phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lars Hofhansl (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (PHOENIX-4165) Do not wait no new memory chunk can be allocated
Date Thu, 07 Sep 2017 04:50:00 GMT

     [ https://issues.apache.org/jira/browse/PHOENIX-4165?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Lars Hofhansl updated PHOENIX-4165:
-----------------------------------
    Attachment: 4165-v3.txt

-v3 includes a test:

100 threads allocate chunk of memory in memory, until each of them fails an allocation. Make
sure that all memory is used. Then all threads free their allocated chunks, make sure everything
was released - I can't think of anything better to test.

This should be good to go.

> Do not wait no new memory chunk can be allocated
> ------------------------------------------------
>
>                 Key: PHOENIX-4165
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4165
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: Lars Hofhansl
>         Attachments: 4165.txt, 4165-v2.txt, 4165-v3.txt
>
>
> Currently the code waits for up to 10s by fault for memory to become "available".
> I think it's better to fail immediately and the let the client retry rather than waiting
on an HBase handler thread.
> In a first iteration we can simply set the max wait time to 0 (or perhaps even -1) so
that we do not attempt to wait but fail immediately. All using code should already deal with
InsufficientMemoryExceptions, since they can already happen right now,
> In a second step I'd suggest to actually remove the waiting code and config option completely.
> [~jamestaylor]



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message