Jackie-Jiang commented on code in PR #8663:
URL: https://github.com/apache/pinot/pull/8663#discussion_r877636573
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/realtime/PinotLLCRealtimeSegmentManager.java:
##########
@@ -1385,4 +1386,89 @@ public void uploadToDeepStoreIfMissing(TableConfig
tableConfig, List<SegmentZKMe
}
}
}
+
+ /**
+ * Creates PROPERTYSTORE and IDEALSTATE entries for each partitionGroup to
resume realtime table consumption
+ * @param tableName
+ */
+ public void resumeRealtimeTableConsumption(String tableName) {
+ TableConfig tableConfig = _helixResourceManager.getTableConfig(tableName,
TableType.REALTIME);
+ String realtimeTableName =
TableNameBuilder.REALTIME.tableNameWithType(tableName);
+
+ InstancePartitions instancePartitions =
+
InstancePartitionsUtils.fetchOrComputeInstancePartitions(_helixResourceManager.getHelixZkManager(),
tableConfig,
+ InstancePartitionsType.CONSUMING);
+ Map<InstancePartitionsType, InstancePartitions> instancePartitionsMap =
+ Collections.singletonMap(InstancePartitionsType.CONSUMING,
instancePartitions);
+ IdealState idealState =
+
HelixHelper.getTableIdealState(_helixResourceManager.getHelixZkManager(),
realtimeTableName);
+ int numReplicas = getNumReplicas(tableConfig, instancePartitions);
+ Map<String, Map<String, String>> instanceStatesMap =
idealState.getRecord().getMapFields();
+
+ PartitionLevelStreamConfig streamConfig = new
PartitionLevelStreamConfig(tableConfig.getTableName(),
+ IngestionConfigUtils.getStreamConfigMap(tableConfig));
+ List<PartitionGroupConsumptionStatus>
currentPartitionGroupConsumptionStatusList =
+ getPartitionGroupConsumptionStatusList(idealState, streamConfig);
+ // Read the smallest offset when a new partition is detected
+ OffsetCriteria originalOffsetCriteria = streamConfig.getOffsetCriteria();
+ streamConfig.setOffsetCriteria(OffsetCriteria.SMALLEST_OFFSET_CRITERIA);
+ List<PartitionGroupMetadata> newPartitionGroupMetadataList =
+ getNewPartitionGroupMetadataList(streamConfig,
currentPartitionGroupConsumptionStatusList);
+ streamConfig.setOffsetCriteria(originalOffsetCriteria);
+
+ int numPartitionGroups = newPartitionGroupMetadataList.size();
+ Map<Integer, SegmentZKMetadata> partitionToLatestSegment =
getLatestSegmentZKMetadataMap(realtimeTableName);
+ for (PartitionGroupMetadata partitionGroupMetadata :
newPartitionGroupMetadataList) {
+ int partitionId = partitionGroupMetadata.getPartitionGroupId();
+ if (partitionToLatestSegment.containsKey(partitionId)
+ &&
idealState.getInstanceStateMap(partitionToLatestSegment.get(partitionId).getSegmentName())
+ .containsValue(SegmentStateModel.CONSUMING)) {
+ // IN_PROGRESS segment already exists, NO-OP
+ LOGGER.info("Skipping partitionGroupId {}, IN_PROGRESS segment already
exists", partitionId);
+ } else {
+ // Step 1: Create PROPERTYSTORE nodes for each partitionId
(IN_PROGRESS)
+ String offset = partitionGroupMetadata.getStartOffset().toString();
+ long newSegmentCreationTimeMs = System.currentTimeMillis();
+
+ SegmentZKMetadata latestSegmentMeta =
partitionToLatestSegment.get(partitionId);
+ int seqNumber = latestSegmentMeta == null ? STARTING_SEQUENCE_NUMBER
+ : new
LLCSegmentName(latestSegmentMeta.getSegmentName()).getSequenceNumber() + 1;
+
+ LLCSegmentName newLLCSegment = new LLCSegmentName(tableName,
partitionId, seqNumber, newSegmentCreationTimeMs);
+ SegmentZKMetadata newSegmentZKMetadata = new
SegmentZKMetadata(newLLCSegment.getSegmentName());
+
newSegmentZKMetadata.setCreationTime(newLLCSegment.getCreationTimeMs());
+ newSegmentZKMetadata.setStartOffset(offset);
+ newSegmentZKMetadata.setNumReplicas(numReplicas);
+
newSegmentZKMetadata.setStatus(CommonConstants.Segment.Realtime.Status.IN_PROGRESS);
+ SegmentPartitionMetadata segmentPartitionMetadata =
+ getPartitionMetadataFromTableConfig(tableConfig, partitionId);
+ if (segmentPartitionMetadata != null) {
+ newSegmentZKMetadata.setPartitionMetadata(segmentPartitionMetadata);
+ }
+
+ FlushThresholdUpdater flushThresholdUpdater =
_flushThresholdUpdateManager
+ .getFlushThresholdUpdater(streamConfig);
+ CommittingSegmentDescriptor committingSegmentDescriptor = new
CommittingSegmentDescriptor(null, offset, 0);
+ flushThresholdUpdater.updateFlushThreshold(streamConfig,
+ newSegmentZKMetadata, committingSegmentDescriptor, null,
+ getMaxNumPartitionsPerInstance(instancePartitions,
numPartitionGroups, numReplicas),
+ newPartitionGroupMetadataList);
+
+ _helixResourceManager.getPropertyStore().set(
Review Comment:
Good point. So it can potentially lead to race condition if 2 runs happen at
exactly the same time, and both create the new segment ZK metadata. One of the
segment ZK metadata will be picked and continue, the other will be left orphan.
@saurabhd336 Let's change `ensureAllPartitionsConsuming` to `synchronized`,
and add a TODO describing this potential race condition. We should probably
clean up the orphan segment ZK metadata if it is not in the ideal state. This
can be handled in a separate PR.
--
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]
