rich7420 commented on code in PR #10895:
URL: https://github.com/apache/ozone/pull/10895#discussion_r4505713168
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeysDeleteRequest.java:
##########
@@ -384,6 +454,48 @@ protected OMClientResponse
getOmClientResponse(OzoneManager ozoneManager,
return omClientResponse;
}
+ /**
+ * Inserts a delete marker for every key that survived validation, reusing
+ * the implementation a single-key delete uses so the two cannot drift.
+ *
+ * @return what each insertion changed, for the response to write out
+ */
+ private List<DeleteMarkerInsertion> insertDeleteMarkers(
+ OzoneManager ozoneManager, OMMetadataManager omMetadataManager,
+ OmBucketInfo omBucketInfo, List<OmKeyInfo> omKeyInfoList,
+ List<String> markerOnlyKeys, long proposedVersionId, long trxnLogIndex)
+ throws IOException {
+
+ List<DeleteMarkerInsertion> insertions = new ArrayList<>();
+ // Applied to the bucket only once every marker is in, so that a batch
+ // failing on a later key leaves the cached bucket as it found it.
+ long pendingNamespace = 0;
+ for (OmKeyInfo currentVersion : omKeyInfoList) {
+ String objectKey = omMetadataManager.getOzoneKey(
+ currentVersion.getVolumeName(), currentVersion.getBucketName(),
+ currentVersion.getKeyName());
+ DeleteMarkerInsertion insertion = insertDeleteMarker(ozoneManager,
Review Comment:
Can we check namespace for the whole batch before caching markers? When the
second key exceeds quota, a subsequent DELETE consumes the first failed marker
and loses the original data version after flush.
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeyCommitRequest.java:
##########
@@ -303,14 +309,29 @@ public OMClientResponse
validateAndUpdateCache(OzoneManager ozoneManager, Execut
}
validateAtomicRewrite(keyToDelete, omKeyInfo, auditMap);
+ final boolean s3Versioning = omBucketInfo.isS3VersioningEnabled();
// Set the UpdateID to current transactionLogIndex
- omKeyInfo = omKeyInfo.toBuilder()
+ OmKeyInfo.Builder committedKeyBuilder = omKeyInfo.toBuilder()
.setExpectedDataGeneration(null)
.addAllMetadata(KeyValueUtil.getFromProtobuf(
commitKeyArgs.getMetadataList()))
.setUpdateID(trxnLogIndex)
- .setDataSize(commitKeyArgs.getDataSize())
- .build();
+ .setDataSize(commitKeyArgs.getDataSize());
+ if (s3Versioning) {
Review Comment:
Can the new data version clear `isDeleteMarker` while preserving the old
marker as history? A real create→commit after DELETE inherits the flag, leaving
the new payload marked as deleted.
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeyCommitRequest.java:
##########
@@ -303,14 +309,29 @@ public OMClientResponse
validateAndUpdateCache(OzoneManager ozoneManager, Execut
}
validateAtomicRewrite(keyToDelete, omKeyInfo, auditMap);
+ final boolean s3Versioning = omBucketInfo.isS3VersioningEnabled();
// Set the UpdateID to current transactionLogIndex
- omKeyInfo = omKeyInfo.toBuilder()
+ OmKeyInfo.Builder committedKeyBuilder = omKeyInfo.toBuilder()
.setExpectedDataGeneration(null)
.addAllMetadata(KeyValueUtil.getFromProtobuf(
commitKeyArgs.getMetadataList()))
.setUpdateID(trxnLogIndex)
- .setDataSize(commitKeyArgs.getDataSize())
- .build();
+ .setDataSize(commitKeyArgs.getDataSize());
+ if (s3Versioning) {
+ // The version identity is frozen when the version is created: an hsync
+ // re-commit keeps updating the same version, so it keeps its
versionId.
+ committedKeyBuilder.setVersionId(isSameHsyncKey
+ ? keyToDelete.getVersionId()
Review Comment:
Can this preserve a legacy null version ID without unboxing? Hsyncing an
unversioned key, enabling versioning, then hsyncing again currently throws an
uncaught NPE here.
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeysDeleteRequest.java:
##########
@@ -259,9 +314,17 @@ public OMClientResponse
validateAndUpdateCache(OzoneManager ozoneManager, Execut
}
final long volumeId = omMetadataManager.getVolumeId(volumeName);
- omClientResponse =
- getOmClientResponse(ozoneManager, omKeyInfoList, dirList, omResponse,
- unDeletedKeys, keyToError, deleteStatus, omBucketInfo, volumeId,
openKeyInfoMap, state);
+ if (markerInsertions != null) {
+ omResponse.setDeleteKeysResponse(DeleteKeysResponse.newBuilder()
+ .setStatus(deleteStatus).setUnDeletedKeys(unDeletedKeys))
Review Comment:
Preserve the existing per-key errors in this response. A mixed batch leaves
a denied key unchanged but returns an empty error map through the quiet
translator, so S3 reports it deleted.
