[
https://issues.apache.org/jira/browse/HBASE-20408?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrew Purtell updated HBASE-20408:
-----------------------------------
Description:
Admin APIs, but not all, should automatically retry to ride over master
failover scenarios. This should be implemented as closely as possible to
existing retry strategy for client APIs, using common code, perhaps refactored.
Note we are not retrying if the API call returns a failure indication. We are
retrying if communication between the admin client and HMaster has been
interrupted, or the HMaster has failed over. We are only retrying until we know
the result status of the request.
Not all admin APIs are safe or can be made idempotent.
was:Admin APIs, but perhaps not all, should automatically retry to ride over
master failover scenarios. This should be implemented as closely as possible to
existing retry strategy for client APIs, using common code, perhaps refactored.
> Retry policy for admin APIs
> ---------------------------
>
> Key: HBASE-20408
> URL: https://issues.apache.org/jira/browse/HBASE-20408
> Project: HBase
> Issue Type: New Feature
> Reporter: Divesh Jain
> Assignee: Divesh Jain
> Priority: Major
>
> Admin APIs, but not all, should automatically retry to ride over master
> failover scenarios. This should be implemented as closely as possible to
> existing retry strategy for client APIs, using common code, perhaps
> refactored.
> Note we are not retrying if the API call returns a failure indication. We are
> retrying if communication between the admin client and HMaster has been
> interrupted, or the HMaster has failed over. We are only retrying until we
> know the result status of the request.
> Not all admin APIs are safe or can be made idempotent.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
