[ 
https://issues.apache.org/jira/browse/IGNITE-2952?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15238833#comment-15238833
 ] 

Denis Magda commented on IGNITE-2952:
-------------------------------------

1. If value or key class is undefined then an exception has to be thrown. I'm 
talking about this line of the code

{noformat}
if (ModelUtil.canCreateInstance(keyClass) && 
ModelUtil.canCreateInstance(valueClass))
{noformat}

2. {personCachePartitions} must be re-calculated each time a new compute is 
going to be sent.

3. Try to call {{close}} explicitly after streaming is finished because I see 
that not all streaming Threads are closed.


> Add yardstick benchmark for cache load testing
> ----------------------------------------------
>
>                 Key: IGNITE-2952
>                 URL: https://issues.apache.org/jira/browse/IGNITE-2952
>             Project: Ignite
>          Issue Type: Test
>            Reporter: Semen Boikov
>            Assignee: Vladislav Pyatkov
>
> Need implement yardstick benchmark which will be used for cache load testing 
> (add it in 'yardstick' module).
> For this load testing nodes will be started with several pre-configured 
> caches. Benchmark on each call of 'test' method should iterate over all 
> configured caches (can get list of caches using Ignite.cacheNames) and 
> execute some random operations:
> - put(All)
> - get(All)
> - invoke(All)
> - remove(All)
> - putIfAbsent
> - replace
> - scan query
> If cache is transactional it also should execute cache operations inside 
> explicitly start transaction with random concurrency/isolation mode.
> This benchmark can be run in scenario when server nodes are restarted, so for 
> explicit transaction need use method IgniteBenchmarkUtils.doInTransaction 
> which has logic for exception handling.
> It should be possible to pre-load some cache data before starting test , 
> there should be special benchmark parameter which specifies how many entries 
> load in caches on start (see IgniteSqlQueryBenchmark.setUp as example of 
> benchmark doing preloading)
> Also it should use non-primitives objects as cache keys/values. Value class 
> should have String, int, long, double, byte[] fields.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to