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeyDeleteRequest.java:
##########
@@ -142,57 +154,73 @@ public OMClientResponse
validateAndUpdateCache(OzoneManager ozoneManager, Execut
OmKeyInfo omKeyInfo =
omMetadataManager.getKeyTable(getBucketLayout()).get(objectKey);
- if (omKeyInfo == null) {
- throw new OMException("Key not found", KEY_NOT_FOUND);
- }
-
- validateIfMatchETag(keyArgs, omKeyInfo);
-
- // Set the UpdateID to current transactionLogIndex
- omKeyInfo = omKeyInfo.toBuilder()
- .setUpdateID(trxnLogIndex)
- .build();
-
- // Update table cache. Put a tombstone entry
- omMetadataManager.getKeyTable(getBucketLayout()).addCacheEntry(
- new CacheKey<>(
- omMetadataManager.getOzoneKey(volumeName, bucketName, keyName)),
- CacheValue.get(trxnLogIndex));
OmBucketInfo omBucketInfo =
getBucketInfo(omMetadataManager, volumeName, bucketName);
- long quotaReleased = sumBlockLengths(omKeyInfo);
- // Empty entries won't be added to deleted table so this key shouldn't
get added to snapshotUsed space.
- boolean isKeyNonEmpty = !OmKeyInfo.isKeyEmpty(omKeyInfo);
- omBucketInfo.decrUsedBytes(quotaReleased, isKeyNonEmpty);
- omBucketInfo.decrUsedNamespace(1L, isKeyNonEmpty);
- OmKeyInfo deletedOpenKeyInfo = null;
-
- // If omKeyInfo has hsync metadata, delete its corresponding open key as
well
- String dbOpenKey = null;
- String hsyncClientId =
omKeyInfo.getMetadata().get(OzoneConsts.HSYNC_CLIENT_ID);
- if (hsyncClientId != null) {
- Table<String, OmKeyInfo> openKeyTable =
omMetadataManager.getOpenKeyTable(getBucketLayout());
- dbOpenKey = omMetadataManager.getOpenKey(volumeName, bucketName,
keyName, hsyncClientId);
- OmKeyInfo openKeyInfo = openKeyTable.get(dbOpenKey);
- if (openKeyInfo != null) {
- openKeyInfo = openKeyInfo.withMetadataMutations(
- metadata -> metadata.put(DELETED_HSYNC_KEY, "true"));
- openKeyTable.addCacheEntry(dbOpenKey, openKeyInfo, trxnLogIndex);
- deletedOpenKeyInfo = openKeyInfo;
- } else {
- LOG.warn("Potentially inconsistent DB state: open key not found with
dbOpenKey '{}'", dbOpenKey);
+ if (omBucketInfo.isS3VersioningEnabled()) {
+ // A delete without a versionId removes no data: a delete marker
+ // becomes the current version and the version it supersedes moves to
+ // the versionedKeyTable. Like S3, the marker is inserted even when the
+ // key does not exist.
+ insertingDeleteMarker = true;
+ omClientResponse = insertDeleteMarker(ozoneManager, omMetadataManager,
Review Comment:
Can both DELETE paths share the existing If-Match check? A stale ETag on an
enabled bucket succeeds and replaces the object with a marker.
##########
hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeysDeleteRequest.java:
##########
@@ -384,6 +454,48 @@ protected OMClientResponse
getOmClientResponse(OzoneManager ozoneManager,
return omClientResponse;
}
+ /**
+ * Inserts a delete marker for every key that survived validation, reusing
+ * the implementation a single-key delete uses so the two cannot drift.
+ *
+ * @return what each insertion changed, for the response to write out
+ */
+ private List<DeleteMarkerInsertion> insertDeleteMarkers(
+ OzoneManager ozoneManager, OMMetadataManager omMetadataManager,
+ OmBucketInfo omBucketInfo, List<OmKeyInfo> omKeyInfoList,
+ List<String> markerOnlyKeys, long proposedVersionId, long trxnLogIndex)
+ throws IOException {
+
+ List<DeleteMarkerInsertion> insertions = new ArrayList<>();
+ // Applied to the bucket only once every marker is in, so that a batch
+ // failing on a later key leaves the cached bucket as it found it.
+ long pendingNamespace = 0;
+ for (OmKeyInfo currentVersion : omKeyInfoList) {
+ String objectKey = omMetadataManager.getOzoneKey(
+ currentVersion.getVolumeName(), currentVersion.getBucketName(),
+ currentVersion.getKeyName());
+ DeleteMarkerInsertion insertion = insertDeleteMarker(ozoneManager,
+ omMetadataManager, omBucketInfo, currentVersion, objectKey,
+ currentVersion.getKeyName(), proposedVersionId,
+ Time.now(), trxnLogIndex, pendingNamespace);
Review Comment:
Stamp and replicate one batch timestamp in `preExecute`, as single DELETE
does. Applying the identical request to two databases with different clocks
currently persists different marker timestamps.
--
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]