[
https://issues.apache.org/jira/browse/HDDS-10731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17839809#comment-17839809
]
Ethan Rose commented on HDDS-10731:
-----------------------------------
AFAIK the version of Ozone that this was encountered on was old, before the
significant changes to block and container delete like the {{isEmpty}} flag
from HDDS-8115 that datanodes use to initiate container deletion. After this
fix empty containers should correctly move to the DELETED state. Once this
happens, HDDS-7126 will make sure the transactions for those deleted containers
get removed. I don't think this version of Ozone had that fix either. [~Sammi]
please double check with the current code and let us know if the issue
described here is still relevant.
> Avoid send block delete commands to DN if the involved container is closed
> and empty
> ------------------------------------------------------------------------------------
>
> Key: HDDS-10731
> URL: https://issues.apache.org/jira/browse/HDDS-10731
> Project: Apache Ozone
> Issue Type: Improvement
> Reporter: Sammi Chen
> Assignee: Sammi Chen
> Priority: Major
>
> During investigate one SCM crash case, we found that the SCM DB deletedBlock
> table has 900M records while customer said the DN's disk utilization looks
> normal and didn't have any major space not reclaimed issue.
> {noformat}
> bin/ozone debug ldb --db=/Users/sammi/workspace/ozone-meta/scm.db scan --cf
> deletedBlocks --count
> 951177554
> {noformat}
> After further checked the deletedBlock table content, there are lots of txs
> whose target container is already closed and empty. Here is one example,
> Container 1087
> {noformat}
> {
> "state" : "CLOSED",
> "stateEnterTime" : 1706614618.156000000,
> "replicationConfig" : {
> "replicationFactor" : "THREE",
> "replicationType" : "RATIS"
> },
> "usedBytes" : 0,
> "numberOfKeys" : 0,
> "lastUsed" : 1713412552.968000000,
> "owner" : "om2",
> "containerID" : 1087,
> "deleteTransactionId" : 2071,
> "sequenceId" : 0
> }
> {noformat}
> Block delete transactions for this container 1087
> {noformat}
> 1564: {
> "memoizedSize" : -1,
> "unknownFields" : {
> "fields" : { }
> },
> "bitField0_" : 7,
> "txID_" : 1564,
> "containerID_" : 1087,
> "localID_" : [ 111677748019211347, 111677748019211366, 111677748019211404 ],
> "count_" : 0,
> "memoizedIsInitialized" : 1,
> "memoizedSerializedSize" : -1,
> "memoizedHashCode" : 0
> }
> 1834: {
> "memoizedSize" : -1,
> "unknownFields" : {
> "fields" : { }
> },
> "bitField0_" : 7,
> "txID_" : 1834,
> "containerID_" : 1087,
> "localID_" : [ 111677748019211432 ],
> "count_" : 0,
> "memoizedIsInitialized" : 1,
> "memoizedSerializedSize" : -1,
> "memoizedHashCode" : 0
> }
> 2071: {
> "memoizedSize" : -1,
> "unknownFields" : {
> "fields" : { }
> },
> "bitField0_" : 7,
> "txID_" : 2071,
> "containerID_" : 1087,
> "localID_" : [ 111677748019211288 ],
> "count_" : 0,
> "memoizedIsInitialized" : 1,
> "memoizedSerializedSize" : -1,
> "memoizedHashCode" : 0
> }
> {noformat}
> For those kinds of txs, we can avoid send the commands to DN, instead we can
> directly delete the txs from the DB.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
