[
https://issues.apache.org/jira/browse/HBASE-18070?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17234952#comment-17234952
]
Huaxiang Sun commented on HBASE-18070:
--------------------------------------
Let me try to answer these questions.
{quote}What problems can be solved by read replicas feature?
{quote}
Assume read replica here means the original read replica design. It provides
high availability of read service with the con that data maybe stale.
{quote}What problems can be solved by this issue but not by the original read
replicas feature?
{quote}
Since the feature only applies to meta table for now, so the discussion will
only apply to meta table. The original meta replicas provides high
availability, but it does not solve the hotspot issue. All reads still go to
the primary meta region and then go to all meta replicas region if there is no
response from the primary meta region within certain time.
{quote}The pros and cons of the old read replicas feature?
{quote}
Again, the discussion is limited to meta replicas feature. The existing meta
replicas provides high availability, it does not solve the hotspot issue. Since
there is no realtime replication of meta wal edits, the gap between the primary
meta and replica meta regions could be big, which means the location data from
meta replica regions is more possible to be stale.
{quote}The pros and cons of the solution for this issue?
{quote}
Hbase-client has logic built-in to deal with stale location data. Based on this
fact, the new meta replicas load balance mode distributes meta location lookup
requests across all meta replica regions, it goes to the primary meta region
only when location from meta replicas regions is stale. To ensure data in meta
replicas region catches up with the primary meta region, this features also
turns on real-time replication of meta data based on phase-2 feature from the
original read replicas feature.
This new mode solves the hotspot issue. The current implementation does not
provide high availability in case of meta replica server failures or similar
(this is the same behavior as there is no meta replicas enabled). This can be
improved with follow-ons.
> Enable memstore replication for meta replica
> --------------------------------------------
>
> Key: HBASE-18070
> URL: https://issues.apache.org/jira/browse/HBASE-18070
> Project: HBase
> Issue Type: New Feature
> Reporter: Hua Xiang
> Assignee: Huaxiang Sun
> Priority: Critical
> Fix For: 3.0.0-alpha-1, HBASE-18070, HBASE-18070.branch-2, 2.5.0
>
>
> Based on the current doc, memstore replication is not enabled for meta
> replica. Memstore replication will be a good improvement for meta replica.
> Create jira to track this effort (feasibility, design, implementation, etc).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
