[jira] [Created] (FLINK-10694) ZooKeeperRunningJobsRegistry Cleanup
Mikhail Pryakhin created FLINK-10694: Summary: ZooKeeperRunningJobsRegistry Cleanup Key: FLINK-10694 URL: https://issues.apache.org/jira/browse/FLINK-10694 Project: Flink Issue Type: Bug Components: DataStream API Affects Versions: 1.6.1 Reporter: Mikhail Pryakhin When a streaming job with Zookeeper-HA enabled gets cancelled all the job-related Zookeeper nodes are not removed. Is there a reason behind that? I noticed that Zookeeper paths are created of type "Container Node" (an Ephemeral node that can have nested nodes) and fall back to Persistent node type in case Zookeeper doesn't support this sort of nodes. But anyway, it is worth removing the job Zookeeper node when a job is cancelled, isn't it? zookeeper version 3.4.10 flink version 1.6.1 # The job is deployed as a YARN cluster with the following properties set {noformat} high-availability: zookeeper high-availability.zookeeper.quorum: high-availability.zookeeper.storageDir: hdfs:/// high-availability.zookeeper.path.root: high-availability.zookeeper.path.namespace: {noformat} # The job is cancelled via flink cancel command. What I've noticed: when the job is running the following directory structure is created in zookeeper {noformat} ///leader/resource_manager_lock ///leader/rest_server_lock ///leader/dispatcher_lock ///leader/5c21f00b9162becf5ce25a1cf0e67cde/job_manager_lock ///leaderlatch/resource_manager_lock ///leaderlatch/rest_server_lock ///leaderlatch/dispatcher_lock ///leaderlatch/5c21f00b9162becf5ce25a1cf0e67cde/job_manager_lock ///checkpoints/5c21f00b9162becf5ce25a1cf0e67cde/041 ///checkpoint-counter/5c21f00b9162becf5ce25a1cf0e67cde ///running_job_registry/5c21f00b9162becf5ce25a1cf0e67cde {noformat} when the job is cancelled the some ephemeral nodes disappear, but most of them are still there: {noformat} ///leader/5c21f00b9162becf5ce25a1cf0e67cde ///leaderlatch/resource_manager_lock ///leaderlatch/rest_server_lock ///leaderlatch/dispatcher_lock ///leaderlatch/5c21f00b9162becf5ce25a1cf0e67cde/job_manager_lock ///checkpoints/ ///checkpoint-counter/ ///running_job_registry/ {noformat} Here is the method [1] responsible for cleaning zookeeper folders up [1] which is called when the job manager has stopped [2]. And it seems it only cleans up the folder running_job_registry, other folders stay untouched. I supposed that everything under the *///* folder is cleaned up when the job is cancelled. [1] [https://github.com/apache/flink/blob/8674b69964eae50cad024f2c5caf92a71bf21a09/flink-runtime/src/main/java/org/apache/flink/runtime/highavailability/zookeeper/ZooKeeperRunningJobsRegistry.java#L107] [2] [https://github.com/apache/flink/blob/f087f57749004790b6f5b823d66822c36ae09927/flink-runtime/src/main/java/org/apache/flink/runtime/dispatcher/Dispatcher.java#L332] -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Created] (FLINK-6950) Add ability to specify single jar files to be shipped to YARN
Mikhail Pryakhin created FLINK-6950: --- Summary: Add ability to specify single jar files to be shipped to YARN Key: FLINK-6950 URL: https://issues.apache.org/jira/browse/FLINK-6950 Project: Flink Issue Type: Improvement Reporter: Mikhail Pryakhin Priority: Minor When deploying a flink job on YARN it is not possible to specify multiple yarnship folders. Often when submitting a flink job, job dependencies and job resources are located in different local folders and both of them should be shipped to YARN cluster. I think it would be great to have an ability to specify jars but not folders that should be shipped to YARN cluster (via the --yarnship-jars option). -- This message was sent by Atlassian JIRA (v6.4.14#64029)
[jira] [Created] (FLINK-6949) Make custom resource files to be shipped to YARN cluster
Mikhail Pryakhin created FLINK-6949: --- Summary: Make custom resource files to be shipped to YARN cluster Key: FLINK-6949 URL: https://issues.apache.org/jira/browse/FLINK-6949 Project: Flink Issue Type: Improvement Components: Client Affects Versions: 1.3.0 Reporter: Mikhail Pryakhin *The problem:* When deploying a flink job on YARN it is not possible to specify custom resource files to be shipped to YARN cluster. *The use case description:* When running a flink job on multiple environments it becomes necessary to pass environment-related configuration files to the job's runtime. It can be accomplished by packaging configuration files within the job's jar. But having tens of different environments one can easily end up packaging as many jar as there are environments. It would be great to have an ability to separate configuration files from the job artifacts. *The possible solution:* add the --yarnship-files option to flink cli to specify files that should be shipped to the YARN cluster. -- This message was sent by Atlassian JIRA (v6.4.14#64029)
[jira] [Created] (FLINK-6912) Consider changing the RichFunction#open method signature to take no arguments.
Mikhail Pryakhin created FLINK-6912: --- Summary: Consider changing the RichFunction#open method signature to take no arguments. Key: FLINK-6912 URL: https://issues.apache.org/jira/browse/FLINK-6912 Project: Flink Issue Type: Sub-task Components: DataSet API, DataStream API Affects Versions: 1.3.0 Reporter: Mikhail Pryakhin Priority: Minor RichFunction#open(org.apache.flink.configuration.Configuration) method takes a Configuration instance as an argument which is always [passed as a new instance|https://github.com/apache/flink/blob/master/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/operators/AbstractUdfStreamOperator.java#L111] bearing no configuration parameters. As I figured out it is a remnant of the past since that method signature originates from the Record API. Consider changing the RichFunction#open method signature to take no arguments as well as actualizing java docs. You can find the complete discussion [here|http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/RichMapFunction-setup-method-td13696.html] -- This message was sent by Atlassian JIRA (v6.4.14#64029)