[
https://issues.apache.org/jira/browse/HIVE-19902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16539384#comment-16539384
]
Alexander Kolbasov commented on HIVE-19902:
-------------------------------------------
[~owen.omalley] Answering your questions above.
# I do not use JMS for the reasons described earlier. The primary reason is
that you separate parts of the execution that you want to measure from parts
that you do want to measure. A typical example:
{code}
for (int i = 0; i < iterations; i++) {
create_table("foo") // measure time for this
drop_table("foo") // do not measure
}
{code}
# I agree that it belongs in standalone-metastore but it is impossible
currently because you can't have any submodules of standalone-metastore. This
is addressed by HIVE-20097. We agreed that once HIVE-20097 is in place this
will move under standalone-metastore.
# As I explained above I can't use JMH for my purposes, so I did implement very
simple framework (it is really tiny).
# Answering your question about goals. My goal is to measure RPC costs for
actual running HMS instances. These RPC costs (in all interesting cases) are
dominated by the cost of server-side implementation (and it is very easy to
adjust for network latencies).
> Provide Metastore micro-benchmarks
> ----------------------------------
>
> Key: HIVE-19902
> URL: https://issues.apache.org/jira/browse/HIVE-19902
> Project: Hive
> Issue Type: Improvement
> Components: Standalone Metastore
> Affects Versions: 3.1.0, 4.0.0
> Reporter: Alexander Kolbasov
> Assignee: Alexander Kolbasov
> Priority: Major
> Attachments: HIVE-19902.01.patch, HIVE-19902.02.patch,
> HIVE-19902.03.patch, HIVE-19902.04.patch
>
>
> It would be very useful to have metastore benchmarks to be able to track perf
> issues.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
