You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ozone.apache.org by "Lokesh Jain (Jira)" <ji...@apache.org> on 2020/03/17 10:56:00 UTC
[jira] [Commented] (HDDS-3214) Unhealthy datanodes repeatedly
participate in pipeline creation
[ https://issues.apache.org/jira/browse/HDDS-3214?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17060823#comment-17060823 ]
Lokesh Jain commented on HDDS-3214:
-----------------------------------
There are three pipelines. One node in each pipeline has been injected by fault. When a write on a datanode fails it sends pipeline close action to the SCM. SCM then destroys the pipeline and allocates a new one. While allocating a pipeline the only available nodes in SCM are the ones involved in the pipeline destroyed. Therefore a new pipeline is created using the same nodes. As a result the client write never succeeds as this cycle continues for all the pipelines in the cluster.
Ideally the datanode failing should be flagged in SCM until a corrective action is taken. The new pipeline should be created using healthy datanodes.
> Unhealthy datanodes repeatedly participate in pipeline creation
> ---------------------------------------------------------------
>
> Key: HDDS-3214
> URL: https://issues.apache.org/jira/browse/HDDS-3214
> Project: Hadoop Distributed Data Store
> Issue Type: Bug
> Reporter: Nilotpal Nandi
> Priority: Major
> Labels: fault_injection
>
> steps taken :
> 1) Mounted noise injection FUSE on all datanodes
> 2) Selected 1 datanode from each open pipeline (factor=3)
> 3) Injected WRITE FAILURE noise with error code - ENOENT on "hdds.datanode.dir" path of list of datanodes selected in step 2)
> 4) start PUT key operation of size 32 MB.
>
> Observation :
> ----------------
> # Commit failed, pipelines were moved to exclusion list.
> # Client retries , new pipeline is created with same set of datanodes. Container creation fails as WRITE FAILURE injection present.
> # Pipeline is closed and the process is repeated for "ozone.client.max.retries" retries.
> Everytime, same set of datanodes are selected for pipeline creation which include 1 unhealthy datanode.
> Expectation - pipeline should have been created by selecting 3 healthy datanodes available.
>
> cc - [~ljain]
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: ozone-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: ozone-issues-help@hadoop.apache.org