[
https://issues.apache.org/jira/browse/ASTERIXDB-1929?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chen Luo updated ASTERIXDB-1929:
--------------------------------
Comment: was deleted
(was: I think the answer is NO. Actually I saw this referencing factory idea
initially was used by ExternalBtree, which also needs to insert anti-matter
tuples during bulkload.
The current workflow of this tuple writer is as follows. When we create a
LSMBTree, we would specify its LeafFrameFactory, which depends on the provided
TupleWriterFactory. Then, BTree would call LeafFrameFactory to create leaf
frames, which calls the TupleWriterFactory to create the corresponding tuple
writer. Thus, the tuple writer is only managed by the leaf frame, and BTree
calls its insertTuple method to do insertion. The problem here is that LSMBTree
or even BTree couldn't track the corresponding TupleWriter. So currently, this
LSMTreeRefrencingTupleWriterFactory would track it and set it to delete mode
when necessary.)
> Factor out isAntimatter from TupleWriter and TupleWriterFactory
> ---------------------------------------------------------------
>
> Key: ASTERIXDB-1929
> URL: https://issues.apache.org/jira/browse/ASTERIXDB-1929
> Project: Apache AsterixDB
> Issue Type: Improvement
> Components: Storage
> Reporter: Yingyi Bu
> Assignee: Chen Luo
>
> isAntimatter is a property of the tuple that we're writing, but not a
> property of the Writer. Hence, we probably should remove setAntimatter(...)
> from the TupleWriter.
> Currently, for each tuple write operation, we need to call setAntimatter() in
> both TupleWriterFactory and TupleWriter. This is painful and potentially
> buggy -- a XxxFactory typically is one per NC.
> Instead, we can add a parameter to write tuple method:
> public int writeTuple(ITupleReference tuple, byte[] targetBuf, int targetOff)
> ->
> public int writeTuple(ITupleReference tuple, byte[] targetBuf, int targetOff,
> boolean isDelete)
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
