Wei-Chiu Chuang created HDDS-15611:
--------------------------------------
Summary: [Docs] Explain the internal workflow of key deletion with
Ozone Snapshot
Key: HDDS-15611
URL: https://issues.apache.org/jira/browse/HDDS-15611
Project: Apache Ozone
Issue Type: Task
Reporter: Wei-Chiu Chuang
h1. Problem
The current System Internals documentation for Key Deletion (delete.md
https://ozone.apache.org/docs/next/system-internals/data-operations/delete)
explains the basic asynchronous deletion path (moving keys to the active
deletedTable and purging blocks via the KeyDeletingService ). However, it
does not explain how key deletion behaves when Ozone Snapshots are enabled on
a bucket.
Specifically, it lacks details on:
1. How deleted keys are statefully partitioned between the active DB and
snapshot DBs.
2. How the KeyDeletingService coordinates key reclaiming across active and
snapshot database checkpoints.
3. How snapshot deletion/reclamation merges and moves deleted keys along the
snapshot chain.
Adding this workflow is essential for operators and developers to understand
space/quota usage and metadata reclaim mechanics in snapshot-enabled
buckets.
h1. Proposed Documentation Updates
We should update the System Internals Delete document to cover the following
snapshot key deletion mechanics:
1. Deleted Key Partitioning at Snapshot Creation:
• Explain that when a snapshot is taken, the active DB is checkpointed.
• Immediately after the checkpoint is created, the snapshot creation
transaction cleans up/purges all keys under the target bucket from the active
DB's deletedTable and deletedDirTable .
• This partitions deleted keys: keys deleted before a snapshot reside in
the snapshot's local deletedTable , while keys deleted after the latest
snapshot reside in the active DB's deletedTable , preventing
double-counting and key duplication.
2. Reclaimable Key Filtering:
• Explain how the KeyDeletingService and DirectoryDeletingService use
the ReclaimableKeyFilter to check if a deleted key is referenced by
prior snapshots before reclaiming its block space.
3. Snapshot Reclamation (Snapshot Delete):
• Explain the workflow of snapshot deletion/purging (via
OMSnapshotMoveDeletedKeysRequest ).
• Detail how deleted keys are moved from the deleted snapshot's
deletedTable to the next snapshot's deletedTable in the chain, or back to the
active DB's deletedTable if it was the last snapshot.
4. Quota Maintenance:
• Explain how snapshotUsedBytes and snapshotUsedNamespace track the
space/namespace of keys pending deletion, and how the Quota Repair tool
recalculates these metrics across all active snapshot databases.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
