[
https://issues.apache.org/jira/browse/IGNITE-7489?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16347017#comment-16347017
]
Ilya Kasnacheev commented on IGNITE-7489:
-----------------------------------------
I don't know, I have tried your test and it seems to check for a very peculiar
case where every object is of the same size. FreeList only contains one
non-empty bucket in this case.
If you change item size to random, suddently you will only see load factor of
0.9.
If you factor all allocated pages in, you only see load factor of 0.999 (this
of course assumes that any non-data pages are full).
I just don't see how it would be a useful metric.
> Weird FillFactor metric fluctuation.
> ------------------------------------
>
> Key: IGNITE-7489
> URL: https://issues.apache.org/jira/browse/IGNITE-7489
> Project: Ignite
> Issue Type: Bug
> Affects Versions: 2.1, 2.2, 2.3
> Reporter: Andrew Mashenkov
> Assignee: Ilya Kasnacheev
> Priority: Major
> Attachments: FillFactorTest.java
>
>
> For now there is no way to get Used\FreeMemory for region in bytes.
> MemoryMetrics.getPagesFillFactor() javadoc says that the method return the
> percentage of space that is still free and can be filled in.
> So, I'd think used memory can be calculated as
> PhysicalMemoryPages*PageSize*FillFactor.
> I've tried to use this, but found weir thing.
>
> PFA a repro.
> There is 2 node grid with no persistence configure. Topology is stable.
> Cache is populated with unique keys (no updates) and observe allocated data
> pages metric grows constantly as expected.
> But the error look too large, used memory (and FillFactor as well) may 2-10+
> time differs.
> E.g. allocated pages, fillFactor, usedMem (bytes):(
> node-0: 13789 0.851563 48096032
> node-1: 14447 0.781250 46230400
> In next second:
> node-0: 13958 0.039063 2233280
> node-1: 14624 0.347656 20824576
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
