How to see total pending containers ?
Hi, Is there a way to see total pending containers in a cluster,so that we know how far behind we are with etl ? There is a pending containers field on the scheduler page under dr. who table,but that is always zero. -- Thanks, Ashwin
Why resource requests are normalized in RM ?
Hi, Anyone knows why resource requests from AMs are normalized to be multiples of yarn.scheduler.minimum-allocation-mb which is 1G by default ? Also is there any problem with reducing yarn.scheduler.minimum-allocation-mb to less than 1G ? /** * Utility method to normalize a list of resource requests, by insuring that * the memory for each request is a multiple of minMemory and is not zero. */ SchedulerUtils.normalizeRequests() -- Thanks, Ashwin
assignMultiple and continuous scheduling in Fair scheduler
Hi, I see these two knobs in Fair scheduler - 'assignMultiple' and 'continuos scheduling'. 1. Are there performance benefits using them ? What are the cons ? 2. Also is there any problem with 'continuous scheduling',I'm asking because this is not mentioned in the FS doc ? -- Thanks, Ashwin
Re: Differences between HistoryServer and Yarn TimeLine server?
Thanks Zhijie ! I had few more questions : 1. I played around with the timeline server ui today which showed the generic application history details, but I couldn't find any page for application specific data. Is the expectation that every application needs to build their own UI using the exposed REST apis and somehow install it with timeline server ? Or am I missing something. 2. Are there REST apis for accessing both generic and framework specific data in 2.4.0 ? 3. Is there an approximate timeframe for timeline server to be feature complete ? 4. Tez doesn't have any job history UI,is there any work being done to integrate Tez with timeline server ? If not,is the timeline server ready for such integration in case someone wants to pick this up ? Thanks, Ashwin On Thu, Apr 24, 2014 at 12:00 AM, Zhijie Shen zs...@hortonworks.com wrote: Ashwin, YARN-321 focuses on the issue in the scope of generic application history service, while YARN-1530 covers the framework specific data service. And yes, the timeline server is going to cover both. We've not such a Jira before, but it is described in YARN-321's design doc. Anyway, I open a Jira (MAPREDUCE-5858) to track this issue. On Wed, Apr 23, 2014 at 11:25 PM, Ashwin Shankar ashwinshanka...@gmail.com wrote: Hi Zhijie, There seems to two umbrella jiras for this - YARN-321 and YARN-1530,can you please let me know what is the difference ? Is timeline server finally going to be YARN321+YARN1530 ? You mentioned that MR is going to integrated with timeline server,is there a jira I can watch ? Thanks, Ashwin On Wed, Apr 23, 2014 at 10:15 PM, Zhijie Shen zs...@hortonworks.comwrote: Sam, You're right. We can definitely integrate MapReduce to use the timeline server to store and serve its specific data, and this is actually our plan. However, it's a big move, and we still need time to get it done. In addition, not to disturb the users that are currently relying on JHS for MR job information, we cannot simply remove JHS from Hadoop. On Wed, Apr 23, 2014 at 8:15 PM, sam liu samliuhad...@gmail.com wrote: Zhijie, I am much clear now. Thanks a lot! As my understanding, besides previous Job History Server, hadoop now has a new timeline server which could restore both the generic YARN application history and the framework specific information. However, I think the timeline server also include the functions of Job History Server, because it can store the framework specific information(of course, include mapreduce framework). In another words, Job History Server is not necessary any more.* If that's the case, why hadoop still include Job History Server?* 2014-04-23 12:56 GMT+08:00 Zhijie Shen zs...@hortonworks.com: In Hadoop 2.4, we have delivered the timeline server at a preview stage, which actually can serve some generic YARN application history as well as the framework specific information. Due to the development logistics, we have created the two concepts: History Server and Timeline Server. To be simple, you can consider the history server of the service of the generic YARN application information, while consider the timeline server of the service of the framework specific information. Importantly, we just have one daemon, which includes both services, and which we'd like to call timeline server (unfortunately, the confusing thing is that the command to start the daemon is historyserver). We're going on working on the timeline server to integrate these two parts, including refactoring the names. BTW, if you mean MapReduce JobHistoryServer by HistoryServer, it's a different daemon, which serves the historic information of MapReduce jobs only. On Tue, Apr 22, 2014 at 8:44 PM, sam liu samliuhad...@gmail.comwrote: Hi Experts, I am confusing on these two concepts. Could you help explain the differences? Thanks! -- Zhijie Shen Hortonworks Inc. http://hortonworks.com/ CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You. -- Zhijie Shen Hortonworks Inc. http://hortonworks.com/ CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution
Re: Differences between HistoryServer and Yarn TimeLine server?
