[
https://issues.apache.org/jira/browse/HBASE-29328?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Anuj Sharma updated HBASE-29328:
--------------------------------
Description:
We are implementing a new hbase command called refresh_hfiles which will pick
up the new edits that has happened on the active cluster post creation of read
replica cluster.
When user create read replica cluster the active cluster can continue to serve
the writes (put, delete flush etc). These can lead to flush or splits or
compactions which will update the store files (HFiles) and its in-memory view
for HStore instance. With refresh_hfiles command read only cluster gets awards
of the changes happened on active cluster post its creation.
This command needs to get executed post refresh_meta command so that newly
created/updated region configuration is safely propagated to read replica which
will avoid inconsistency and data loss on read replica cluster.
There will be three flavors of the command:
1. For no parameter: {_}refresh_hfiles{_}, refresh hfiles for all user table
2. For table name as parameter : {_}refresh_hfiles "TABLE_NAME" =>
"table_name"{_}, refresh hfiles for a table
3. For namespace as parameter: {_}refresh_hfiles "NAMESPACE" => "namespace"{_},
refresh hfiles for all tables under namespace
See [Design
Doc|https://docs.google.com/document/d/1EI0lsURX1BZhv3DYgMvZCl4EUy-ADJRkHUc1PjzZtj0/edit?tab=t.0#heading=h.udx6mn3dj24c]
for more details.
> Implement new HBase command: refresh_hfiles
> -------------------------------------------
>
> Key: HBASE-29328
> URL: https://issues.apache.org/jira/browse/HBASE-29328
> Project: HBase
> Issue Type: Sub-task
> Reporter: Rahul Agarkar
> Assignee: Anuj Sharma
> Priority: Major
> Labels: pull-request-available
>
> We are implementing a new hbase command called refresh_hfiles which will pick
> up the new edits that has happened on the active cluster post creation of
> read replica cluster.
> When user create read replica cluster the active cluster can continue to
> serve the writes (put, delete flush etc). These can lead to flush or splits
> or compactions which will update the store files (HFiles) and its in-memory
> view for HStore instance. With refresh_hfiles command read only cluster gets
> awards of the changes happened on active cluster post its creation.
> This command needs to get executed post refresh_meta command so that newly
> created/updated region configuration is safely propagated to read replica
> which will avoid inconsistency and data loss on read replica cluster.
> There will be three flavors of the command:
> 1. For no parameter: {_}refresh_hfiles{_}, refresh hfiles for all user table
> 2. For table name as parameter : {_}refresh_hfiles "TABLE_NAME" =>
> "table_name"{_}, refresh hfiles for a table
> 3. For namespace as parameter: {_}refresh_hfiles "NAMESPACE" =>
> "namespace"{_}, refresh hfiles for all tables under namespace
> See [Design
> Doc|https://docs.google.com/document/d/1EI0lsURX1BZhv3DYgMvZCl4EUy-ADJRkHUc1PjzZtj0/edit?tab=t.0#heading=h.udx6mn3dj24c]
> for more details.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
