This is an automated email from the ASF dual-hosted git repository. guoyangze pushed a commit to branch master in repository https://gitbox.apache.org/repos/asf/flink.git
commit 817e3f2b964fa5d86e207d6aa3065f139ec84402 Author: Xiangyu Feng <xiangyu...@gmail.com> AuthorDate: Wed Nov 1 19:16:08 2023 +0800 [FLINK-33235][doc] Update OLAP Quickstart doc --- docs/content/docs/dev/table/olap_quickstart.md | 181 +++++++++++++------------ docs/content/docs/dev/table/overview.md | 2 +- docs/static/fig/olap-architecture.svg | 21 +++ 3 files changed, 117 insertions(+), 87 deletions(-) diff --git a/docs/content/docs/dev/table/olap_quickstart.md b/docs/content/docs/dev/table/olap_quickstart.md index e0b3afba2bc..e5084e065c2 100644 --- a/docs/content/docs/dev/table/olap_quickstart.md +++ b/docs/content/docs/dev/table/olap_quickstart.md @@ -1,5 +1,5 @@ --- -title: "Quickstart for Flink OLAP" +title: "OLAP Quickstart" weight: 91 type: docs aliases: @@ -24,32 +24,41 @@ specific language governing permissions and limitations under the License. --> -# Introduction +# OLAP Quickstart -Flink OLAP has already added to [Apache Flink Roadmap](https://flink.apache.org/roadmap/). It means Flink can not only support streaming and batch computing, but also support OLAP(On-Line Analytical Processing). This page will show how to quickly set up a Flink OLAP service, and will introduce some best practices. +OLAP (OnLine Analysis Processing) is a key technology in the field of data analysis, it is generally used to perform complex queries on large data sets with latencies in seconds. Now Flink can not only support streaming and batch computing, but also supports users to deploy it as an OLAP computing service. This page will show you how to quickly set up a local Flink OLAP service, and will also introduce some best practices helping you deploy Flink OLAP service in production. -## Architecture +## Architecture Introduction +This chapter will introduce you to the overall architecture of Flink OLAP service and the advantages of using it. -The Flink OLAP service consists of three parts: Client, Flink SQL Gateway, Flink Session Cluster. +### Architecture -* **Client**: Could be any client that can interact with Flink SQL Gateway, such as SQL client, Flink JDBC driver and so on. -* **Flink SQL Gateway**: The SQL Gateway provides an easy way to submit the Flink Job, look up the metadata, and analyze table stats. -* **Flink Session Cluster**: We choose session clusters to run OLAP queries, mainly to avoid the overhead of cluster startup. +Flink OLAP service consists of three parts: Client, Flink SQL Gateway and Flink Session Cluster. -## Advantage +* **Client**: Could be any client that can interact with [Flink SQL Gateway]({{< ref "docs/dev/table/sql-gateway/overview" >}}), such as [SQL Client]({{< ref "docs/dev/table/sqlClient" >}}), [Flink JDBC Driver]({{< ref "docs/dev/table/jdbcDriver" >}}) and so on. +* **Flink SQL Gateway**: The SQL Gateway provides an easy way to parse the sql query, look up the metadata, analyze table stats, optimize the plan and submit JobGraphs to cluster. +* **Flink Session Cluster**: OLAP queries run on [session cluster]({{< ref "/docs/deployment/resource-providers/native_kubernetes#starting-a-flink-session-on-kubernetes" >}}), mainly to avoid the overhead of cluster startup. + +{{< img src="/fig/olap-architecture.svg" alt="Illustration of Flink OLAP Architecture" width="85%" >}} + +### Advantage * **Massively Parallel Processing** - * Flink OLAP runs naturally as an MPP(Massively Parallel Processing) system, which supports low-latency ad-hoc queries + * Flink OLAP runs naturally as a massively parallel processing system, which enables planners to easily adjust the job parallelism to fulfill queries' latency requirement under different data sizes. +* **Elastic Resource Management** + * Flink's resource management supports min/max scaling, which means the session cluster can allocate the resource according to workload dynamically. * **Reuse Connectors** - * Flink OLAP can reuse rich connectors in Flink ecosystem. + * Flink OLAP can reuse the rich [Connectors]({{< ref "docs/connectors/table/overview" >}}) in Flink ecosystem. * **Unified Engine** * Unified computing engine