[
https://issues.apache.org/jira/browse/HDFS-5434?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13853579#comment-13853579
]
Eric Sirianni commented on HDFS-5434:
-------------------------------------
Good question! To be frank, we haven't fully implemented append yet. Our
current design approach relies on shared storage (see HDFS-5318) in our
{{FsDatasetSpi}} plugin in order to provide a multi-node pipeline in the append
case for {{repcount=1}}. With shared storage, the single physical replica is
reported via _multiple_ DataNodes to the NameNode. For append, the NameNode
should include _all_ those DataNodes in the append pipeline (see caveat below).
Note that this requires some _out-of-band_ coordination in our
{{FsDatasetSpi}} plugin in order to actually persist the appended data to the
shared replica in a consistent manner.
So, to summarize, we would not rely on the {{BlockPlacementPolicy}} extension
to enforce a multiinode append pipeline with {{repcount=1}}. Instead, we would
rely on shared storage and multiple replica reporting to accomplish this. I
realize that this asymmetry somewhat invalidates my earlier assertion that a
general solution for divorcing the repcount from the pipeline length is
achievable.
Let me know if this makes sense or if any clarifications are needed - I may be
assuming too much context here.
h5. Caveat:
Actually only those on Storages reported as {{READ_WRITE}} should be included
in the append pipeline. This may be a gap in the current NameNode append code
- I'll follow up on this. This also illustrates why your suggestion on
HDFS-5318 of reporting only a _single_ {{READ_WRITE}} node for a given shared
replica may be problematic - we wouldn't get a multi-node pipeline for append
in this case.
> Write resiliency for replica count 1
> ------------------------------------
>
> Key: HDFS-5434
> URL: https://issues.apache.org/jira/browse/HDFS-5434
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 2.2.0
> Reporter: Buddy
> Priority: Minor
> Attachments: BlockPlacementPolicyMinPipelineSize.java,
> BlockPlacementPolicyMinPipelineSizeWithNodeGroup.java, HDFS_5434.patch
>
>
> If a file has a replica count of one, the HDFS client is exposed to write
> failures if the data node fails during a write. With a pipeline of size of
> one, no recovery is possible if the sole data node dies.
> A simple fix is to force a minimum pipeline size of 2, while leaving the
> replication count as 1. The implementation for this is fairly non-invasive.
> Although the replica count is one, the block will be written to two data
> nodes instead of one. If one of the data nodes fails during the write, normal
> pipeline recovery will ensure that the write succeeds to the surviving data
> node.
> The existing code in the name node will prune the extra replica when it
> receives the block received reports for the finalized block from both data
> nodes. This results in the intended replica count of one for the block.
> This behavior should be controlled by a configuration option such as
> {{dfs.namenode.minPipelineSize}}.
> This behavior can be implemented in {{FSNameSystem.getAdditionalBlock()}} by
> ensuring that the pipeline size passed to
> {{BlockPlacementPolicy.chooseTarget()}} in the replication parameter is:
> {code}
> max(replication, ${dfs.namenode.minPipelineSize})
> {code}
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
