[
https://issues.apache.org/jira/browse/HDFS-16333?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei-Chiu Chuang updated HDFS-16333:
-----------------------------------
Component/s: erasure-coding
> fix balancer bug when transfer an EC block
> ------------------------------------------
>
> Key: HDFS-16333
> URL: https://issues.apache.org/jira/browse/HDFS-16333
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: balancer & mover, erasure-coding
> Reporter: qinyuren
> Assignee: qinyuren
> Priority: Major
> Labels: pull-request-available
> Fix For: 3.4.0, 3.3.2, 3.2.4
>
> Attachments: image-2021-11-18-17-25-13-089.png,
> image-2021-11-18-17-25-50-556.png, image-2021-11-18-17-28-03-155.png
>
> Time Spent: 6h 40m
> Remaining Estimate: 0h
>
> We set the EC policy to (6+3) and we also have nodes that were
> decommissioning when we executed balancer.
> With the balancer running, we find many error logs as follow.
> !image-2021-11-18-17-25-13-089.png|width=858,height=135!
> Node A wants to transfer an EC block to node B, but we found that the block
> is not on node A. The FSCK command to show the block status as follow
> !image-2021-11-18-17-25-50-556.png|width=607,height=189!
> In the dispatcher. getBlockList function
> !image-2021-11-18-17-28-03-155.png!
>
> Assume that the location of the an EC block in storageGroupMap look like this
> indices:[0, 1, 2, 3, 4, 5, 6, 7, 8]
> node:[a, b, c, d, e, f, g, h, i]
> after decommission operation, the internal block on indices[1] were
> decommission to another node.
> indices:[0, 1, 2, 3, 4, 5, 6, 7, 8]
> node:[a, {color:#FF0000}j{color}, c, d, e, f, g, h, i]
> the location of indices[1] change from node {color:#FF0000}b{color} to node
> {color:#FF0000}j{color}.
>
> When the balancer get the block location and check it with the location in
> storageGroupMap.
> If a node is not found in storageGroupMap, it will not be add to block
> locations.
> In this case, node {color:#FF0000}j {color}will not be added to the block
> locations, while the indices is not updated.
> Finally, the block location may look like this,
> indices:[0, 1, 2, 3, 4, 5, 6, 7, 8]
> {color:#FF0000}block.location:[a, c, d, e, f, g, h, i]{color}
> the location of the nodes does not match their indices
>
> Solution:
> we should update the indices and match with the nodes
> {color:#FF0000}indices:[0, 2, 3, 4, 5, 6, 7, 8]{color}
> {color:#FF0000}block.location:[a, c, d, e, f, g, h, i]{color}
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
