prashantwason commented on code in PR #8837: URL: https://github.com/apache/hudi/pull/8837#discussion_r1266694440
########## hudi-client/hudi-client-common/src/main/java/org/apache/hudi/metadata/HoodieBackedTableMetadataWriter.java: ########## @@ -1019,14 +1016,23 @@ public void update(HoodieRollbackMetadata rollbackMetadata, String instantTime) } } + // lets apply a delta commit with DT's rb instant containing following records: + // a. any log files as part of RB commit metadata that was added + // b. log files added by the commit in DT being rolled back. By rolled back, we mean, a rollback block will be added and does not mean it will be deleted. + // both above list should only be added to FILES partition. + + String rollbackInstantTime = createRollbackTimestamp(instantTime); + processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(engineContext, metadataMetaClient.getActiveTimeline(), + dataMetaClient, rollbackMetadata, rollbackInstantTime)); + if (deltacommitsSinceCompaction.containsInstant(deltaCommitInstant)) { LOG.info("Rolling back MDT deltacommit " + commitInstantTime); - if (!getWriteClient().rollback(commitInstantTime, instantTime)) { + if (!getWriteClient().rollback(commitInstantTime, rollbackInstantTime)) { throw new HoodieMetadataException("Failed to rollback deltacommit at " + commitInstantTime); } } else { LOG.info(String.format("Ignoring rollback of instant %s at %s since there are no corresponding deltacommits on MDT", - commitInstantTime, instantTime)); + commitInstantTime, rollbackInstantTime)); Review Comment: I think this should still be instantTime for the log message to make sense. Ignoring rollback of instant %s at %s. The last %s is supposed to be the instantTime at which the rollback happened. rollbackInstantTime is the "created" instantTime for MDT and not the instantTime on the DT. ########## hudi-common/src/main/java/org/apache/hudi/metadata/HoodieTableMetadataUtil.java: ########## @@ -653,37 +608,57 @@ private static void processRestoreMetadata(HoodieActiveTimeline metadataTableTim */ public static Map<MetadataPartitionType, HoodieData<HoodieRecord>> convertMetadataToRecords( HoodieEngineContext engineContext, HoodieActiveTimeline metadataTableTimeline, - HoodieRollbackMetadata rollbackMetadata, MetadataRecordsGenerationParams recordsGenerationParams, - String instantTime, Option<String> lastSyncTs, boolean wasSynced) { + HoodieTableMetaClient dataTableMetaClient, + HoodieRollbackMetadata rollbackMetadata, String instantTime) { final Map<MetadataPartitionType, HoodieData<HoodieRecord>> partitionToRecordsMap = new HashMap<>(); - Map<String, List<String>> partitionToDeletedFiles = new HashMap<>(); - Map<String, Map<String, Long>> partitionToAppendedFiles = new HashMap<>(); - List<HoodieRecord> filesPartitionRecords = - convertMetadataToRollbackRecords(metadataTableTimeline, rollbackMetadata, partitionToDeletedFiles, partitionToAppendedFiles, instantTime, lastSyncTs, wasSynced); + List<HoodieRecord> filesPartitionRecords = convertMetadataToRollbackRecords(rollbackMetadata, instantTime); + + List<HoodieRecord> reAddedRecords = getHoodieRecordsForLogFilesFromRollbackPlan(dataTableMetaClient, instantTime); + filesPartitionRecords.addAll(reAddedRecords); final HoodieData<HoodieRecord> rollbackRecordsRDD = engineContext.parallelize(filesPartitionRecords, 1); + partitionToRecordsMap.put(MetadataPartitionType.FILES, rollbackRecordsRDD); return partitionToRecordsMap; } + private static List<HoodieRecord> getHoodieRecordsForLogFilesFromRollbackPlan(HoodieTableMetaClient dataTableMetaClient, String instantTime) { + /*List<HoodieInstant> instants = dataTableMetaClient.reloadActiveTimeline().filterRequestedRollbackTimeline() + .filter(instant -> instant.getTimestamp().equals(instantTime) && instant.isRequested()).getInstants();*/ + + HoodieInstant rollbackInstant = new HoodieInstant(HoodieInstant.State.REQUESTED, HoodieTimeline.ROLLBACK_ACTION, instantTime); + + // HoodieInstant rollbackInstant = instants.get(0); + HoodieInstant requested = HoodieTimeline.getRollbackRequestedInstant(rollbackInstant); + try { + HoodieRollbackPlan rollbackPlan = TimelineMetadataUtils.deserializeAvroMetadata( + dataTableMetaClient.getActiveTimeline().readRollbackInfoAsBytes(requested).get(), HoodieRollbackPlan.class); + + Map<String, Map<String, Long>> partitionToLogFilesMap = new HashMap<>(); + + rollbackPlan.getRollbackRequests().forEach(rollbackRequest -> { + partitionToLogFilesMap.computeIfAbsent(rollbackRequest.getPartitionPath(), s -> new HashMap<>()); + // fetch only log files that are expected to be RB'd in DT as part of this rollback. these log files will not be deleted, but rendered + // invalid once rollback is complete. + partitionToLogFilesMap.get(rollbackRequest.getPartitionPath()).putAll(rollbackRequest.getLogBlocksToBeDeleted()); Review Comment: can save an additional get() call here by chaining the putAll to the output of computeIfAbsent ########## hudi-common/src/main/java/org/apache/hudi/metadata/HoodieTableMetadataUtil.java: ########## @@ -653,37 +608,57 @@ private static void processRestoreMetadata(HoodieActiveTimeline metadataTableTim */ public static Map<MetadataPartitionType, HoodieData<HoodieRecord>> convertMetadataToRecords( HoodieEngineContext engineContext, HoodieActiveTimeline metadataTableTimeline, - HoodieRollbackMetadata rollbackMetadata, MetadataRecordsGenerationParams recordsGenerationParams, - String instantTime, Option<String> lastSyncTs, boolean wasSynced) { + HoodieTableMetaClient dataTableMetaClient, + HoodieRollbackMetadata rollbackMetadata, String instantTime) { final Map<MetadataPartitionType, HoodieData<HoodieRecord>> partitionToRecordsMap = new HashMap<>(); - Map<String, List<String>> partitionToDeletedFiles = new HashMap<>(); - Map<String, Map<String, Long>> partitionToAppendedFiles = new HashMap<>(); - List<HoodieRecord> filesPartitionRecords = - convertMetadataToRollbackRecords(metadataTableTimeline, rollbackMetadata, partitionToDeletedFiles, partitionToAppendedFiles, instantTime, lastSyncTs, wasSynced); + List<HoodieRecord> filesPartitionRecords = convertMetadataToRollbackRecords(rollbackMetadata, instantTime); + + List<HoodieRecord> reAddedRecords = getHoodieRecordsForLogFilesFromRollbackPlan(dataTableMetaClient, instantTime); + filesPartitionRecords.addAll(reAddedRecords); final HoodieData<HoodieRecord> rollbackRecordsRDD = engineContext.parallelize(filesPartitionRecords, 1); + partitionToRecordsMap.put(MetadataPartitionType.FILES, rollbackRecordsRDD); return partitionToRecordsMap; Review Comment: Collections.singletonMap? -- 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: commits-unsubscr...@hudi.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org