errose28 opened a new pull request, #9987:
URL: https://github.com/apache/ozone/pull/9987

   ## What changes were proposed in this pull request?
   
   Currently there is no coordination between reconciliation and container 
scanner by design since they are both long running background processes. As 
part of this design, there was a known corner case that could cause the 
checksum to deviate before eventually converging if scanner and reconciliation 
ran at the same time:
   
   The container scanner builds the merkle tree in memory as it is running, and 
then flushes it to disk when it finishes scanning the container. If the 
container is updated (via reconciliation) while the scanner is running, the 
scanner may not see these updates and as a result persist a stale tree at the 
end of its scan. A subsequent container scan would then see the fully updated 
container contents and write the correct tree, restoring the system to a stable 
state.
   
   This change prevents this case by having the container scanner check the 
data checksum when it starts, and then again before it persists the tree. If 
the checksum has changed, the scanner knows its work is now invalid and it must 
rescan the container to build a new tree.
   
   A maximum number of retries per container is added so that the scanner does 
not get stuck in a loop on a single container and fail to process others. When 
the max retries is hit, the scan will be skipped without any affect on the 
container and its last scan timestamp will not be updated. This will cause the 
container to be retried first after the whole container set is finished 
scanning and the next round of scanning begins.
   
   Changes generated by Cursor with review and modifications by me.
   
   ## What is the link to the Apache JIRA
   
   HDDS-14908
   
   (Please replace this section with the link to the Apache JIRA)
   
   ## How was this patch tested?
   
   Unit tests added for the following cases:
   - Retry with eventual success.
   - Retry without success requiring the scan to be skipped.
   - Container iterator correctly orders containers when some timestamps are 
not updated during processing, causing skipped containers to be tried first in 
the next iteration.
   
   This change does not apply to the metadata scanner, which does not read or 
write container data checksums.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to