accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Keith Turner (JIRA)" <>
Subject [jira] [Commented] (ACCUMULO-1602) put the KeyExtent in a major compactors' thread name for easier debugging
Date Tue, 23 Jul 2013 20:04:48 GMT


Keith Turner commented on ACCUMULO-1602:

bq. Don't tablets already have fixed length id's? I thought this was what corresponded to
the HDFS directory.

KeyExtent has a getUUID() method which basically hashes the extent. I think the monitor uses
this.  Older versions of Accumulo may have used this for the HDFS dir, newer versions use
a counter from zookeeper.  The tablets directory could be used instead of the hash, it would
be shorter and should be unique.

bq. Isn't the point of hashes that they are one-way and hard to reverse?

The utility I was thinking of would scan the metadata table hashing each tablets extent until
it finds the extent in question.  If the tablets dir were used, then a special utility would
not be needed.  Could just grep the metadata table.

There is one issue with using the tablet dir, it will not distinguish tablets that split.
 One of the children from a split will have the same tablet dir as the parent.  Sometimes
when debugging its important to distinguish the parent from the child.

We could generate a unique id for each tablet and store this in a column in the metadata table.
 This allows the tablet to be found with grep in the Accumulo shell.  The monitor could display
this id.

> put the KeyExtent in a major compactors' thread name for easier debugging
> -------------------------------------------------------------------------
>                 Key: ACCUMULO-1602
>                 URL:
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: tserver
>            Reporter: Eric Newton
>            Priority: Minor
>              Labels: newbie
>             Fix For: 1.6.0

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see:

View raw message