voonhous commented on code in PR #19674: URL: https://github.com/apache/hudi/pull/19674#discussion_r4230494522
########## packaging/hudi-spark-native-bundle/pom.xml: ########## @@ -0,0 +1,505 @@ +<?xml version="1.0" encoding="UTF-8"?> +<!-- + 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. +--> +<project xmlns="http://maven.apache.org/POM/4.0.0"; xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"; xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd";> + <parent> + <artifactId>hudi</artifactId> + <groupId>org.apache.hudi</groupId> + <version>1.3.0-SNAPSHOT</version> + <relativePath>../../pom.xml</relativePath> + </parent> + <modelVersion>4.0.0</modelVersion> + <artifactId>hudi-spark${sparkbundle.version}-native-bundle_${scala.binary.version}</artifactId> + <packaging>jar</packaging> + + <properties> + <checkstyle.skip>true</checkstyle.skip> + <main.basedir>${project.parent.basedir}</main.basedir> + <skipTests>true</skipTests> + <javax.servlet.version>3.1.0</javax.servlet.version> + <!-- Apache DataFusion Comet releases one artifact per Spark minor version, and does not + cover every Spark version Hudi builds against. Default to producing nothing, so that a + build for an uncovered Spark version yields no bundle rather than one named "native" that + carries no native code. A sources jar is still produced, but nothing is installed or + deployed. The comet-spark<version> profiles below opt in. --> + <hudi.spark.native.bundle.skip>true</hudi.spark.native.bundle.skip> + <hudi.spark.native.bundle.jar.phase>none</hudi.spark.native.bundle.jar.phase> + <maven.install.skip>${hudi.spark.native.bundle.skip}</maven.install.skip> + <maven.deploy.skip>${hudi.spark.native.bundle.skip}</maven.deploy.skip> + </properties> + + <build> + <plugins> + <plugin> + <groupId>org.apache.rat</groupId> + <artifactId>apache-rat-plugin</artifactId> + </plugin> + <plugin> + <groupId>org.apache.maven.plugins</groupId> + <artifactId>maven-jar-plugin</artifactId> + <executions> + <execution> + <id>default-jar</id> + <!-- none when this module produces no bundle. Without it maven-jar-plugin still emits + a jar named as if it were a bundle, holding nothing but the generated META-INF. + skipIfEmpty does not help: remote-resources populates target/classes regardless. --> + <phase>${hudi.spark.native.bundle.jar.phase}</phase> + </execution> + </executions> + </plugin> + <plugin> + <groupId>org.apache.maven.plugins</groupId> + <artifactId>maven-shade-plugin</artifactId> + <version>${maven-shade-plugin.version}</version> + <executions> + <execution> + <phase>package</phase> + <goals> + <goal>shade</goal> + </goals> + <configuration> + <skip>${hudi.spark.native.bundle.skip}</skip> + <createSourcesJar>${shadeSources}</createSourcesJar> + <dependencyReducedPomLocation>${project.build.directory}/dependency-reduced-pom.xml + </dependencyReducedPomLocation> + <transformers> + <transformer implementation="org.apache.maven.plugins.shade.resource.ApacheLicenseResourceTransformer" /> + <transformer implementation="org.apache.maven.plugins.shade.resource.ApacheNoticeResourceTransformer"> + <addHeader>true</addHeader> + </transformer> + <transformer implementation="org.apache.maven.plugins.shade.resource.IncludeResourceTransformer"> + <resource>META-INF/LICENSE</resource> + <file>target/classes/META-INF/LICENSE</file> + </transformer> + <transformer implementation="org.apache.maven.plugins.shade.resource.AppendingTransformer"> + <resource>META-INF/services/org.apache.spark.sql.sources.DataSourceRegister</resource> + </transformer> + <transformer implementation="org.apache.maven.plugins.shade.resource.ServicesResourceTransformer"/> + </transformers> + <artifactSet> + <includes combine.children="append"> + <include>org.apache.hudi:hudi-hadoop-common</include> + <include>org.apache.hudi:hudi-common</include> + <include>org.apache.hudi:hudi-client-common</include> + <include>org.apache.hudi:hudi-spark-client</include> + <include>org.apache.hudi:hudi-spark-common_${scala.binary.version}</include> + <include>org.apache.hudi:hudi-spark_${scala.binary.version}</include> + <include>org.apache.hudi:${hudi.spark.module}_${scala.binary.version}</include> + <include>org.apache.hudi:${hudi.spark.common.module}</include> + <include>org.apache.hudi:hudi-hive-sync</include> + <include>org.apache.hudi:hudi-sync-common</include> + <include>org.apache.hudi:hudi-hadoop-mr</include> + <include>org.apache.hudi:hudi-timeline-service</include> + + <!-- NOTE: Comet must never be relocated. Its native library is bound through JNI + symbols named Java_org_apache_comet_Native_*, and the shared objects are + resolved as classpath resources under org/apache/comet/<os>/<arch>/, so a + relocation of org.apache.comet builds cleanly and then fails on the first + native call. The same applies to org.apache.arrow.c, which Comet carries + unshaded for the Arrow C data interface JNI bindings. --> + <include>org.apache.datafusion:*</include> + + <include>javax.servlet:javax.servlet-api</include> + <include>com.beust:jcommander</include> + <include>io.javalin:javalin</include> + <!