[
https://issues.apache.org/jira/browse/PHOENIX-1314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14172653#comment-14172653
]
Lars Hofhansl commented on PHOENIX-1314:
----------------------------------------
Thinking about this a bit over night... The details are tricky, aren't they?
* Will we ever make different decisions in different stores (column families)?
* If so, do we *have* to seek backwards or will otherwise miss rows?
* If we have to seek backwards, can we still guarantee that the rows get to the
client in sorted order?
* Will we even need seek backwards to a previous row? Or is it just in the way
that we create the KVs that causes problems? In other words, can we change the
assert to check that we do not go backwards in terms of row keys?
* This will get tricky if one column family does not have any columns for a
row. Then do we need to seek to the minimum of the hints we found, or the
maximum?
bq. first store scanner reseek to varchar7 second store scanner still scans
kvs less than hint
That's not the case, though, right? They seek to the same KV, no? (Maybe I am
not interpreting the log right).
> Assertion tripped for skip scan with two unit tests
> ---------------------------------------------------
>
> Key: PHOENIX-1314
> URL: https://issues.apache.org/jira/browse/PHOENIX-1314
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.1
> Reporter: James Taylor
> Assignee: rajeshbabu
> Attachments: PHOENIX-1314.patch, PHOENIX-1314_3.0.patch,
> PHOENIX-1314_v3.patch, Phoenix-1314_3.0.patch
>
>
> After checking in a pretty sizeable change
> (https://git-wip-us.apache.org/repos/asf?p=phoenix.git;a=commit;h=d018cc1c6e01d9836de6e67af4f8b91de3269bfd),
> the assert in SkipScanFilter.setNextCellHint() started getting tripped for
> the following unit tests:
> - DeleteIT.testDeleteAllFromTableWithIndexNoAutoCommitNoSalting()
> - MutableIndexIT.testCoveredColumnUpdatesWithLocalIndex()
> The tests seem to pass without this assert (not sure what HBase does if the
> seek next hint is the same as the current row), but it would be good to get
> to the bottom of this. Would you mind taking a look, [~rajesh23]? It's
> possibly related to PHOENIX-1313.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
