[
https://issues.apache.org/jira/browse/HDDS-9367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17770247#comment-17770247
]
Ethan Rose commented on HDDS-9367:
----------------------------------
Some ideas to improve error handling:
# Make errors to the RocksDB update fatal, like it is in other components.
# Restore the old OM DB on error.
** Ideally each time an OM snapshot is requested the same error would be hit,
so we would not have to worry about the cause rolling off in the logs.
** However, Recon would appear to be working but never updated. This might
confuse users.
> Improve RocksDB related error handling within Recon
> ---------------------------------------------------
>
> Key: HDDS-9367
> URL: https://issues.apache.org/jira/browse/HDDS-9367
> Project: Apache Ozone
> Issue Type: Improvement
> Reporter: Ethan Rose
> Priority: Major
>
> For issues like HDDS-8734, Recon will close the RocksDB instance
> [here|https://github.com/apache/ozone/blob/2c578c3ae73417d2fd2f24ab356632ddb5c427fb/hadoop-ozone/recon/src/main/java/org/apache/hadoop/ozone/recon/recovery/ReconOmMetadataManagerImpl.java#L123],
> but farther up
> [here|https://github.com/apache/ozone/blob/2a826133d681e3ed8a789c77b6c6f8c622e4c743/hadoop-ozone/recon/src/main/java/org/apache/hadoop/ozone/recon/spi/impl/OzoneManagerServiceProviderImpl.java#L543]
> the exception is swallowed. This leaves Recon running but the OM DB
> inaccessible. It will return 500 from APIs needing that DB and continue to
> log errors, but the initial cause of the DB close could be lost if the logs
> roll off from all the errors. A Recon instance could remain in this
> half-working state for quite some time if it is not actively being used.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
