[
https://issues.apache.org/jira/browse/KAFKA-1229?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14218561#comment-14218561
]
Troy Collinsworth edited comment on KAFKA-1229 at 11/19/14 10:01 PM:
---------------------------------------------------------------------
In a virtualized environment like Google Compute or AWS, if the ZooKeeper
quorum membership node dies and a new node is started to replace it, the new
node will likely have a new hostnames/IP addresses. If nodes are rolling
replaced to update security patches they will have new hostnames/IP addresses.
To avoid having to restart Kafka which would cause leadership changes and
partition migration, it would be ideal if Kafka could reload the config or at
least have the ZooKeeper quorum membership list updated dynamically without
restarting the Kafka brokers.
ZooKeeper 3.5-ALPHA has dynamic reconfiguration that manages the ZooKeeper
quorum membership in ZooKeeper znodes. Referencing or dynamically updating the
Kafka ZooKeeper quorum membership list from the ZooKeeper znodes would
consolidate and solve this issue.
was (Author: tcollinsworth):
In a virtualized environment like Google Compute or AWS, if the ZooKeeper
quorum membership node dies and a new node is started to replace it, the new
node will likely have a new hostnames/IP addresses. If nodes are rolling
replaced to update security patches they will have new hostnames/IP addresses.
To avoid having to restart Kafka which would cause leadership changes and
partition migration, it would be ideal if Kafka could reload the config or at
least have the ZooKeeper quorum membership list updated dynamically without
restarting the Kafka brokers.
ZooKeeper 3.5-ALPHA has dynamic reconfiguration that manages the ZooKeeper
quorum membership in ZooKeeper znodes. referencing or dynamically updating the
Kafka ZooKeeper quorum membership list from the ZooKeeper znodes would
consolidate and solve this issue.
> Reload broker config without a restart
> --------------------------------------
>
> Key: KAFKA-1229
> URL: https://issues.apache.org/jira/browse/KAFKA-1229
> Project: Kafka
> Issue Type: Wish
> Components: config
> Affects Versions: 0.8.0
> Reporter: Carlo Cabanilla
> Priority: Minor
>
> In order to minimize client disruption, ideally you'd be able to reload
> broker config without having to restart the server. On *nix system the
> convention is to have the process reread its configuration if it receives a
> SIGHUP signal.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
