[
https://issues.apache.org/jira/browse/HDDS-13388?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Andika updated HDDS-13388:
-------------------------------
Summary: Support clearing a bucket by using DeleteRange (was: Support
emptying a bucket by using DeleteRange)
> Support clearing a bucket by using DeleteRange
> ----------------------------------------------
>
> Key: HDDS-13388
> URL: https://issues.apache.org/jira/browse/HDDS-13388
> Project: Apache Ozone
> Issue Type: Improvement
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
>
> We had a bucket which contains 500 million keys and was requested to be
> deleted entirely. We decided to delete by using our internal lifecycle
> service which will send delete requests for each of these keys.
> However, after deleting the keys in the bucket and when we tried to delete
> the bucket (which triggers OMBucketDeleteRequest), the OM is stuck for a few
> minutes, causing unavailability. Furthermore, when we tried to list this
> bucket, it causes the OM CPU to be maxed out due to the high number of block
> decompressed by the iterator. This is also accompanied with
> * High DB Mutex Wait
> * High DB Number Iter Skip
> * High last level read bytes
> * High non-last level read bytes
> The main suspect is that there are a lot of tombstones for most of these 500
> million of keys in the bottom levels which have not been deleted. The main
> way we plan to handle this is to manually compact this (HDDS-12533,
> HDDS-12310, and HDDS-12819).
> However, looking back, we can instead use DeleteRange to simply deprecating a
> bucket to reduce the tombstones (since no users are going to write to the
> bucket anymore). We can try to support admin-only OM request that will
> trigger this.
> Ref: https://rocksdb.org/blog/2018/11/21/delete-range.html
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
