[
https://issues.apache.org/jira/browse/HBASE-11777?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14116454#comment-14116454
]
Anoop Sam John commented on HBASE-11777:
----------------------------------------
I see.. you mean here Ram?
StoreFileScanner#setCurrentCell
{code}
if(this.cur != null && this.reader.isBulkLoaded() && cur.getSequenceId() <=
0) {
KeyValue curKV = KeyValueUtil.ensureKeyValue(cur);
curKV.setSequenceId(this.reader.getSequenceID());
{code}
Compactor#performCompaction
{code}
for (Cell c : kvs) {
KeyValue kv = KeyValueUtil.ensureKeyValue(c);
if (cleanSeqId && kv.getSequenceId() <= smallestReadPoint) {
kv.setSequenceId(0);
{code}
Up in the scan layer we set the seqId and here we can get an object of
BufferedDataBlockEncoder.ClonedSeekerState. I know the 1st one is added
recently. But both these existing code were not good wrt performance. We added
BufferedDataBlockEncoder.ClonedSeekerState to avoid need for copy value part
also when we scan encoded data. But see the call to KeyValueUtil.ensureKeyValue
which will recreate KV and copy both key and value parts!!!
I can correct this by BufferedDataBlockEncoder.ClonedSeekerState implementing
SettableSequenceId and now we wont be having KeyValueUtil.ensureKeyValue.
Also suggest we can get rid of this KeyValueUtil.ensureKeyValue. We can work
on that with sub tasks later.
> Find a way to use KV.setSequenceId() on Cells on the server-side read path
> --------------------------------------------------------------------------
>
> Key: HBASE-11777
> URL: https://issues.apache.org/jira/browse/HBASE-11777
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 0.99.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 0.99.0, 2.0.0
>
> Attachments: CellWithSequenceNumber.java, HBASE-11777.patch,
> HBASE-11777_V2.patch, HBASE-11777_V3.patch
>
>
> Over in HBASE-11591 there was a need to set the sequenceId of the HFile to
> the bulk loaded KVs. Since we are trying to use the concept of Cells in the
> read path if we need to use setSequenceId(), then the Cell has to be
> converted to KV and only KeyValue impl has the accessor setSequenceId().
> [~anoop.hbase] suggested if we can use a Server side impl of Cell and have
> these accessors in them.
> This JIRA aims to solve this and see the related code changes that needs to
> be carried out for this.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
