[
https://issues.apache.org/jira/browse/HBASE-11004?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nick Dimiduk updated HBASE-11004:
---------------------------------
Attachment: spans.png
spans.txt
HBASE-11004.00.patch
Here's an initial patch. It requires a couple tweaks I made to htrace, [pull
request|https://github.com/cloudera/htrace/pull/27].
I've verified it works, doesn't leak spans, etc via the included changes to
HLogPerfEval. For example this command
{noformat}
./bin/hbase org.apache.hadoop.hbase.regionserver.wal.HLogPerformanceEvaluation
-Dhbase.trace.spanreceiver.classes=org.htrace.impl.LocalFileSpanReceiver
-Dhbase.local-file-span-rec
eiver.path=/tmp/spans.txt -threads 1 -iterations 100 -syncInterval 10
{noformat}
Produces this highly illegible chat, also attached. More convincing, a grep
through the spans.txt reports 211 total entries: 100 loop iterations, 100
appends, 10 syncs, and 1 thread. Increasing to 2 threads produces 422 entries
of similarly doubled proportions.
> Extend traces through FSHLog#sync
> ---------------------------------
>
> Key: HBASE-11004
> URL: https://issues.apache.org/jira/browse/HBASE-11004
> Project: HBase
> Issue Type: Improvement
> Components: wal
> Reporter: Nick Dimiduk
> Assignee: Nick Dimiduk
> Attachments: HBASE-11004.00.patch, spans.png, spans.txt
>
>
> Changes introduced in HBASE-8755 decouple wal append from wal sync. A gap was
> left in the tracing of these requests. I believe this means are spans are
> decoupled from the work happening over on HDFS-5274. This ticket is to close
> the air-gap between threads.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
