zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jordan Zimmerman <jor...@jordanzimmerman.com>
Subject ANN: Exhibitor 1.4.0
Date Sun, 14 Oct 2012 01:23:54 GMT
* When shutting down ZooKeeper, attempt a simple kill first before trying kill -9.

* Issue 51: Include log4j in standalone version

* Reworked how CLI options are displayed in the standalone version. There are so many now
it was
getting confusing.

* Issue 54: Added a new REST API to the Cluster APIs to get the status/state of the entire
See the wiki for details.

* Issue 55: Added a way to specify the ACL in the standalone version. Run with "--help" to
how to set the options.

* Issue 57: Boy - just when you think you know jQuery and Javascript... I didn't realize that
need to be destroyed explicitly and that event handlers have to be unbound explicitly. This
for 99% of the weirdness with the buttons in the UI and the problems with the 4LTR dialog.
I've also
added cache defeating query parameters to all AJAX URLs.

* Issue 58: Added a ZookeeperConfigProvider. The idea here is to use a second Zookeeper cluster
as a datastore for Exhibitor's shared config. You can run Exhibitor on the shared config ensemble
but it will need to be run without shared config. Please see the Github wiki for details.

* The change for Issue 58 involved more CLI options for config. For clarity, I've broken the
options into groups - a group for each type. A new *required* CLI option has been added: "configtype".
It must be either "file", "s3", "zookeeper" or "none". Configtype "none" is a special purpose
to be used for running Exhibitor on the ZooKeeper cluster that is storing config for the main

* Issue 63: Restart, Stop, Start buttons on the control panels now better reflect the state
of the
"Automatic Instance Restarts" switch. If the switch is On, the button is "Restart". If the
switch is
Off, the button is "Stop" if the instance is running and "Start" if the instance is not running.

* Issue 61: The "Additional Config" section should retain the order that it was entered instead
being alphabetized.

* Issue 59: Support a separate Transaction Log directory in the config. It can be blank and
default to the Snapshot directory.

* Updates to the control panel were not warning if another process updated the config at the
same time.

* Automatic instance management checkbox moved to the Config tab.

* Lots of UI/logging tweaks and improvements.

* Major reworking of automatic instance management and some work on rolling config:

    Automatic Instance Management (AIM) no longer uses heartbeats to determine instance state.
    a REST call is periodically made to each instance in the cluster. This should be more

    Instead of making changes one-by-one, AIM now reconfigures the entire cluster. i.e. if
an instance
    is new and it notices there is an instance that's been removed, both changes are made
in one shot.

    The previous config value DEAD_INSTANCE_PERIOD_MS has been removed. In its place is
    AUTO_MANAGE_INSTANCES_SETTLING_PERIOD_MS. AIM waits until the ensemble state (quorum,
members, etc.)
    has been without change for this period before making any changes. Also, the ensemble
must be
    in Quorum for AIM to make changes.

    Previously, Rolling Config Changes would wait for Quorum before the config change advances
    the next instance. However it would only wait up to 4 attempts. If the number of attempts
    exceeded it would advance anyway. This has been changed. If Quorum is not achieved within
    attempts the rolling config change is canceled and the change is force applied. The hope
    is that the ensemble will then settle.

View raw message