for Streaming/Batch/OLAP. -# Deploying in Local Mode +## Deploying in Local Mode + +In this chapter, you will learn how to build Flink OLAP services locally. -## Downloading Flink +### Downloading Flink -The same as [Local Installation]({{< ref "docs/try-flink/local_installation" >}}). Flink runs on all UNIX-like environments, i.e. Linux, Mac OS X, and Cygwin (for Windows). We need to have at least Java 11 installed, Java 17 is more recommended in OLAP scenario. To check the Java version installed, type in your terminal: +The same as [Local Installation]({{< ref "docs/try-flink/local_installation" >}}). Flink runs on all UNIX-like environments, i.e. Linux, Mac OS X, and Cygwin (for Windows). User need to have at __Java 11__ installed. To check the Java version installed, user can type in the terminal: ``` java -version @@ -61,7 +70,7 @@ Next, [Download](https://flink.apache.org/downloads/) the latest binary release tar -xzf flink-*.tgz ``` -## Starting a local cluster +### Starting a local cluster To start a local cluster, run the bash script that comes with Flink: @@ -69,9 +78,9 @@ To start a local cluster, run the bash script that comes with Flink: ./bin/start-cluster.sh ``` -You should be able to navigate to the web UI at localhost:8081 to view the Flink dashboard and see that the cluster is up and running. +You should be able to navigate to the web UI at http://localhost:8081 to view the Flink dashboard and see that the cluster is up and running. -## Start a SQL Client CLI +### Start a SQL Client CLI You can start the CLI with an embedded gateway by calling: @@ -79,7 +88,7 @@ You can start the CLI with an embedded gateway by calling: ./bin/sql-client.sh ``` -## Running Queries +### Running Queries You could simply execute queries in CLI and retrieve the results. @@ -102,98 +111,98 @@ GROUP BY buyer ORDER BY total_cost LIMIT 3; ``` -And then you could find job detail information in web UI at localhost:8081. +And then you could find job detail information in web UI at http://localhost:8081. -# Deploying in Production +## Deploying in Production This section guides you through setting up a production ready Flink OLAP service. -## Cluster Deployment +### Client -In production, we recommend to use Flink Session Cluster, Flink SQL Gateway and Flink JDBC Driver to build an OLAP service. +#### Flink JDBC Driver -### Session Cluster +You should use Flink JDBC Driver when submitting queries to SQL Gateway since it provides low-level connection management. When used in production, you should pay attention to reuse the JDBC connection to avoid frequently creating/closing sessions in the Gateway and then reduce the E2E query latency. For detailed information, please refer to the [Flink JDBC Driver]({{ <ref "docs/dev/table/jdbcDriver"> }}). -For Flink Session Cluster, we recommend to deploy Flink on native Kubernetes using session mode. Kubernetes is a popular container-orchestration system for automating computer application deployment, scaling, and management. By deploying on native Kubernetes, Flink Session Cluster is able to dynamically allocate and de-allocate TaskManagers. For more information, please refer to [Native Kubernetes]({{< ref "docs/deployment/resource-providers/native_kubernetes">}}). +### Cluster Deployment -### SQL Gateway +In production, you should use Flink Session Cluster, Flink SQL Gateway to build an OLAP service. -For Flink SQL Gateway, we recommend deploying it as a stateless microservice and register this on the service discovery component. For more information, please refer to the [SQL Gateway Overview]({{< ref "docs/dev/table/sql-gateway/overview">}}). +#### Session Cluster -### Flink JDBC Driver +For Flink Session Cluster, you can deploy it on Native Kubernetes using session mode. Kubernetes is a popular container-orchestration system for automating computer application deployment, scaling, and management. By deploying on Native Kubernetes, Flink Session Cluster is able to dynamically allocate and de-allocate TaskManagers. For more information, please refer to [Native Kubernetes]({{ < ref "docs/deployment/resource-providers/native_kubernetes"> }}). Furthermore, you can config the [...] -When submitting queries to SQL Gateway, we recommend