yashmayya commented on code in PR #15930: URL: https://github.com/apache/pinot/pull/15930#discussion_r2122564948
########## pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/assignment/instance/ImplicitRealtimeTablePartitionSelector.java: ########## @@ -0,0 +1,79 @@ +/** + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ +package org.apache.pinot.controller.helix.core.assignment.instance; + +import com.google.common.annotations.VisibleForTesting; +import javax.annotation.Nullable; +import org.apache.pinot.common.assignment.InstancePartitions; +import org.apache.pinot.spi.config.table.TableConfig; +import org.apache.pinot.spi.config.table.assignment.InstanceReplicaGroupPartitionConfig; +import org.apache.pinot.spi.stream.StreamConsumerFactoryProvider; +import org.apache.pinot.spi.stream.StreamMetadataProvider; +import org.apache.pinot.spi.utils.IngestionConfigUtils; + + +/** + * Variation of {@link InstanceReplicaGroupPartitionSelector} that uses the number of partitions from the stream + * to determine the number of partitions in each replica group. + */ +public class ImplicitRealtimeTablePartitionSelector extends InstanceReplicaGroupPartitionSelector { + private final TableConfig _tableConfig; + private final int _numPartitions; + + public ImplicitRealtimeTablePartitionSelector(TableConfig tableConfig, + InstanceReplicaGroupPartitionConfig replicaGroupPartitionConfig, String tableNameWithType, + @Nullable InstancePartitions existingInstancePartitions, boolean minimizeDataMovement) { + this(tableConfig, replicaGroupPartitionConfig, tableNameWithType, existingInstancePartitions, minimizeDataMovement, + StreamConsumerFactoryProvider.create(IngestionConfigUtils.getFirstStreamConfig(tableConfig)) + .createStreamMetadataProvider( + ImplicitRealtimeTablePartitionSelector.class.getSimpleName() + "-" + tableNameWithType)); + } + + @VisibleForTesting + ImplicitRealtimeTablePartitionSelector(TableConfig tableConfig, + InstanceReplicaGroupPartitionConfig replicaGroupPartitionConfig, String tableNameWithType, + @Nullable InstancePartitions existingInstancePartitions, boolean minimizeDataMovement, + StreamMetadataProvider streamMetadataProvider) { + super(replicaGroupPartitionConfig, tableNameWithType, existingInstancePartitions, minimizeDataMovement); + _tableConfig = tableConfig; + _numPartitions = getStreamNumPartitions(streamMetadataProvider); + } + + private int getStreamNumPartitions(StreamMetadataProvider streamMetadataProvider) { + try (streamMetadataProvider) { + return streamMetadataProvider.fetchPartitionCount(10_000L); + } catch (Exception e) { + throw new RuntimeException("Failed to retrieve partition info for table: " + _tableNameWithType, e); + } + } + + @Override + protected int getNumPartitions() { + return _numPartitions; + } + + @Override + protected int getNumInstancesPerPartition(int numInstancesPerReplicaGroup) { + if (_tableConfig.isUpsertEnabled()) { + return 1; // For upsert enabled tables, we enforce one instance per partition Review Comment: > We usually try to treat COMPLETED like OFFLINE though, so I'm just trying to think about whether it makes sense to fetch the numPartitions from the stream or not? I guess it should be fine, if a user wants to use this with this new instance selector, they can configure it to do so. Even for `COMPLETED` segments, wouldn't having the number of partitions in the replica group align with the stream / segment partitions be a good thing? This would have the same effect of minimizing partition and data movement across instances during repartitioning and scale ups / scale downs when using the minimize data movement algorithm. > The special handling for keeping one server per partition only applies to CONSUMING instance partitions, so we don't really need this for non-consuming instance partitions I'm not really sure I follow why this partition selector only applies to `CONSUMING` instance partitions? Why shouldn't users be able to configure it for `COMPLETED` instance partitions in a non-upsert RT table and get the benefit of aligning their RG partitions with the stream / segment partitions? -- 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...@pinot.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org For additional commands, e-mail: commits-h...@pinot.apache.org
