client_guide.md

layout	displayTitle	title	description	license
page	Uniffle Shuffle Client Guide	Uniffle Shuffle Client Guide	Uniffle Shuffle Client Guide	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.

Uniffle Shuffle Client Guide

Uniffle is designed as a unified shuffle engine for multiple computing frameworks, including Apache Spark, Apache Hadoop and Apache Tez. Uniffle has provided pluggable client plugins to enable remote shuffle in Spark, MapReduce and Tez.

Deploy & client specific configuration

Refer to the following documents on how to deploy Uniffle client plugins with Spark, MapReduce and Tez. Client specific configurations are also listed in each document.

Client	Link
Spark	Deploy Spark Client Plugin & Configurations
MapReduce	Deploy MapReduce Client Plugin & Configurations
Tez	Deploy Tez Client Plugin & Configurations

Common Configuration

The important configuration of client is listed as following. These configurations are shared by all types of clients.

Property Name	Default	Description
<client_type>.rss.coordinator.quorum	-	Coordinator quorum
<client_type>.rss.writer.buffer.size	3m	Buffer size for single partition data
<client_type>.rss.storage.type	-	Supports MEMORY_LOCALFILE, MEMORY_HDFS, MEMORY_LOCALFILE_HDFS
<client_type>.rss.client.read.buffer.size	14m	The max data size read from storage
<client_type>.rss.client.send.threadPool.size	5	The thread size for send shuffle data to shuffle server
<client_type>.rss.client.assignment.tags	-	The comma-separated list of tags for deciding assignment shuffle servers. Notice that the SHUFFLE_SERVER_VERSION will always as the assignment tag whether this conf is set or not
<client_type>.rss.client.data.commit.pool.size	The number of assigned shuffle servers	The thread size for sending commit to shuffle servers
<client_type>.rss.client.assignment.shuffle.nodes.max	-1	The number of required assignment shuffle servers. If it is less than 0 or equals to 0 or greater than the coordinator's config of "rss.coordinator.shuffle.nodes.max", it will use the size of "rss.coordinator.shuffle.nodes.max" default
<client_type>.rss.client.io.compression.codec	lz4	The compression codec is used to compress the shuffle data. Default codec is lz4. Other options areZSTD and SNAPPY.
<client_type>.rss.client.io.compression.zstd.level	3	The zstd compression level, the default level is 3
<client_type>.rss.client.shuffle.data.distribution.type	NORMAL	The type of partition shuffle data distribution, including normal and local_order. The default value is normal. Now this config is only valid in Spark3.x
<client_type>.rss.estimate.task.concurrency.dynamic.factor	1.0	Between 0 and 1, used to estimate task concurrency, when the client is spark, it represents how likely is this part of the resource between spark.dynamicAllocation.minExecutors and spark.dynamicAllocation.maxExecutors to be allocated, when the client is mr, it represents how likely the resources of map and reduce are satisfied. Effective when <client_type>.rss.estimate.server.assignment.enabled=true or Coordinator's rss.coordinator.select.partition.strategy is CONTINUOUS.
<client_type>.rss.estimate.server.assignment.enabled	false	Support mr and spark, whether to enable estimation of the number of ShuffleServers that need to be allocated based on the number of concurrent tasks.
<client_type>.rss.estimate.task.concurrency.per.server	80	It takes effect when rss.estimate.server.assignment.enabled=true, how many tasks are concurrently assigned to a ShuffleServer.
<client_type>.rss.client.max.concurrency.of.per-partition.write	-	The maximum number of files that can be written concurrently to a single partition is determined. This value will only be respected by the remote shuffle server if it is greater than 0.
<client_type>.rss.client.rpc.timeout.ms	60000	Timeout in milliseconds for RPC calls.
<client_type>.rss.client.rpc.maxAttempts	3	When we fail to send RPC calls, we will retry for maxAttempts times.
<client_type>.rss.client.rpc.netty.pageSize	4096	The value of pageSize for PooledByteBufAllocator when using gRPC internal Netty on the client-side. This configuration will only take effect when rss.rpc.server.type is set to GRPC_NETTY.
<client_type>.rss.client.rpc.netty.maxOrder	3	The value of maxOrder for PooledByteBufAllocator when using gRPC internal Netty on the client-side. This configuration will only take effect when rss.rpc.server.type is set to GRPC_NETTY.
<client_type>.rss.client.rpc.netty.smallCacheSize	1024	The value of smallCacheSize for PooledByteBufAllocator when using gRPC internal Netty on the client-side. This configuration will only take effect when rss.rpc.server.type is set to GRPC_NETTY.