Hi Zhijie, There seems to two umbrella jiras for this - YARN-321 and YARN-1530,can you please let me know what is the difference ? Is timeline server finally going to be YARN321+YARN1530 ? You mentioned that MR is going to integrated with timeline server,is there a jira I can watch ? Thanks, Ashwin On Wed, Apr 23, 2014 at 10:15 PM, Zhijie Shen zs...@hortonworks.com wrote: Sam, You're right. We can definitely integrate MapReduce to use the timeline server to store and serve its specific data, and this is actually our plan. However, it's a big move, and we still need time to get it done. In addition, not to disturb the users that are currently relying on JHS for MR job information, we cannot simply remove JHS from Hadoop. On Wed, Apr 23, 2014 at 8:15 PM, sam liu samliuhad...@gmail.com wrote: Zhijie, I am much clear now. Thanks a lot! As my understanding, besides previous Job History Server, hadoop now has a new timeline server which could restore both the generic YARN application history and the framework specific information. However, I think the timeline server also include the functions of Job History Server, because it can store the framework specific information(of course, include mapreduce framework). In another words, Job History Server is not necessary any more.* If that's the case, why hadoop still include Job History Server?* 2014-04-23 12:56 GMT+08:00 Zhijie Shen zs...@hortonworks.com: In Hadoop 2.4, we have delivered the timeline server at a preview stage, which actually can serve some generic YARN application history as well as the framework specific information. Due to the development logistics, we have created the two concepts: History Server and Timeline Server. To be simple, you can consider the history server of the service of the generic YARN application information, while consider the timeline server of the service of the framework specific information. Importantly, we just have one daemon, which includes both services, and which we'd like to call timeline server (unfortunately, the confusing thing is that the command to start the daemon is historyserver). We're going on working on the timeline server to integrate these two parts, including refactoring the names. BTW, if you mean MapReduce JobHistoryServer by HistoryServer, it's a different daemon, which serves the historic information of MapReduce jobs only. On Tue, Apr 22, 2014 at 8:44 PM, sam liu samliuhad...@gmail.com wrote: Hi Experts, I am confusing on these two concepts. Could you help explain the differences? Thanks! -- Zhijie Shen Hortonworks Inc. http://hortonworks.com/ CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You. -- Zhijie Shen Hortonworks Inc. http://hortonworks.com/ CONFIDENTIALITY NOTICE NOTICE: This message is intended for the use of the individual or entity to which it is addressed and may contain information that is confidential, privileged and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, you are hereby notified that any printing, copying, dissemination, distribution, disclosure or forwarding of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and delete it from your system. Thank You. -- Thanks, Ashwin
Re: Setting debug log level for individual daemons
Yes, thank you Stanley ! Ashwin On Tue, Apr 15, 2014 at 8:01 PM, Stanley Shi s...@gopivotal.com wrote: Is this what you are looking for? http://hadoop.apache.org/docs/r2.3.0/hadoop-project-dist/hadoop-common/CommandsManual.html#daemonlog Regards, *Stanley Shi,* On Wed, Apr 16, 2014 at 2:06 AM, Ashwin Shankar ashwinshanka...@gmail.com wrote: Thanks Gordon and Stanley, but this would require us to bounce the process. Is there a way to change log levels without bouncing the process ? On Tue, Apr 15, 2014 at 3:23 AM, Gordon Wang gw...@gopivotal.com wrote: Put the following line in the log4j setting file. log4j.logger.org.apache.hadoop.yarn.server.resourcemanager=DEBUG,console On Tue, Apr 15, 2014 at 8:33 AM, Ashwin Shankar ashwinshanka...@gmail.com wrote: Hi, How do we set log level to debug for lets say only Resource manager and not the other hadoop daemons ? -- Thanks, Ashwin -- Regards Gordon Wang -- Thanks, Ashwin -- Thanks, Ashwin
Re: Setting debug log level for individual daemons
