[
https://issues.apache.org/jira/browse/KUDU-3686?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexey Serbin updated KUDU-3686:
--------------------------------
Description:
As of time of writing, after modifying the set of Kudu masters in a cluster
(e.g., adding new masters when going from single- to multi-master
configurations) it's necessary to restart all the tablet servers in the
cluster, and that's documented as Item 7 in the [Migrating to Multiple Kudu
Masters
process|https://kudu.apache.org/docs/administration.html#perform-the-migration]
process.
However, if a tablet server still heartbeats to at least one master that's a
part of the current set of Kudu masters in the cluster, it's possible to
piggy-back the current list of Kudu masters with the {{TSHeartbeatResponsePB}}
that's sent as a response from master to the tablet server.
Implementing that would be a useful operability improvement.
was:
As of time of writing, after modifying the set of Kudu masters in a cluster
(e.g., adding new masters when going from single- to multi-master
configurations) it's necessary to restart all the tablet servers in the
cluster, and that's
[documented|https://kudu.apache.org/docs/administration.html#perform-the-migration]
at least for the [Migrating to Multiple Kudu
Masters|https://kudu.apache.org/docs/administration.html#migrate_to_multi_master]
process.
However, if a tablet server still heartbeats to at least one master that's a
part of the current set of Kudu masters in the cluster, it's possible to
piggy-back the current list of Kudu masters with the {{TSHeartbeatResponsePB}}
that's sent as a response from master to the tablet server.
Implementing that would be a useful operability improvement.
> Tablet servers to automatically refresh list of Kudu masters in the cluster
> ---------------------------------------------------------------------------
>
> Key: KUDU-3686
> URL: https://issues.apache.org/jira/browse/KUDU-3686
> Project: Kudu
> Issue Type: Improvement
> Components: master, tserver
> Reporter: Alexey Serbin
> Priority: Major
> Labels: operability, supportability
>
> As of time of writing, after modifying the set of Kudu masters in a cluster
> (e.g., adding new masters when going from single- to multi-master
> configurations) it's necessary to restart all the tablet servers in the
> cluster, and that's documented as Item 7 in the [Migrating to Multiple Kudu
> Masters
> process|https://kudu.apache.org/docs/administration.html#perform-the-migration]
> process.
> However, if a tablet server still heartbeats to at least one master that's a
> part of the current set of Kudu masters in the cluster, it's possible to
> piggy-back the current list of Kudu masters with the
> {{TSHeartbeatResponsePB}} that's sent as a response from master to the tablet
> server.
> Implementing that would be a useful operability improvement.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
