[
https://issues.apache.org/jira/browse/PHOENIX-7109?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17788812#comment-17788812
]
Aron Attila Meszaros commented on PHOENIX-7109:
-----------------------------------------------
The above mentioned commit had the effect to not use PREAD readtype on hbase
scans, but use the default (stream) type instead. I suppose that is the reason
for the partial empty results returned by hbase.
Seems like the Default readtype first uses PREAD, but after a threshold
switches to STREAM, this can be the reason for repeating step 3 is necessary.
The threshold can be modified by the "hbase.storescanner.pread.max.bytes"
attribute. For test purposes I set a higher threshold, which also resolved the
issue.
> Incorrect query results when using OFFSET
> -----------------------------------------
>
> Key: PHOENIX-7109
> URL: https://issues.apache.org/jira/browse/PHOENIX-7109
> Project: Phoenix
> Issue Type: Bug
> Components: phoenix
> Affects Versions: 5.1.3
> Reporter: Aron Attila Meszaros
> Assignee: Aron Attila Meszaros
> Priority: Major
>
> Steps to reproduce:
> # create table test (i1 integer not null, i2 integer not null, i3 integer,
> i4 integer, CONSTRAINT pk PRIMARY KEY (i1, i2));
> # upsert into test select rand()*40, rand()*10000000, null, rand()*100 from
> SYSTEM.CATALOG limit 130000;
> # upsert into test select rand()*40, rand()*10000000, null, rand()*100 from
> test limit 130000;
> Repeat step 3 a few times, then run
> {noformat}
> select * from test where i1=10 and i4=20 offset X;{noformat}
> where X is bigger than the number of rows satisfying the condition.
> There are rows returned by phoenix (expected result is empty set.)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
