[
https://issues.apache.org/jira/browse/HBASE-18070?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17232521#comment-17232521
]
Michael Stack edited comment on HBASE-18070 at 11/16/20, 4:49 AM:
------------------------------------------------------------------
{quote}This meta replica Load Balance mode should increase throughput as well.
The previous design is for High availability, so it is not for throughput. In
this new mode, load balance is introduced so read is offloaded mostly from
primary meta and distributed among multiple meta replica regions. In case of
the newly PE test, all location data is static, so if there are enough clients
does location lookup, it could exhaust the non-meta-replica region server, with
meta replica LB, it can support n(replica regon #) * read requests/second.
{quote}
[~huaxiangsun] yes... if enough load but we don't seem to be driving enough
load.. we seem bottlenecked on client, but improving perf is secondary?
{quote}With that said, the PE test should not increase primary read numbers at
all (there is no fall-back-to-primary path in the test), I will take a look
tomorrow to see why there are so many reads going through the primary meta
region, could be something wrong in the path.
{quote}
Yes, would be good to figure where the primary load is coming from. TODO.
[~zhangduo]
{quote}First you said you do not need to prove that the meta lookup request has
been distributed to meta replicas, as it has been done by the Meta Read
Replicas feature which has been done years ago.
{quote}
The 'hedged read' scenario has been in place forever, yes.
{quote}Then you said the feature is about distributing load of meta, which
directly objects your first argument?
{quote}
Distributing load is the 'LoadBalance' configuration, which is new and not the
'HedgedRead'.
{quote}Is it because my English is too bad? Could anyone help me to understand
better? Thanks a lot.
{quote}
Keep asking questions. I'll try and explain. I may not be doing a good job of
it. My impression is that you are want us to make reports around HA which would
be good to have but not the thrust of this issue. I think it would be good to
do but don't want it to be obstacle to commit. Thanks.
was (Author: stack):
{quote}This meta replica Load Balance mode should increase throughput as well.
The previous design is for High availability, so it is not for throughput. In
this new mode, load balance is introduced so read is offloaded mostly from
primary meta and distributed among multiple meta replica regions. In case of
the newly PE test, all location data is static, so if there are enough clients
does location lookup, it could exhaust the non-meta-replica region server, with
meta replica LB, it can support n(replica regon #) * read requests/second.
{quote}
[~huaxiangsun] yes... if enough load but we don't seem to be driving enough
load.. we seem bottlenecked on client, but improving perf is secondary?
{quote}With that said, the PE test should not increase primary read numbers at
all (there is no fall-back-to-primary path in the test), I will take a look
tomorrow to see why there are so many reads going through the primary meta
region, could be something wrong in the path.
{quote}
Yes, would be good to figure where the primary load is coming from. TODO.
[~zhangduo]
{quote}First you said you do not need to prove that the meta lookup request has
been distributed to meta replicas, as it has been done by the Meta Read
Replicas feature which has been done years ago.
{quote}
The 'hedged read' scenario has been in place forever, yes.
{quote}Then you said the feature is about distributing load of meta, which
directly objects your first argument?
{quote}
Distributing load is the 'LoadBalance' configuration, which is new and not the
'HedgedRead'.
{quote}Is it because my English is too bad? Could anyone help me to understand
better? Thanks a lot.
{quote}
Keep asking questions. I'll try and explain. My impression is that you are want
us to make reports around HA which would be good to have but not the thrust of
this issue. I think it would be good to do but don't want it to be obstacle to
commit. Thanks.
> 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, 2.4.0, HBASE-18070, HBASE-18070.branch-2
>
>
> 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)
