[
https://issues.apache.org/jira/browse/HBASE-21032?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16587958#comment-16587958
]
Andrey Elenskiy commented on HBASE-21032:
-----------------------------------------
Even though we don't use read replicas, it seems like it's the same issue. The
last row loaded from hbase:meta was:
18/08/21 19:46:21 INFO assignment.RegionStateStore: Load hbase:meta entry
region=2ad4d95f7b9d9ba0d746b8da50a7f9a7, regionState=OPEN,
lastHost=regionserver-0,16020,1534538897512,
regionLocation=regionserver-3,16020,1533765879856, openSeqNum=172
Which, based on region encoded name, is hbase:namespace. However, the
regionserver is different.
I think I'm going to halt deploying the patched version for now. Instead, I've
run App.java that I've attached and it looks like it behaves as expected.
> ScanResponses contain only one cell each
> ----------------------------------------
>
> Key: HBASE-21032
> URL: https://issues.apache.org/jira/browse/HBASE-21032
> Project: HBase
> Issue Type: Bug
> Components: Performance, Scanners
> Affects Versions: 2.1.0
> Environment: HBase 2.1.0
> Hadoop 2.8.4
> Java 8
> Reporter: Andrey Elenskiy
> Assignee: Andrey Elenskiy
> Priority: Major
> Fix For: 3.0.0, 2.2.0, 2.1.1
>
> Attachments: App.java, HBASE-21032-v1.patch, HBASE-21032-v1.patch,
> HBASE-21032.patch
>
>
> I have a long row with a bunch of columns that I'm scanning with
> setAllowPartialResults(true). In the response I'm getting the first partial
> ScanResponse being around 2MB with multiple cells while all of the consequent
> ones being 1 cell per ScanResponse. After digging more, I found that each of
> those single cell ScanResponse partials are preceded by a heartbeat (zero
> cells). This results in two requests per cell to a regionserver.
> I've attached code to reproduce it on hbase version 2.1.0 (it works as
> expected on 2.0.0 and 2.0.1).
> [^App.java]
> I'm fairly certain it's a serverside issue as
> [gohbase|https://github.com/tsuna/gohbase] client is having the same issue. I
> have not tried to reproduce this with multi-row scan.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
