[
https://issues.apache.org/jira/browse/HBASE-15537?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15225562#comment-15225562
]
Yu Li commented on HBASE-15537:
-------------------------------
AFAIK which DNs to use for each block is decided by
{{BlockPlacementPolicy#chooseTarget}} in HDFS, and I just used the default
policy when testing multiwal with FSHLog.
>From the current implement of {{FanOutOneBlockAsyncDFSOutput}} the block
>placement seems to follow the same way as pipeline write (#createOutput still
>depends on NameNodeRpcServer#addBlock for block locations):
{code}
locatedBlock = namenode.addBlock(src, client.getClientName(), null, null,
stat.getFileId(),
null);
List<Channel> datanodeList = new ArrayList<>();
futureList = connectToDataNodes(conf, clientName, locatedBlock, 0L, 0L,
PIPELINE_SETUP_CREATE,
summer, eventLoop);
for (Future<Channel> future : futureList) {
// fail the creation if there are connection failures since we are
fail-fast. The upper
// layer should retry itself if needed.
datanodeList.add(future.syncUninterruptibly().getNow());
}
{code}
So IMHO we could still believe NN for DNs load balance and don't need to worry
about this in our testing. Or could you further clarify your concern if I
misunderstood the question [~stack]? Thanks.
> Add multi WAL support for AsyncFSWAL
> ------------------------------------
>
> Key: HBASE-15537
> URL: https://issues.apache.org/jira/browse/HBASE-15537
> Project: HBase
> Issue Type: Sub-task
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Fix For: 2.0.0
>
> Attachments: HBASE-15537-v3.patch, HBASE-15537-v4.patch,
> HBASE-15537.patch, HBASE-15537_v2.patch
>
>
> The multi WAL should not be bound with {{FSHLog}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
