[
https://issues.apache.org/jira/browse/KUDU-2105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16133455#comment-16133455
]
Todd Lipcon commented on KUDU-2105:
-----------------------------------
Do you have some kind of short doc to share about what approach you're thinking
of here? Looking briefly at the list above, I think we have already checked
several of those boxes without any new framework beyond the existing
ctest/minicluster/etc?
Here are a few of the existing approaches we have for your bullet points:
bq. 1. To allow node density testing without multi node cluster support.
See dense_node-itest
bq. 2. To allow general perf testing without multi node cluster support.
See things like full_stack-insert-scan-test
bq. 3. To be able to run performance tests which have less variance (due to
network noise).
Our various perf microbenchmarks from benchmarks.sh all run without network
noise
bq. 4. (nice, but not needed) be able to simulate various characteristics such
as network reliability, rpc call failures, under stress without needing to
modify processes on other nodes.
We do a bunch of this in raft_consensus-itest and other tests.
> Create single node stress test framework
> ----------------------------------------
>
> Key: KUDU-2105
> URL: https://issues.apache.org/jira/browse/KUDU-2105
> Project: Kudu
> Issue Type: Test
> Components: perf
> Reporter: Edward Fancher
> Assignee: Edward Fancher
>
> It would be useful to have single node stress test support.
> 1. To allow node density testing without multi node cluster support.
> 2. To allow general perf testing without multi node cluster support.
> 3. To be able to run performance tests which have less variance (due to
> network noise).
> 4. (nice, but not needed) be able to simulate various characteristics such as
> network reliability, rpc call failures, under stress without needing to
> modify processes on other nodes.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
