[
https://issues.apache.org/jira/browse/CASSANDRA-1551?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12986808#action_12986808
]
Jon Hermes commented on CASSANDRA-1551:
---------------------------------------
- NPE is caused by the columns being null (see also: a deleted row) in the
return set of the get. In all cases, this can be ignored before calculating
row.cf.columnCount(). In order for listEPH() to be able to see a difference
between getting 0 columns and getting a dead row, it gets 1 column and throws
it away.
- The "garbage" that looked like [][] was 4 bytes representing the ipaddr (127,
0, 0, and 1, for example) being parsed as a UTF-8 string. This was because
deleteHFE() (which was only exposed in this patch, not changed) was calculating
the row key in an older way. HHOM has a row for every endpoint, and the row key
is the bytes of the string representation of the ipaddr ("127.0.0.1", or 9
bytes, not 4). Fixing this removes the garbage and implies that removetoken'ing
a node has never deleted hints for it (hints locations as stored in the
HintsCF, not the actual hints themselves)!
The latter implication is confirmed, so this is a new feature and a bugfix.
> create "tell me what nodes you have hints for" jmx api
> ------------------------------------------------------
>
> Key: CASSANDRA-1551
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1551
> Project: Cassandra
> Issue Type: New Feature
> Components: Tools
> Reporter: Jonathan Ellis
> Assignee: Jon Hermes
> Priority: Minor
> Fix For: 0.7.2
>
> Attachments: 1551-v2.txt, 1551-v3.txt, 1551.txt
>
> Original Estimate: 4h
> Remaining Estimate: 4h
>
> we can do this efficiently in 0.7 due to new HH schema. in 0.6 this would
> require scanning all hints so probably not worth it.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