using Flink JDBC Driver since it provides low-level connection management. When used in production, we need to pay attention to reuse the JDBC connection to avoid frequently creating/closing sessions in the Gateway. For more information, please refer to the [Flink JDBC Driver]({{{<ref "docs/dev/table/jdbcDriver">}}}). +#### SQL Gateway -## Datasource Configurations +For Flink SQL Gateway, you should deploy it as a stateless microservice and register the instance on service discovery component. Through this way, client can balance the query between instances easily. For more information, please refer to [SQL Gateway Overview]({{< ref "docs/dev/table/sql-gateway/overview">}}). -### Catalogs +### Datasource Configurations -In OLAP scenario, we recommend using FileCatalogStore in the catalog configuration introduced in [FLIP-295](https://cwiki.apache.org/confluence/display/FLINK/FLIP-295%3A+Support+lazy+initialization+of+catalogs+and+persistence+of+catalog+configurations). As a long running service, Flink OLAP cluster's catalog information will not change frequently and can be re-used cross sessions. For more information, please refer to the [Catalog Store]({{< ref "docs/dev/table/catalogs#catalog-store">}}). +#### Catalogs -### Connectors +In OLAP scenario, you should configure `FileCatalogStore` provided by [Catalogs]({{< ref "docs/dev/table/catalogs">}}) as the catalog used by cluster. As a long-running service, Flink OLAP cluster's catalog information will not change frequently and should be re-used cross sessions to reduce the cold-start cost. For more information, please refer to the [Catalog Store]({{< ref "docs/dev/table/catalogs#catalog-store">}}). -Both Session Cluster and SQL Gateway rely on connectors to analyze table stats and read data from the configured data source. To add connectors, please refer to the [Connectors and Formats]({{< ref "docs/connectors/table/overview">}}). +#### Connectors -## Cluster Configurations +Both Session Cluster and SQL Gateway rely on connectors to analyze table stats and read data from the configured data source. To add connectors, please refer to the [Connectors]({{< ref "docs/connectors/table/overview">}}). -In OLAP scenario, we picked out a few configurations that can help improve user usability and query performance. +### Recommended Cluster Configurations -### SQL&Table Options +In OLAP scenario, appropriate configurations that can greatly help users improve the overall usability and query performance. Here are some recommended production configurations: -| Parameters | Default | Recommended | -|:-------------------------------------|:--------|:------------| -| table.optimizer.join-reorder-enabled | false | true | -| pipeline.object-reuse | false | true | +#### SQL&Table Options -### Runtime Options +| Parameters | Default | Recommended | +|:---------------------------------------------------------------------------------------------------------------|:--------|:------------| +| [table.optimizer.join-reorder-enabled]({{<ref "docs/dev/table/config#table-optimizer-join-reorder-enabled">}}) | false | true | +| [pipeline.object-reuse]({{< ref "docs/deployment/config#pipeline-object-reuse" >}}) | false | true | -| Parameters | Default | Recommended | -|:-----------------------------|:-----------------------|:------------------------------------------------------------------------------------------------------------------------------------------| -| execution.runtime-mode | STREAMING | BATCH | -| execution.batch-shuffle-mode | ALL_EXCHANGES_BLOCKING | ALL_EXCHANGES_PIPELINED | -| env.java.opts.all | {default value} | {default value} -XX:PerMethodRecompilationCutoff=10000 -XX:PerBytecodeRecompilationCutoff=10000-XX:ReservedCodeCacheSize=512M -XX:+UseZGC | -| JDK Version | 11 | 17 | +#### Runtime Options -We strongly recommend using JDK17 with ZGC in OLAP scenario in order to provide zero gc stw and solve the issue described in [FLINK-32746](https://issues.apache.org/jira/browse/FLINK-32746). +| Parameters | Default | Recommended | +|:--------------------------------------------------------------------------------------------------|:-----------------------|:------------------------------------------------------------------------------------------------------------------------------------------| +| [execution.runtime-mode]({{< ref "docs/deployment/config#execution-runtime-mode" >}}) | STREAMING | BATCH | +| [execution.batch-shuffle-mode]({{< ref "docs/deployment/config#execution-batch-shuffle-mode" >}}) | ALL_EXCHANGES_BLOCKING | ALL_EXCHANGES_PIPELINED | +| [env.java.opts.all]({{< ref "docs/deployment/config#env-java-opts-all" >}}) | {default value} | {default value} -XX:PerMethodRecompilationCutoff=10000 -XX:PerBytecodeRecompilationCutoff=10000-XX:ReservedCodeCacheSize=512M -XX:+UseZGC | +| JDK Version | 11 | 17 | -### Scheduling Options +Using JDK17 within ZGC can greatly help optimize the metaspace garbage collection issue, detailed information can be found in [FLINK-32746](https://issues.apache.org/jira/browse/FLINK-32746). Meanwhile, ZGC can provide close to zero application pause time when collecting garbage objects in memory. Additionally, OLAP queries need to be executed in `BATCH` mode because both `Pipelined` and `Blocking` edges may appear in the execution plan of an OLAP query. Batch scheduler allows queries to [...] + +#### Scheduling Options | Parameters | Default | Recommended | -|:---------------------------------------------------------|:------------------|:------------------| -| jobmanager.scheduler | Default | Default | -| jobmanager.execution.failover-strategy | region | full | -| restart-strategy.type | (none) | disable | -| jobstore.type | File | Memory | -| jobstore.max-capacity | Integer.MAX_VALUE | 500 | - -We would like to highlight the usage of `PipelinedRegionSchedulingStrategy`. Since many OLAP queries will have blocking edges in their jobGraph. - -### Network Options - -| Parameters | Default | Recommended | -|:------------------------------------|:-----------|:---------------| -| rest.server.numThreads | 4 | 32 | -| web.refresh-interval | 3000 | 300000 | -| pekko.framesize | 10485760b | 104857600b | - -### ResourceManager Options - -| Parameters | Default | Recommended | -|:-------------------------------------|:----------|:---------------| -| kubernetes.jobmanager.replicas | 1 | 2 | -| kubernetes.jobmanager.cpu.amount | 1.0 | 16.0 | -| jobmanager.memory.process.size | (none) | 65536m | -| jobmanager.memory.jvm-overhead.max | 1g | 6144m | -| kubernetes.taskmanager.cpu.amount | (none) | 16 | -| taskmanager.numberOfTaskSlots | 1 | 32 | -| taskmanager.memory.process.size | (none) | 65536m | -| taskmanager.memory.managed.size | (none) | 65536m | - -We prefer to use large taskManager pods in OLAP since this can put more computation in local and reduce network/deserialization/serialization overhead. Meanwhile, since JobManager is a single point of calculation in OLAP scenario, we also prefer large pod. - -# Future Work -There is a big margin for improvement in Flink OLAP, both in usability and query performance, and we trace all of them in underlying tickets. +|:------------------------------------------------------------------------------------------------------------------------|:------------------|:--------| +| [jobmanager.scheduler]({{< ref "docs/deployment/config#jobmanager-scheduler" >}}) | Default | Default | +| [jobmanager.execution.failover-strategy]({{< ref "docs/deployment/config#jobmanager-execution-failover-strategy-1" >}}) | region | full | +| [restart-strategy.type]({{< ref "docs/deployment/config#restart-strategy-type" >}}) | (none) | disable | +| [jobstore.type]({{< ref "docs/deployment/config#jobstore-type" >}}) | File | Memory | +| [jobstore.max-capacity]({{< ref "docs/deployment/config#jobstore-max-capacity" >}}) | Integer.MAX_VALUE | 500 | + + +#### Network Options + +| Parameters | Default | Recommended | +|:--------------------------------------------------------------------------------------|:-----------|:---------------| +| [rest.server.numThreads]({{< ref "docs/deployment/config#rest-server-numthreads" >}}) | 4 | 32 | +| [web.refresh-interval]({{< ref "docs/deployment/config#web-refresh-interval" >}}) | 3000 | 300000 | +| [pekko.framesize]({{< ref "docs/deployment/config#pekko-framesize" >}}) | 10485760b | 104857600b | + +#### ResourceManager Options + +| Parameters | Default | Recommended | +|:-------------------------------------------------------------------|:--------|:----------------------------------------| +| [kubernetes.jobmanager.replicas]({{< ref "docs/deployment/config#kubernetes-jobmanager-replicas" >}}) | 1 | 2 | +| [kubernetes.jobmanager.cpu.amount]({{< ref "docs/deployment/config#kubernetes-jobmanager-cpu-amount" >}}) | 1.0 | 16.0 | +| [jobmanager.memory.process.size]({{< ref "docs/deployment/config#jobmanager-memory-process-size" >}}) | (none) | 32g | +| [jobmanager.memory.jvm-overhead.max]({{< ref "docs/deployment/config#jobmanager-memory-jvm-overhead-max" >}}) | 1g | 3g | +| [kubernetes.taskmanager.cpu.amount]({{< ref "docs/deployment/config#kubernetes-taskmanager-cpu-amount" >}}) | (none) | 16 | +| [taskmanager.numberOfTaskSlots]({{< ref "docs/deployment/config#taskmanager-numberoftaskslots" >}}) | 1 | 32 | +| [taskmanager.memory.process.size]({{< ref "docs/deployment/config#taskmanager-memory-process-size" >}}) | (none) | 65536m | +| [taskmanager.memory.managed.size]({{< ref "docs/deployment/config#taskmanager-memory-managed-size" >}}) | (none) | 16384m | +| [slotmanager.number-of-slots.min]({{< ref "docs/deployment/config#slotmanager-number-of-slots-min" >}}) | 0 | {taskManagerNumber * numberOfTaskSlots} | + +You can configure `slotmanager.number-of-slots.min` to a proper value as the reserved resource pool serving OLAP queries. TaskManager should configure with a large resource specification in OLAP scenario since this can put more computations in local and reduce network/deserialization/serialization overhead. Meanwhile, as a single point of calculation in OLAP, JobManager also prefer large resource specification. + +## Future Work +Flink OLAP is now part of [Apache Flink Roadmap](https://flink.apache.org/what-is-flink/roadmap/), which means the community will keep putting efforts to improve Flink OLAP, both in usability and query performance. Relevant work are traced in underlying tickets: - https://issues.apache.org/jira/browse/FLINK-25318 -- https://issues.apache.org/jira/browse/FLINK-32898 - -Furthermore, we are adding relevant OLAP benchmarks to the Flink repository such as [flink-benchmarks](https://github.com/apache/flink-benchmarks). \ No newline at end of file +- https://issues.apache.org/jira/browse/FLINK-32898 \ No newline at end of file diff --git a/docs/content/docs/dev/table/overview.md b/docs/content/docs/dev/table/overview.md index 6715d064c2f..69bc97a1ccd 100644 --- a/docs/content/docs/dev/table/overview.md +++ b/docs/content/docs/dev/table/overview.md @@ -60,7 +60,7 @@ Where to go next? * [Built-in Functions]({{< ref "docs/dev/table/functions/systemFunctions" >}}): Supported functions in Table API and SQL. * [SQL Client]({{< ref "docs/dev/table/sqlClient" >}}): Play around with Flink SQL and submit a table program to a cluster without programming knowledge. * [SQL Gateway]({{< ref "docs/dev/table/sql-gateway/overview" >}}): A service that enables the multiple clients to execute SQL from the remote in concurrency. -* [SQL JDBC Driver]({{< ref "docs/dev/table/jdbcDriver" >}}): A JDBC Driver that submits SQL statements to sql-gateway. * [OLAP Quickstart]({{< ref "docs/dev/table/olap_quickstart" >}}): A quickstart to show how to set up a Flink OLAP service. +* [SQL JDBC Driver]({{< ref "docs/dev/table/jdbcDriver" >}}): A JDBC Driver that submits SQL statements to sql-gateway. {{< top >}} diff --git a/docs/static/fig/olap-architecture.svg b/docs/static/fig/olap-architecture.svg new file mode 100644 index 00000000000..2b39666ab86 --- /dev/null +++ b/docs/static/fig/olap-architecture.svg @@ -0,0 +1,21 @@ +<?xml version="1.0" encoding="UTF-8" standalone="no"?> +<!-- +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. +--> + +<svg version="1.1" viewBox="0.0 0.0 960.0 720.0" fill="none" stroke="none" stroke-linecap="square" stroke-miterlimit="10" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns="http://www.w3.org/2000/svg"><clipPath id="p.0"><path d="m0 0l960.0 0l0 720.0l-960.0 0l0 -720.0z" clip-rule="nonzero"/></clipPath><g clip-path="url(#p.0)"><path fill="#000000" fill-opacity="0.0" d="m0 0l960.0 0l0 720.0l-960.0 0z" fill-rule="evenodd"/><path fill="#000000" fill-opacity="0.0" d="m231.58206 85.755905l274.42 [...] \ No newline at end of file