-- Spark only has mortbay jetty --> + <include>org.eclipse.jetty:*</include> + <include>org.eclipse.jetty.websocket:*</include> + <include>org.jetbrains.kotlin:*</include> + <include>org.rocksdb:rocksdbjni</include> + <!-- Bundle Jackson JSR310 library since it is not present in spark 2.x. For spark 3.x this will + bundle the same JSR310 version that is included in spark runtime --> + <include>com.fasterxml.jackson.datatype:jackson-datatype-jsr310</include> + <include>com.lmax:disruptor</include> + <include>com.github.davidmoten:guava-mini</include> + <include>com.github.davidmoten:hilbert-curve</include> + <include>com.github.ben-manes.caffeine:caffeine</include> + <include>org.apache.parquet:parquet-avro</include> + <include>com.twitter:chill-protobuf</include> + + <!-- NOTE: unlike hudi-spark-bundle, dropwizard metrics are neither bundled nor Review Comment: **major:** Without dropwizard, Hudi's classes in this jar keep raw `com.codahale`, but `hudi-utilities-slim-bundle` and `hudi-aws-bundle` are compiled against `org.apache.hudi.com.codahale` and take `Metrics` from the paired Spark bundle. With `hoodie.metrics.on=true`, slim + native fails with `NoSuchMethodError` on `Metrics.getRegistry` (`HoodieStreamerMetrics:97`), and the CloudWatch reporter's reflective constructor lookup fails. This came from option (a) of my round-1 comment. Could this module instead shade the already-relocated `hudi-spark-bundle` jar plus Comet, with no relocations of its own? `CometSource` keeps Spark's raw names, Hudi keeps its relocated copy, and this pom stops having to track `hudi-spark-bundle`'s. ########## packaging/bundle-validation/validate.sh: ########## @@ -432,3 +524,15 @@ if [[ ${SCALA_PROFILE} != 'scala-2.13' ]]; then fi echo "::warning::validate.sh done validating metaserver bundle" fi + +# Runs last, so a failure here cannot skip the validation of another bundle. Review Comment: **major:** This block is unreachable in PR CI. Since #19655 (ef07f0f, 08-25) the only active `validate-bundles` entry is flink2.2, so `:462` runs the flink test and `exit 0`s before getting here. [Job 112572884710](https://github.com/apache/hudi/actions/runs/37553065321/job/112572884710) at 68cc511 staged the native jar but logged no native line, so that green says nothing about Comet. Could we move this call above the flink 2.x early exit, and correct the `bot.yml:1338` comment that says the Spark 3.5 lane runs it? ########## packaging/bundle-validation/validate.sh: ########## @@ -69,6 +76,91 @@ use_default_java_runtime () { export JAVA_HOME=${DEFAULT_JAVA_HOME} } +## +# Function to test the spark native bundle, which carries Apache DataFusion Comet. +# +# Comet ships class file version 61 bytecode and a glibc linked libcomet.so, so this only runs on +# the Java 17 pass. +# +# A Hudi path read looks its table up in the session catalog, and the Hive metastore configured in +# hive-site.xml only runs inside test_spark_hadoop_mr_bundles, so the session uses the in-memory +# catalog. +# +# env vars (defined in container): +# SPARK_HOME: path to the spark directory +## +test_spark_native_bundle () { + local outputDir=/tmp/spark-native-bundle + rm -rf $outputDir + change_java_runtime_version + echo "::warning::validate.sh Writing and querying Hudi tables with Comet enabled" + $SPARK_HOME/bin/spark-shell --jars $JARS_DIR/spark-native.jar \ + --conf 'spark.plugins=org.apache.spark.CometPlugin' \ + --conf 'spark.sql.extensions=org.apache.spark.sql.hudi.HoodieSparkSessionExtension,org.apache.comet.CometSparkSessionExtensions' \ + --conf 'spark.shuffle.manager=org.apache.spark.sql.comet.execution.shuffle.CometShuffleManager' \ + --conf 'spark.comet.enabled=true' \ + --conf 'spark.comet.exec.enabled=true' \ + --conf 'spark.comet.convert.parquet.enabled=true' \ + --conf 'spark.comet.explain.fallback.enabled=true' \ + --conf 'spark.comet.metrics.enabled=true' \ Review Comment: **major:** This conf is inert, so the round-1 `CometSource` `AbstractMethodError` could regress unnoticed. Comet 1.0.0 `Plugins.scala:58-61` returns before `registerCometMetrics` unless `spark.memory.offHeap.enabled` or `spark.comet.exec.onHeap.enabled` is set, and neither is here (nor in the module README usage block). Could we add `spark.memory.offHeap.enabled=true` and `spark.memory.offHeap.size=1g` here and in the README, plus `spark.comet.nativeLoadRequired=true` so a libcomet load failure surfaces the real error? -- 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]
