zuston commented on code in PR #53: URL: https://github.com/apache/incubator-uniffle/pull/53#discussion_r937294464
########## common/src/main/java/org/apache/uniffle/common/provider/HadoopAccessorProvider.java: ########## @@ -0,0 +1,265 @@ +/* + * 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.uniffle.common.provider; + +import java.io.Closeable; +import java.io.IOException; +import java.security.PrivilegedExceptionAction; +import java.util.Map; +import java.util.concurrent.Executors; +import java.util.concurrent.ScheduledExecutorService; +import java.util.concurrent.TimeUnit; + +import com.google.common.annotations.VisibleForTesting; +import com.google.common.collect.Maps; +import org.apache.commons.lang3.StringUtils; +import org.apache.hadoop.conf.Configuration; +import org.apache.hadoop.fs.FileSystem; +import org.apache.hadoop.fs.LocalFileSystem; +import org.apache.hadoop.fs.Path; +import org.apache.hadoop.security.SecurityUtil; +import org.apache.hadoop.security.UserGroupInformation; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import org.apache.uniffle.common.util.ThreadUtils; + +/** + * The HadoopAccessorProvider will provide the only entrypoint to get the hadoop filesystem whether + * the hadoop cluster is kerberized or not. + * <p> + * It should be initialized when the shuffle server/coordinator starts. And in client, there is no need + * to login with keytab at startup, the authentication of client side should be managed by computing framework + * like Spark/MR. + */ +public class HadoopAccessorProvider implements Closeable { + private static final Logger LOGGER = LoggerFactory.getLogger(HadoopAccessorProvider.class); + + private static volatile HadoopAccessorProvider provider; + + private final boolean kerberosEnabled; + // Only valid when in kerberized cluster. + private Map<String, UserGroupInformation> cache; + private ScheduledExecutorService scheduledExecutorService; + + private HadoopAccessorProvider(SecurityInfo securityInfo) throws Exception { + if (securityInfo == null) { + this.kerberosEnabled = false; + return; + } + + this.kerberosEnabled = true; + + String keytabFile = securityInfo.getKeytabFilePath(); + String principal = securityInfo.getPrincipal(); + long reLoginIntervalSec = securityInfo.getReloginIntervalSec(); + + if (StringUtils.isEmpty(keytabFile)) { + throw new Exception("When hadoop kerberos is enabled, keytab must be set"); + } + + if (StringUtils.isEmpty(principal)) { + throw new Exception("When hadoop kerberos is enabled, principal must be set"); + } + + if (reLoginIntervalSec <= 0) { + throw new Exception("The relogin interval must be negative"); + } + + Configuration conf = new Configuration(false); + conf.set("hadoop.security.authentication", "kerberos"); + + UserGroupInformation.setConfiguration(conf); + UserGroupInformation.loginUserFromKeytab(principal, keytabFile); + + cache = Maps.newConcurrentMap(); + + LOGGER.info("Got Kerberos ticket, keytab [{}], principal [{}], user [{}]", + keytabFile, principal, UserGroupInformation.getLoginUser()); + + scheduledExecutorService = Executors.newSingleThreadScheduledExecutor( + ThreadUtils.getThreadFactory("Kerberos-relogin-%d") + ); + scheduledExecutorService.scheduleAtFixedRate( + HadoopAccessorProvider::kerberosRelogin, + reLoginIntervalSec, + reLoginIntervalSec, + TimeUnit.SECONDS); + } + + @VisibleForTesting + static void kerberosRelogin() { + try { + LOGGER.info("Renewing kerberos token."); + UserGroupInformation.getLoginUser().checkTGTAndReloginFromKeytab(); + } catch (Throwable t) { + LOGGER.error("Error in token renewal task: ", t); + } + } + + private static UserGroupInformation getProxyUser(String user) throws IOException { + provider.cache.putIfAbsent( + user, + UserGroupInformation.createProxyUser( + user, UserGroupInformation.getLoginUser() + ) + ); + return provider.cache.get(user); + } + + private static FileSystem getInternalFileSystem( + final String user, + final boolean retrievedByProxyUser, + final Path path, + final Configuration configuration) throws Exception { + if (provider == null) { + throw new Exception("HadoopAccessorProvider