[ 
https://issues.apache.org/jira/browse/STORM-876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14990636#comment-14990636
 ] 

ASF GitHub Bot commented on STORM-876:
--------------------------------------

Github user d2r commented on a diff in the pull request:

    https://github.com/apache/storm/pull/845#discussion_r43953169
  
    --- Diff: storm-core/src/clj/backtype/storm/daemon/nimbus.clj ---
    @@ -1103,21 +1230,39 @@
          (.set_reset_log_level_timeout_epoch log-config (coerce/to-long 
timeout))
          (.unset_reset_log_level_timeout_epoch log-config))))
     
    +(defmethod blob-sync-code :distributed [conf nimbus]
    +  (if (not (is-leader nimbus :throw-exception false))
    +    (let [storm-cluster-state (:storm-cluster-state nimbus)
    +          nimbus-host-port-info (:nimbus-host-port-info nimbus)
    +          blob-store-key-list (into [] (get-key-set-from-blob-store 
(:blob-store nimbus)))
    --- End diff --
    
    It would be better to have a `nimbus/get-key-list-from-blob-store` 
function, and have `nimbus/get-key-set-from-blob-store` make a list out of it, 
rather than doing a list -> set -> list conversion.


> Dist Cache: Basic Functionality
> -------------------------------
>
>                 Key: STORM-876
>                 URL: https://issues.apache.org/jira/browse/STORM-876
>             Project: Apache Storm
>          Issue Type: Improvement
>          Components: storm-core
>            Reporter: Robert Joseph Evans
>            Assignee: Robert Joseph Evans
>         Attachments: DISTCACHE.md, DistributedCacheDesignDocument.pdf
>
>
> Basic functionality for the Dist Cache feature.
> As part of this a new API should be added to support uploading and 
> downloading dist cache items.  storm-core.ser, storm-conf.ser and storm.jar 
> should be written into the blob store instead of residing locally. We need a 
> default implementation of the blob store that does essentially what nimbus 
> currently does and does not need anything extra.  But having an HDFS backend 
> too would be great for scalability and HA.
> The supervisor should provide a way to download and manage these blobs and 
> provide a working directory for the worker process with symlinks to the 
> blobs.  It should also allow the blobs to be updated and switch the symlink 
> atomically to point to the new blob once it is downloaded.
> All of this is already done by code internal to Yahoo! we are in the process 
> of getting it ready to push back to open source shortly.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to