[
https://issues.apache.org/jira/browse/HDFS-8287?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14734591#comment-14734591
]
Tsz Wo Nicholas Sze commented on HDFS-8287:
-------------------------------------------
> ... we needed to copy current cell buffer for ParityGenerateTask. If there
> are any better way to avoid double using of cell buffer rather than copying,
> please let me know that. ...
(I somehow overlooked your comment earlier.) We could avoid copying buffers
using wait-nodify; see below
{code}
synchronized CellBuffers flip() throws InterruptedIOException { //renamed
from setReadyToCurrent
if (bufReady == null) {
try {
wait();
} catch(InterruptedException ie) {
throw DFSUtil.toInterruptedIOException("flip interrupted.", ie);
}
}
CellBuffers tmp = bufCurrent;
bufCurrent = bufReady;
bufCurrent.clear();
bufReady = null;
return tmp;
}
synchronized void releaseBuf(CellBuffers buf) {
bufReady = buf;
notifyAll();
}
//we also need to synchronized, check null and wait-nodify in other methods
{code}
{code}
void writeParityCells() throws IOException {
final CellBuffers cb = doubleCellBuffer.flip();
//encode the data cells
final ByteBuffer[] buffers = cb.getBuffers();
final byte[][] checkSumArrays =
doubleCellBuffer.getCheckSumArrays().clone();
// Create parity packets asynchronously.
completionService.submit(new Callable<Void>() {
@Override
public Void call() throws Exception {
encode(encoder, numDataBlocks, buffers);
for (int i = numDataBlocks; i < numAllBlocks; i++) {
try {
writeParity(i, buffers[i], checkSumArrays[i]);
} catch (IOException e) {
LOG.warn("Caught exception ", e);
}
}
return null;
}
});
doubleCellBuffer.releaseBuf(cb);
}
{code}
> DFSStripedOutputStream.writeChunk should not wait for writing parity
> ---------------------------------------------------------------------
>
> Key: HDFS-8287
> URL: https://issues.apache.org/jira/browse/HDFS-8287
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: hdfs-client
> Reporter: Tsz Wo Nicholas Sze
> Assignee: Kai Sasaki
> Attachments: HDFS-8287-HDFS-7285.00.patch,
> HDFS-8287-HDFS-7285.01.patch, HDFS-8287-HDFS-7285.02.patch,
> HDFS-8287-HDFS-7285.03.patch, HDFS-8287-HDFS-7285.04.patch,
> HDFS-8287-HDFS-7285.05.patch, HDFS-8287-HDFS-7285.06.patch,
> HDFS-8287-HDFS-7285.07.patch
>
>
> When a stripping cell is full, writeChunk computes and generates parity
> packets. It sequentially calls waitAndQueuePacket so that user client cannot
> continue to write data until it finishes.
> We should allow user client to continue writing instead but not blocking it
> when writing parity.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
