[
https://issues.apache.org/jira/browse/HBASE-13291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14390055#comment-14390055
]
stack commented on HBASE-13291:
-------------------------------
bq. There can be a way to turn off the cell level TTL?
We are not even getting that far. Its just the check to see if any tags at all.
bq. But because it was checking if there was tags it appeared in the profiler
for me.
Tags are not enabled. When I next the Cell out of block with
HFileReaderV3#ScannerV3#next, we skip the tags because a flag on the hfile says
there are none. Good. This isCellTTLExpired is higher up in the merge with no
file context to check. It figures there are no tags by looking for a tag length
of non-zero. To figure tag length, it first must parse the key and value ints.
This work shows in two profilers: flight recorder and linux perf.
bq. Ah, fuck, that is my fault, sorry.
We need the TTL check for DLR. Our key layout is a bit of hack. Hopefully we
can get into place where we can do new layout soon so we don't have to do this
kinda gymnastics just to see if the Cell tagged.
> Lift the scan ceiling
> ---------------------
>
> Key: HBASE-13291
> URL: https://issues.apache.org/jira/browse/HBASE-13291
> Project: HBase
> Issue Type: Improvement
> Components: Scanners
> Affects Versions: 1.0.0
> Reporter: stack
> Assignee: stack
> Attachments: 13291.hacks.txt, 13291.inlining.txt, Screen Shot
> 2015-03-26 at 12.12.13 PM.png, Screen Shot 2015-03-26 at 3.39.33 PM.png,
> hack_to_bypass_bb.txt, nonBBposAndInineMvccVint.txt, q (1).png, traces.7.svg,
> traces.filterall.svg, traces.nofilter.svg, traces.small2.svg,
> traces.smaller.svg
>
>
> Scanning medium sized rows with multiple concurrent scanners exhibits
> interesting 'ceiling' properties. A server runs at about 6.7k ops a second
> using 450% of possible 1600% of CPUs when 4 clients each with 10 threads
> doing scan 1000 rows. If I add '--filterAll' argument (do not return
> results), then we run at 1450% of possible 1600% possible but we do 8k ops a
> second.
> Let me attach flame graphs for two cases. Unfortunately, there is some
> frustrating dark art going on. Let me try figure it... Filing issue in
> meantime to keep score in.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
