[
https://issues.apache.org/jira/browse/HDFS-9850?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15526978#comment-15526978
]
Anu Engineer commented on HDFS-9850:
------------------------------------
bq. This corner case was already there and not introduced by this jira
I am not convinced that this statement is correct. Let us enumerate the cases.
{panel:title=Old behaviour}
# If a source or destination disk is removed *while* diskbalancer is running --
There is a failure -- crash perhaps ?, I will check that by running the test
case and the plan status flag will be stuck in PLAN_UNDER_PROGRESS.
# if a source or destination disk is removed *after* diskbalancer is completely
run. A queryStatus would return the status that it remembers.
{panel}
As far as I understand (Please correct me if I am wrong), the new behaviour is
as follows.
{panel:title=New behaviour}
# If a source or destination disk is removed *while* diskbalancer is running --
There is a failure and the error message is set appropriately, but plan status
flag will be stuck in PLAN_UNDER_PROGRESS -- and details of error is the error
message.
# if a source or destination disk is removed *after* diskbalancer is completely
run. _A queryStatus would return an error saying that disk is missing_.
{panel}
The fact that some disk got removed *after* disk balancer was run should not be
any consequence to Disk balancer. I am not arguing about the first case at all.
My concern is about *second point* -- behaviour of queryStatus after
diskbalancer has finished running. I am arguing that it should not fail if a
disk is missing -- when querystatus is done.
> DiskBalancer : Explore removing references to FsVolumeSpi
> ----------------------------------------------------------
>
> Key: HDFS-9850
> URL: https://issues.apache.org/jira/browse/HDFS-9850
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: balancer & mover
> Affects Versions: 3.0.0-alpha2
> Reporter: Anu Engineer
> Assignee: Manoj Govindassamy
> Attachments: HDFS-9850.001.patch, HDFS-9850.002.patch,
> HDFS-9850.003.patch
>
>
> In HDFS-9671, [~arpitagarwal] commented that we should explore the
> possibility of removing references to FsVolumeSpi at any point and only deal
> with storage ID. We are not sure if this is possible, this JIRA is to explore
> if that can be done without issues.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
