[
https://issues.apache.org/jira/browse/FLINK-7970?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
shuai.xu updated FLINK-7970:
----------------------------
Description:
Now all slot allocation is one by one from execution to slot pool. If batch
allocate a number of slots to slot pool, then slot pool can assign its cached
slots according to the global slot requests of all executions, so it can make
an optimal matching between slot and execution, this is especially usefully for
failover. For example, it can assign a slot to the execution whose state is
just on the machine when failover.
!sp.jpg|thumbnail!
was:
Now all slot allocation is one by one from execution to slot pool. If batch
allocate a number of slots to slot pool, then slot pool can assign its cached
slots according to the global slot requests of all executions, so it can make
an optimal matching between slot and execution, this is especially usefully for
failover. For example, it can assign a slot to the execution whose state is
just on the machine when failover.
> SlotPool support batch allocating slots
> ---------------------------------------
>
> Key: FLINK-7970
> URL: https://issues.apache.org/jira/browse/FLINK-7970
> Project: Flink
> Issue Type: Improvement
> Components: JobManager
> Reporter: shuai.xu
> Assignee: shuai.xu
> Labels: flip-6
> Attachments: sp.jpg
>
>
> Now all slot allocation is one by one from execution to slot pool. If batch
> allocate a number of slots to slot pool, then slot pool can assign its cached
> slots according to the global slot requests of all executions, so it can make
> an optimal matching between slot and execution, this is especially usefully
> for failover. For example, it can assign a slot to the execution whose state
> is just on the machine when failover.
> !sp.jpg|thumbnail!
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
