accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Keith Turner (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-2570) Consider API additions to support non-batch oriented reads/writes
Date Thu, 27 Mar 2014 22:35:17 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-2570?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13950041#comment-13950041
] 

Keith Turner commented on ACCUMULO-2570:
----------------------------------------

The BatchWriter implementation has an optimization for writing a single Mutation to a tserver
where it will do it in one RPC instead of multiple RPCs.

> Consider API additions to support non-batch oriented reads/writes
> -----------------------------------------------------------------
>
>                 Key: ACCUMULO-2570
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-2570
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: client, tserver
>            Reporter: Josh Elser
>
> Primary Read API : Scanner and BatchScanner
> Primary Write API: BatchWriter
> Both the Read calls are oriented to a scanning a collection of key-values (unless someone
knows of the Range.exact method), and have internals that are focused on multiple, repeated
fetching of batches.
> For writes, Mutations are buffered internally and are focused around sending batches
of updates.
> While these classes do support pointed accesses (write a single Mutation, get the Value
from a single Key), the API is not succinct and may incur unnecessary penalty due to the implementation
also needing to support batch operations.
> It would be nice to consider some new access patterns that are more focused around very
pointed accesses that reduce the API complexity for simple cases. Extremely hypothetical example
> {code}
>   Value v = connector.get("tableName", new Key("row", "cf", "cq", "viz");
>   connector.put("tableName", new Key("row", "cf", "cq", "viz"), new Value("value"));
> {code}
> The implementation could focus on specifically handling this pointed accesses in the
most efficient way without worrying about also handling batch operations.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message