[ https://issues.apache.org/jira/browse/SPARK-26311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16715938#comment-16715938 ]
ASF GitHub Bot commented on SPARK-26311: ---------------------------------------- HeartSaVioR commented on a change in pull request #23260: [SPARK-26311][YARN] New feature: custom log URL for stdout/stderr URL: https://github.com/apache/spark/pull/23260#discussion_r240446659 ########## File path: resource-managers/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala ########## @@ -246,13 +246,56 @@ private[yarn] class ExecutorRunnable( sys.env.get("SPARK_USER").foreach { user => val containerId = ConverterUtils.toString(c.getId) val address = c.getNodeHttpAddress - val baseUrl = s"$httpScheme$address/node/containerlogs/$containerId/$user" - env("SPARK_LOG_URL_STDERR") = s"$baseUrl/stderr?start=-4096" - env("SPARK_LOG_URL_STDOUT") = s"$baseUrl/stdout?start=-4096" + sparkConf.get(config.CUSTOM_LOG_URL) match { + case Some(customUrl) => + val pathVariables = ExecutorRunnable.buildPathVariables(httpScheme, address, + YarnConfiguration.getClusterId(conf), containerId, user) + val envNameToFileNameMap = Map("SPARK_LOG_URL_STDERR" -> "stderr", + "SPARK_LOG_URL_STDOUT" -> "stdout") + val logUrls = ExecutorRunnable.replaceLogUrls(customUrl, pathVariables, + envNameToFileNameMap) + + logUrls.foreach { case (envName, url) => + env(envName) = url + } + case None => + val baseUrl = s"$httpScheme$address/node/containerlogs/$containerId/$user" + env("SPARK_LOG_URL_STDERR") = s"$baseUrl/stderr?start=-4096" + env("SPARK_LOG_URL_STDOUT") = s"$baseUrl/stdout?start=-4096" + } } } env } } + +private[yarn] object ExecutorRunnable { + val LOG_URL_PATTERN_HTTP_SCHEME = "{{HttpScheme}}" Review comment: Ah OK. I'm in favor of avoiding to use string constant directly, but not strong opinion on it. Will address. And yes I can put them in a single method, but placing a new method into class will bring unnecessary burden to the test code, since ExecutorRunnable receives lots of parameters to be instantiated. If we want to add an end-to-end test (instantiating YARN cluster and running executors) we still need to instantiate ExecutorRunnable (I think we are already covering it from here [1]), but if we just want to make sure the logic works properly, we might want to keep this as new object and add a test against the object to avoid instantiating ExecutorRunnable. WDYT? 1. https://github.com/apache/spark/blob/05cf81e6de3d61ddb0af81cd179665693f23351f/resource-managers/yarn/src/test/scala/org/apache/spark/deploy/yarn/YarnClusterSuite.scala#L460 ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org > [YARN] New feature: custom log URL for stdout/stderr > ---------------------------------------------------- > > Key: SPARK-26311 > URL: https://issues.apache.org/jira/browse/SPARK-26311 > Project: Spark > Issue Type: Improvement > Components: YARN > Affects Versions: 2.4.0 > Reporter: Jungtaek Lim > Priority: Major > > Spark has been setting static log URLs for YARN application, which points to > NodeManager webapp. Normally it would work for both running apps and finished > apps, but there're also other approaches on maintaining application logs, > like having external log service which enables to avoid application log url > to be a deadlink when NodeManager is not accessible. (Node decommissioned, > elastic nodes, etc.) > Spark can provide a new configuration for custom log url on YARN mode, which > end users can set it properly to point application log to external log > service. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org