should be initialized."); + } + + if (retrievedByProxyUser && StringUtils.isEmpty(user)) { + throw new Exception("User must be set when security is enabled"); + } + if (retrievedByProxyUser && !provider.kerberosEnabled) { + String msg = String.format("There is need to be interactive with secured DFS by user: %s, path: %s " + + "but the HadoopAccessProvider's kerberos config is disabled and can't retrieve the " + + "secured filesystem", user, path); + throw new Exception(msg); + } + + // For local file systems, return the raw local file system, such calls to flush() + // actually flushes the stream. + try { + FileSystem fs; + if (retrievedByProxyUser) { + LOGGER.info("Fetching the proxy user ugi of {} when getting filesystem of [{}]", user, path); + UserGroupInformation proxyUserUGI = provider.getProxyUser(user); + fs = proxyUserUGI.doAs(new PrivilegedExceptionAction<FileSystem>() { + @Override + public FileSystem run() throws Exception { + return path.getFileSystem(configuration); + } + }); + } else { + fs = path.getFileSystem(configuration); + } + if (fs instanceof LocalFileSystem) { + LOGGER.debug("{} is local file system", path); + return ((LocalFileSystem) fs).getRawFileSystem(); + } + return fs; + } catch (IOException e) { + LOGGER.error("Fail to get filesystem of {}", path); + throw e; + } + } + + /** + * The only entrypoint is to get the hadoop filesystem instance and is compatible with + * the kerberos security. + * + * When to invoke this method? + * 1. For shuffle server side, it needs to get filesystem before writing the shuffle data to secured HDFS + * with the spark job's user auth. + * + * @param user + * @param path + * @param conf + * @return + * @throws Exception + */ + public static FileSystem getFilesystem( + final String user, + final Path path, + final Configuration conf) throws Exception { + UserGroupInformation.AuthenticationMethod authenticationMethod = + SecurityUtil.getAuthenticationMethod(conf); + boolean securityEnable = authenticationMethod != UserGroupInformation.AuthenticationMethod.SIMPLE; + return getInternalFileSystem(user, securityEnable, path, conf); + } + + /** + * The method is to return the Hadoop Filesystem directly which is not retrieved by + * ugi proxy user. + * + * When to invoke this method? + * 1. In client side, spark shuffle-reader getting filesystem before reading shuffle data stored in HDFS. + * 2. In shuffle-server/coordinator side, it reads the config file stored in HDFS. + * + * @param path + * @param configuration + * @return + * @throws Exception + */ + public static FileSystem getFileSystem( + final Path path, + final Configuration configuration) throws Exception { + return getInternalFileSystem(null, false, path, configuration); + } + + /** + * For kerberized cluster access + * @param securityInfo + * @throws Exception + */ + public static void init(SecurityInfo securityInfo) throws Exception { + if (provider == null) { + synchronized (HadoopAccessorProvider.class) { + if (provider == null) { + final HadoopAccessorProvider hadoopAccessorProvider = new HadoopAccessorProvider(securityInfo); + provider = hadoopAccessorProvider; + } + } + } + LOGGER.info("The {} has been initialized, kerberos enable: {}", + HadoopAccessorProvider.class.getSimpleName(), + provider.kerberosEnabled); + } + + /** + * No need to + * For non-kerberized cluster access, like client side reading shuffle data. + */ + public static void init() throws Exception { + init(null); + } + + @VisibleForTesting + public static void cleanup() throws Exception { + if (provider != null) { + provider.close(); + provider = null; + } + } + + @Override + public void close() throws IOException { + if (cache != null) { + for (UserGroupInformation ugi : cache.values()) { + try { + FileSystem.closeAllForUGI(ugi); + } catch (IOException ioException) { + LOGGER.error("Exception occurred while closing filesystems for {}", ugi.getUserName(), ioException); + } + } + cache.clear(); Review Comment: No cache is also OK. But i think the number of users may be not too much. -- 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]
