[
https://issues.apache.org/jira/browse/HBASE-1501?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jonathan Gray updated HBASE-1501:
---------------------------------
Attachment: HBench-v1.tar.gz
Contains 5 assorted tests, each with timings for sequential inserts, random reads, and full scans.
There are a couple static finals in HBench.java that define how long to sleep between calls to HBA flush and major compact (after puts, before gets/scans). There is also a boolean to enable/disable that from happening. Right now I believe it is disabled.
When enabled, you will sleep the specified amount of time and then after the HBA calls and sleeps the program holds for stdin input (return) before continuing (if you want to watch the logs and ensure things are over).
To uncover some of the compaction weirdness and halfhfile stuff, play with enabling and disabling the flush/compact boolean and the sleeping. With no sleep and no manual flushes, I have issues reading from any of the tables which have enough inserts to cause a split. Also, calling the flush/compact and continuing before it's done does some interesting things as well.
Specifically, the avgFat() and wide() cases have both given me problems.
Benchmark Tool for 0.20 API
---------------------------
Key: HBASE-1501
URL:
https://issues.apache.org/jira/browse/HBASE-1501Project: Hadoop HBase
Issue Type: Test
Affects Versions: 0.20.0
Reporter: Jonathan Gray
Assignee: Jonathan Gray
Priority: Trivial
Attachments: HBench-v1.tar.gz
I wrote a set of classes to aid in benchmarking puts, gets, and scans for various table schemas and patterns.
We can turn this in to something as part of contrib down the road, but for now this is just an internal tool I'm sharing to help debug.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
