication Slave UI Dialog not showing properly.
Editing a previously created backup schedule alters the hostname, and backup job fails.
Number counter on ‘Alarms’ and ‘Logs’ tabs doesn’t make sense.
User Management – refresh/reload button and corrected text for CREATE USER.
Controller:
Backup fix to support xtrabackup 2.3.
Start-up bugs to initialise internal host structures.
netcat port defaults to 9999 (and impossible to change).
Cluster failure with Unknown database some_schema message.
Remove Node: wsrep_cluster_address is not updated.
Corrected printout in backup
Corrected sampling of wsrep_flow_cntr_sent/recv
Patch release: Oct 16th, 2015
Build number:
clustercontrol-1.2.11-808
clustercontrol-cmonapi-1.2.11-128
clustercontrol-controller-1.2.11-974
This is a our best release yet for Postgres with a number of improvements.
Create a new Postgres Node/cluster from the “Create Database Node” dialog or add an new node with a few click.
You can now easily add a new replication slave for your Postgres master node.
The replication peformance and status is shown on the overview page for the slave.
You can restore a backup created by ClusterControl on a specific node.
Create your own dashboard with stats to chart/graph on the overview page like MySQL based clusters.
DB performance charts on the Nodes page.
View database status and variables on your postgres nodes side by side.
Enabled DevStudio, i.e., our JavaScript based advisors which was introduced in 1.2.10 for Postgres as well.
Create your own postgres advisors/DB minions for alarms or email notifications.
MaxScale for MySQL based clusters. MariaDB MaxScale is an open-source, database-centric proxy that works with MariaDB Enterprise, MariaDB Enterprise Cluster, MariaDB 5.5, MariaDB 10 and Oracle MySQL.
Deploy MaxScale instance for round-robin or read/write splitter with a customizable configuration.
Add an existing running MaxScale instance.
Send commands to max admin and view the output in ClusterControl.
MySQL Based Clusters.
You can now use CoderShip as the Galera vendor for Create Cluster and Database node.
Create a MySQL Replication Master Node from the Create DB Node dialog. Currently only Percona as vendor is supported.
Add/Register an existing running MySQL slave without stopping and provisioning the dataset from the cluster.
Create Cluster and Database Node now support using “internal repositories” for environments where you do not have internet access and have internal repostory servers instead.
Removed the limit of only being able to chart 8 DB stats. You can now arrange the charts in a layout with 2 or 3 columns and chart up to 20 stats.
Fixes to Clone Cluster and the UI notification system/look.
Backup individual schemas.
Option to enable ‘wsrep_desync’ during backup for Galera clusters to workaround stalls/issues with FLUSH TABLES WITH READ LOCK. Puts the backup node into ‘Donor/Desynced’ state during the backup.
Manage Email Notifications for all users at once
New Database Logs page
We have a new page specifically for database logs that you access from Logs →Database Logs.
A tree view lists your DB nodes so you can simply pick the nodes that you want to check the mysql error log for
Revamped Configuration Management.
New implementation and look using our JS engine and a set of js scripts.
Group Changes. Automatically change and persist individual database variables across your DB nodes at once. If it’s a dynamic variable we’ll change it directly on the nodes.
Revamped MySQL User Management.
New implementation and look using our JS Engine and a set of js scripts.
We’ removed the old implementation where we maintained users created from ClusterControl separately.
Users and privileges are set directly and retrieved from your cluster so you are always in sync.
Create your users across more than one cluster at once.
HAProxy and KeepAlived
You can now add existing running HAProxy and Keepalived instances that have been installed outside of ClusterControl.