[
https://issues.apache.org/jira/browse/HBASE-9769?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13801529#comment-13801529
]
Lars Hofhansl commented on HBASE-9769:
--------------------------------------
We should test end-to-end, not a microbenchmark of StoreScanner.
Note that you cannot exercise the seeking code in checkVersion without
returning data to the client, in which case network IO will dominate. If you
filter KVs out with a filter before that checkVersion is never called, if the
filter returns INCLUDE it'll call checkVersion and incur a seek. Only with a
coprocessor would it be possible to exercise checkVersion and avoid the network
IO.
Also note that in your filter case you'd still get the
SEEK_NEXT_ROW/SEEK_NEXT_COL in ScanWildcardColumnTracker.checkVersion for each
column that you included.
When you get a chance, could you check out the last patch on HBASE-9778? Maybe
you could run it through your micro StoreScanner test, I'd be curious how it
compares.
Generally, if that the column tracker code is not efficient we should fix that
rather than circumventing it completely with a filter.
> Improve performance of a Scanner with explicit column list when rows are
> small/medium size
> ------------------------------------------------------------------------------------------
>
> Key: HBASE-9769
> URL: https://issues.apache.org/jira/browse/HBASE-9769
> Project: HBase
> Issue Type: Improvement
> Components: Scanners
> Affects Versions: 0.98.0, 0.94.12, 0.96.0
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
> Attachments: 9769-0.94-sample1.txt, 9769-0.94-sample2.txt,
> 9769-0.94-sample.txt, 9769-94.txt, 9769-94-v2.txt, 9769-trunk-v1.txt,
> 9769-trunk-v2.txt, 9769-trunk-v3.txt, 9769-trunk-v4.txt
>
>
--
This message was sent by Atlassian JIRA
(v6.1#6144)
