[
https://issues.apache.org/jira/browse/HBASE-613?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12599261#action_12599261
]
Jim Kellerman commented on HBASE-613:
-------------------------------------
This is ugly.
Because InternalScanner specifies next as:
{code}
public boolean next(HStoreKey key, SortedMap<byte[], byte[]> results)
{code}
it is only possible to return one timestamp for all the results of a row.
Consequently, Cells on the client side are meaningless with respect to
timestamp. MemcacheScanner.next sets the timestamp in the HStoreKey to the
timestamp requested when the scanner was created. So if the timestamp requested
when the scanner was created was HConstants.LATEST_TIMESTAMP, that is what gets
set. Similarly, if a specific timestamp was set when the scanner was created,
older entries in the cache may not be found.
So what I propose is to change InternalScanner.next to be:
{code}
public byte[] next(SortedMap<byte[], Cell> results)
{code}
where the return value is the row key or null if there are not results and the
results map key is the column name and value is the Cell (value, timestamp)
pair. This should make it easier to determine what results should be returned.
> Timestamp-anchored scanning fails to find all records
> -----------------------------------------------------
>
> Key: HBASE-613
> URL: https://issues.apache.org/jira/browse/HBASE-613
> Project: Hadoop HBase
> Issue Type: Bug
> Components: client
> Reporter: stack
> Assignee: Jim Kellerman
> Fix For: 0.2.0
>
> Attachments: TestTimestampScanning.java
>
>
> If I add 3 versions of a cell and then scan across the first set of added
> cells using a timestamp that should only get values from the first upload, a
> bunch are missing (I added 100k on each of the three uploads). I thought it
> the fact that we set the number of cells found back to 1 in HStore when we
> move off current row/column but that doesn't seem to be it. I also tried
> upping the MAX_VERSIONs on my table and that seemed to have no effect. Need
> to look closer.
> Build a unit test because replicating on cluster takes too much time.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
