[
https://issues.apache.org/jira/browse/FLINK-6469?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16499962#comment-16499962
]
ASF GitHub Bot commented on FLINK-6469:
---------------------------------------
Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/5448
Okay, after taking a look, I think we need to add a few changes:
- We need to add an additional `MemoryUnit.parse()` method that takes the
"default" unit, so that we parse the old heap sizes such that they are in MB if
nothing else is specified.
- We should either change the return value of `getMebiBytes()` to `int`
or have a `getMebiBytesAsInt()` method that uses a
`MathUtils.checkedDownCast()` to avoid unnoticed overflow errors.
Open question: As we are changing the value type of the heap size config
options, should we deprecate the current config keys and introduce new ones
(like `jobmanager.heap-size`)?
> Configure Memory Sizes with units
> ---------------------------------
>
> Key: FLINK-6469
> URL: https://issues.apache.org/jira/browse/FLINK-6469
> Project: Flink
> Issue Type: Improvement
> Components: Core
> Reporter: Stephan Ewen
> Assignee: vinoyang
> Priority: Major
>
> Currently, memory sizes are configured by pure numbers, the interpretation is
> different from configuration parameter to parameter.
> For example, heap sizes are configured in megabytes, network buffer memory is
> configured in bytes, alignment thresholds are configured in bytes.
> I propose to configure all memory parameters the same way, with units similar
> to time. The JVM itself configured heap size similarly: {{Xmx5g}} or
> {{Xmx2000m}}.
> {code}
> 10000 -> bytes
> 10 kb
> 64 mb
> 1 gb
> ...
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
