[
https://issues.apache.org/jira/browse/MESOS-3736?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14965463#comment-14965463
]
Gilbert Song commented on MESOS-3736:
-------------------------------------
This is a note referring to current consideration after in person discussion
with Jojy. Thanks to [~jojy].
*Still considering two question:
1. Handling simultaneous failure. If the first request is called and is
written into the hashmap. All the other requests will be waiting for the future
of the first request. But because its return type is 'Future<vector<string>>',
if its future status is 'FAILED/DISCARDED', the other requests will be waiting
forever.
2. The current hashmap uses 'stringify(image::name)' as key, but it may not be
unique because there is chance that layer_ids can be changed. One solution is
to have 'stringify(image)' as key.
> Support docker local store pull same image simultaneously
> ----------------------------------------------------------
>
> Key: MESOS-3736
> URL: https://issues.apache.org/jira/browse/MESOS-3736
> Project: Mesos
> Issue Type: Improvement
> Reporter: Gilbert Song
> Assignee: Gilbert Song
>
> The current local store implements get() using the local puller. For all
> requests of pulling same docker image at the same time, the local puller just
> untar the image tarball as many times as those requests are, and cp all of
> them to the same directory, which wastes time and bear high demand of
> computation. We should be able to support the local store/puller only do
> these for the first time, and the simultaneous pulling request should wait
> for the promised future and get it once the first pulling finishes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