Thanks Gordon and Stanley, but this would require us to bounce the process. Is there a way to change log levels without bouncing the process ? On Tue, Apr 15, 2014 at 3:23 AM, Gordon Wang gw...@gopivotal.com wrote: Put the following line in the log4j setting file. log4j.logger.org.apache.hadoop.yarn.server.resourcemanager=DEBUG,console On Tue, Apr 15, 2014 at 8:33 AM, Ashwin Shankar ashwinshanka...@gmail.com wrote: Hi, How do we set log level to debug for lets say only Resource manager and not the other hadoop daemons ? -- Thanks, Ashwin -- Regards Gordon Wang -- Thanks, Ashwin
Setting debug log level for individual daemons
Hi, How do we set log level to debug for lets say only Resource manager and not the other hadoop daemons ? -- Thanks, Ashwin
Resetting dead datanodes list
Hi, Hadoop-1's name node UI displays dead datanodes even if those instances are terminated and are not part of the cluster anymore. Is there a way to reset the dead datenode list without bouncing namenode ? This would help me in my script(which would run nightly) which parses the html page,terminates dead datanodes and resize the cluster. -- Thanks, Ashwin
Job fails if I change HADOOP_USER_NAME
Hi, I'm writing a new feature in Fair scheduler and wanted to test it out by running jobs submitted by different users from my laptop. My sleep job runs fine as long as the user name is my mac user name. If I change my hadoop user name by setting HADOOP_USER_NAME, my jobs fail with the exception *org.apache.hadoop.util.Shell$ExitCodeException.* I also tried creating a new user account on my laptop and running a job as that user but I get the same exception. Please let me know if any of you have come across this. I tried changing ulimits max proc(to 1024),but doesn't solve the problem. Here is the stack trace : Job job_1395389889916_0001 failed with state FAILED due to: Application application_1395389889916_0001 failed 3 times due to AM Container for appattempt_1395389889916_0001_03 exited with exitCode: 1 due to: Exception from container-launch: org.apache.hadoop.util.Shell$ExitCodeException: org.apache.hadoop.util.Shell$ExitCodeException: at org.apache.hadoop.util.Shell.runCommand(Shell.java:505) at org.apache.hadoop.util.Shell.run(Shell.java:418) at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:650) at org.apache.hadoop.yarn.server.nodemanager.DefaultContainerExecutor.launchContainer(DefaultContainerExecutor.java:195) -- Thanks, Ashwin
Re: Job fails if I change HADOOP_USER_NAME
Hi Rohit, How I enable debug for AM container logs ? and to which location are they written to ? I tried changing log4j.prop and can see DEBUGs for RM,NM etc but I don't see AM related debug logs. Thanks, Ashwin On Fri, Mar 21, 2014 at 3:05 AM, Rohith Sharma K S rohithsharm...@huawei.com wrote: Hi The below stack trace is generic for any am launcher failed to launch. Can debug on AM container logs, so get proper stacktrace.? Thanks Regards Rohith Sharma K S *From:* Ashwin Shankar [mailto:ashwinshanka...@gmail.com] *Sent:* 21 March 2014 14:02 *To:* user@hadoop.apache.org *Subject:* Job fails if I change HADOOP_USER_NAME Hi, I'm writing a new feature in Fair scheduler and wanted to test it out by running jobs submitted by different users from my laptop. My sleep job runs fine as long as the user name is my mac user name. If I change my hadoop user name by setting HADOOP_USER_NAME, my jobs fail with the exception *org.apache.hadoop.util.Shell$ExitCodeException.* I also tried creating a new user account on my laptop and running a job as that user but I get the same exception. Please let me know if any of you have come across this. I tried changing ulimits max proc(to 1024),but doesn't solve the problem. Here is the stack trace : Job job_1395389889916_0001 failed with state FAILED due to: Application application_1395389889916_0001 failed 3 times due to AM Container for appattempt_1395389889916_0001_03 exited with exitCode: 1 due to: Exception from container-launch: org.apache.hadoop.util.Shell$ExitCodeException: org.apache.hadoop.util.Shell$ExitCodeException: at org.apache.hadoop.util.Shell.runCommand(Shell.java:505) at org.apache.hadoop.util.Shell.run(Shell.java:418) at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:650) at org.apache.hadoop.yarn.server.nodemanager.DefaultContainerExecutor.launchContainer(DefaultContainerExecutor.java:195) -- Thanks, Ashwin -- Thanks, Ashwin