Notice:

1. <client_type> should be mapreduce tez or spark

2. <client_type>.rss.coordinator.quorum is compulsory, and other configurations are optional when coordinator dynamic configuration is enabled.

Client Quorum Setting

Uniffle supports client-side quorum protocol to tolerant shuffle server crash. This feature is client-side behaviour, in which shuffle writer sends each block to multiple servers, and shuffle readers could fetch block data from one of server. Since sending multiple replicas of blocks can reduce the shuffle performance and resource consumption, we designed it as an optional feature.

Property Name	Default	Description
<client_type>.rss.data.replica	1	The max server number that each block can be send by client in quorum protocol
<client_type>.rss.data.replica.write	1	The min server number that each block should be send by client successfully
<client_type>.rss.data.replica.read	1	The min server number that metadata should be fetched by client successfully

Notice:

1. spark.rss.data.replica.write + spark.rss.data.replica.read > spark.rss.data.replica

Recommended examples:

1. Performance First (default)

spark.rss.data.replica 1
spark.rss.data.replica.write 1
spark.rss.data.replica.read 1

2. Fault-tolerant First

spark.rss.data.replica 3
spark.rss.data.replica.write 2
spark.rss.data.replica.read 2

Netty Setting

Property Name	Default	Description
<client_type>.rss.client.type	GRPC	Supports GRPC_NETTY, GRPC. The default value is GRPC. But we recommend using GRPC_NETTY to enable Netty on the client side for better stability and performance.
<client_type>.rss.client.netty.io.mode	NIO	Netty EventLoopGroup backend, available options: NIO, EPOLL.
<client_type>.rss.client.netty.client.connection.timeout.ms	600000	Connection active timeout.
<client_type>.rss.client.netty.client.threads	0	Number of threads used in the client thread pool. Default is 0, Netty will use the number of (available logical cores * 2) as the number of threads.
<client_type>.rss.client.netty.client.prefer.direct.bufs	true	If true, we will prefer allocating off-heap byte buffers within Netty.
<client_type>.rss.client.netty.client.pooled.allocator.enabled	true	If true, we will use PooledByteBufAllocator to allocate byte buffers within Netty, otherwise we'll use UnpooledByteBufAllocator.
<client_type>.rss.client.netty.client.shared.allocator.enabled	true	A flag indicating whether to share the ByteBuf allocators between the different Netty channels when enabling Netty. If enabled then only three ByteBuf allocators are created: one PooledByteBufAllocator where caching is allowed, one PooledByteBufAllocator where not and one UnpooledByteBufAllocator. When disabled, a new allocator is created for each transport client.
<client_type>.rss.client.netty.client.connections.per.peer	2	Suppose there are 100 executors, spark.rss.client.netty.client.connections.per.peer = 2, then each ShuffleServer will establish a total of (100 * 2) connections with multiple clients.
<client_type>.rss.client.netty.client.receive.buffer	0	Receive buffer size (SO_RCVBUF). Note: the optimal size for receive buffer and send buffer should be latency * network_bandwidth. Assuming latency = 1ms, network_bandwidth = 10Gbps, buffer size should be ~ 1.25MB. Default is 0, the operating system automatically estimates the receive buffer size based on default settings.
<client_type>.rss.client.netty.client.send.buffer	0	Send buffer size (SO_SNDBUF).