R: spark 1.2 writing on parquet after a join never ends - GC problems
Could anyone figure out what is going in my spark cluster? Thanks in advance Paolo Inviata dal mio Windows Phone Da: Paolo Plattermailto:paolo.plat...@agilelab.it Inviato: 06/02/2015 10:48 A: user@spark.apache.orgmailto:user@spark.apache.org Oggetto: spark 1.2 writing on parquet after a join never ends - GC problems Hi all, I’m experiencing a strange behaviour of spark 1.2. I’ve a 3 node cluster + the master. each node has: 1 HDD 7200 rpm 1 TB 16 GB RAM 8 core I configured executors with 6 cores and 10 GB each ( spark.storage.memoryFraction = 0.6 ) My job is pretty simple: val file1 = sc.parquetFile(“path1”) //19M rows val file2 = sc.textFile(“path2”) //12K rows val join = file1.as(‘f1’).join(file2.as(‘f2’), LeftOuter, Some(“f1.field”.attr === ”f2.field”.attr)) join.map( _.toCaseClass() ).saveAsParquetFile( “path3” ) When I perform this job into the spark-shell without writing on parquet file, but performing a final count to execute the pipeline, it’s pretty fast. When I submit the application to the cluster with the saveAsParquetFile instruction, task execution slows progressively and it never ends. I debugged this behaviour and I found that the cause is the executor’s disconnection due to missing heartbeat. Missing heatbeat in my opinion is related to GC (I report to you a piece of GC log from one of the executors) 484.861: [GC [PSYoungGen: 2053788K-718157K(2561024K)] 7421222K-6240219K(9551872K), 2.6802130 secs] [Times: user=1.94 sys=0.60, real=2.68 secs] 497.751: [GC [PSYoungGen: 2560845K-782081K(2359808K)] 8082907K-6984335K(9350656K), 4.8611660 secs] [Times: user=3.66 sys=1.55, real=4.86 secs] 510.654: [GC [PSYoungGen: 2227457K-625664K(2071552K)] 8429711K-7611342K(9062400K), 22.5727850 secs] [Times: user=3.34 sys=2.43, real=22.57 secs] 533.745: [Full GC [PSYoungGen: 625664K-0K(2071552K)] [ParOldGen: 6985678K-2723917K(6990848K)] 7611342K-2723917K(9062400K) [PSPermGen: 62290K-6 K(124928K)], 56.9075910 secs] [Times: user=65.28 sys=5.91, real=56.90 secs] 667.637: [GC [PSYoungGen: 1445376K-623184K(2404352K)] 4169293K-3347101K(9395200K), 11.7959290 secs] [Times: user=1.58 sys=0.60, real=11.79 secs] 690.936: [GC [PSYoungGen: 1973328K-584256K(2422784K)] 4697245K-3932841K(9413632K), 39.3594850 secs] [Times: user=2.88 sys=0.96, real=39.36 secs] 789.891: [GC [PSYoungGen: 1934400K-585552K(2434048K)] 5282985K-4519857K(9424896K), 17.4456720 secs] [Times: user=2.65 sys=1.36, real=17.44 secs] 814.697: [GC [PSYoungGen: 1951056K-330109K(2426880K)] 5885361K-4851426K(9417728K), 20.9578300 secs] [Times: user=1.64 sys=0.81, real=20.96 secs] 842.968: [GC [PSYoungGen: 1695613K-180290K(2489344K)] 6216930K-4888775K(9480192K), 3.2760780 secs] [Times: user=0.40 sys=0.30, real=3.28 secs] 886.660: [GC [PSYoungGen: 1649218K-427552K(2475008K)] 6357703K-5239028K(9465856K), 5.4738210 secs] [Times: user=1.47 sys=0.25, real=5.48 secs] 897.979: [GC [PSYoungGen: 1896480K-634144K(2487808K)] 6707956K-5874208K(9478656K), 23.6440110 secs] [Times: user=2.63 sys=1.11, real=23.64 secs] 929.706: [GC [PSYoungGen: 2169632K-663200K(2199040K)] 7409696K-6538992K(9189888K), 39.3632270 secs] [Times: user=3.36 sys=1.71, real=39.36 secs] 1006.206: [GC [PSYoungGen: 2198688K-655584K(2449920K)] 8074480K-7196224K(9440768K), 98.5040880 secs] [Times: user=161.53 sys=6.71, real=98.49 secs] 1104.790: [Full GC [PSYoungGen: 655584K-0K(2449920K)] [ParOldGen: 6540640K-6290292K(6990848K)] 7196224K-6290292K(9440768K) [PSPermGen: 62247K-6224 7K(131072K)], 610.0023700 secs] [Times: user=1630.17 sys=27.80, real=609.93 secs] 1841.916: [Full GC [PSYoungGen: 1440256K-0K(2449920K)] [ParOldGen: 6290292K-6891868K(6990848K)] 7730548K-6891868K(9440768K) [PSPermGen: 62266K-622 66K(131072K)], 637.4852230 secs] [Times: user=2035.09 sys=36.09, real=637.40 secs] 2572.012: [Full GC [PSYoungGen: 1440256K-509513K(2449920K)] [ParOldGen: 6891868K-6990703K(6990848K)] 8332124K-7500217K(9440768K) [PSPermGen: 62275K -62275K(129024K)], 698.2497860 secs] [Times: user=2261.54 sys=37.63, real=698.26 secs] 3326.711: [Full GC It might seem that the writing file operation is too slow and it’s a bottleneck, but then I tried to chenge my algorithm in the following way : val file1 = sc.parquetFile(“path1”) //19M rows val file2 = sc.textFile(“path2”) //12K rows val bFile2 = sc.broadcast( file2.collect.groupBy( f2 = f2.filed ) ) //broadcast of the smaller file as Map() file1.map( f1 = ( f1, bFile2.value( f1.field ).head ) ) //manual join .map( _toCaseClass() ) .saveAsParquetFile( “path3” ) in this way the task is fast and ends without problems, so now I’m pretty confused. * Join works well if I use count as final action * Parquet write is working well without previous join operation * Parquet write after join never ends and I detected GC problems Anyone can figure out what it’s happening ? Thanks Paolo
Re: Installing a python library along with ec2 cluster
Hi, You can make a image of ec2 with all the python libraries installed and create a bash script to export python_path in the /etc/init.d/ directory. Then you can launch the cluster with this image and ec2.py Hope this can be helpful Cheers Gen On Sun, Feb 8, 2015 at 9:46 AM, Chengi Liu chengi.liu...@gmail.com wrote: Hi, I want to install couple of python libraries (pip install python_library) which I want to use on pyspark cluster which are developed using the ec2 scripts. Is there a way to specify these libraries when I am building those ec2 clusters? Whats the best way to install these libraries on each ec2 node? Thanks
Re: no space left at worker node
Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb. If yes, you should 1. stop hdfs 2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful. Cheers Gen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Mesos coarse mode not working (fine grained does)
Hi, I’m trying to get coarse mode to work under mesos(0.21.0), I thought this would be a trivial change as Mesos was working well in fine-grained mode. However the mesos tasks fail, I can’t pinpoint where things go wrong. This is a mesos stderr log from a slave: Fetching URI 'http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:57:45.415575 25720 fetcher.cpp:126] Downloading 'http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' to '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:58:09.146960 25720 fetcher.cpp:64] Extracted resource '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' into '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151’ Mesos slaves' stdout are empty. And I can confirm the spark distro is correctly extracted: $ ls spark-1.2.0-bin-hadoop2.4 spark-1.2.0-bin-hadoop2.4.tgz stderr stdout The spark-submit log is here: http://pastebin.com/ms3uZ2BK Mesos-master http://pastebin.com/QH2Vn1jX Mesos-slave http://pastebin.com/DXFYemix Can somebody pinpoint me to logs, etc to further investigate this, I’m feeling kind of blind. Furthermore, do the executors on mesos inherit all configs from the spark application/submit? E.g. I’ve given my executors 20GB of memory through a spark-submit —conf” parameter. Should these settings also be present in the spark-1.2.0-bin-hadoop2.4.tgz distribution’s configs? If, in order to be helped here, I need to present more logs etc, please let me know. Regards, Hans van den Bogert - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
RE: no space left at worker node
Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: no space left at worker node
Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not. The reply of Micheal is good solution about this type of problem. You can check his site. Cheers Gen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: # LABEL=/ / ext4defaults,noatime 1 1 tmpfs /dev/shmtmpfs defaults0 0 devpts /dev/ptsdevpts gid=5,mode=620 0 0 sysfs /syssysfs defaults0 0 proc/proc procdefaults0 0 /dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0 /dev/sdc/mnt2 auto defaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow -- Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb. If yes, you should 1. stop hdfs 2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful. Cheers Gen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
RE: no space left at worker node
Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not.The reply of Micheal is good solution about this type of problem. You can check his site. CheersGen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
RE: no space left at worker node
Thanks Michael. I didn't edit core-site.xml. We use the default one. I only saw hdaoop.tmp.dir in core-site.xml, pointing to /mnt/ephemeral-hdfs. How can I edit the config file? Best regards, Ey-Chih Date: Sun, 8 Feb 2015 16:51:32 + From: m_albert...@yahoo.com To: gen.tan...@gmail.com; eyc...@hotmail.com CC: user@spark.apache.org Subject: Re: no space left at worker node You might want to take a look in core-site.xml, andsee what is listed as usable directories (hadoop.tmp.dir, fs.s3.buffer.dir). It seems that on S3, the root disk is relatively small (8G), but the config files list a mnt directory under it. Somehow the system doesn't balance between the very small space it has under the root disk and the larger disks, so the root disk fills up while the others are unused. At my site, we wrote a boot script to edit these problem out of the config before hadoop starts. -Mike From: gen tang gen.tan...@gmail.com To: ey-chih chow eyc...@hotmail.com Cc: user@spark.apache.org user@spark.apache.org Sent: Sunday, February 8, 2015 6:09 AM Subject: Re: no space left at worker node Hi,I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use?If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfsHope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote:Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: Mesos coarse mode not working (fine grained does)
I wasn’t thorough, the complete stderr includes: g++: /usr/lib64/libaprutil-1.so: No such file or directory g++: /usr/lib64/libapr-1.so: No such file or directoryn (including that trailing ’n') Though I can’t figure out how the process indirection is going from the frontend spark application to mesos executors and where this shared library error comes from. Hope someone can shed some light, Thanks On 08 Feb 2015, at 14:15, Hans van den Bogert hansbog...@gmail.com wrote: Hi, I’m trying to get coarse mode to work under mesos(0.21.0), I thought this would be a trivial change as Mesos was working well in fine-grained mode. However the mesos tasks fail, I can’t pinpoint where things go wrong. This is a mesos stderr log from a slave: Fetching URI 'http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:57:45.415575 25720 fetcher.cpp:126] Downloading 'http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' to '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:58:09.146960 25720 fetcher.cpp:64] Extracted resource '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' into '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151’ Mesos slaves' stdout are empty. And I can confirm the spark distro is correctly extracted: $ ls spark-1.2.0-bin-hadoop2.4 spark-1.2.0-bin-hadoop2.4.tgz stderr stdout The spark-submit log is here: http://pastebin.com/ms3uZ2BK Mesos-master http://pastebin.com/QH2Vn1jX Mesos-slave http://pastebin.com/DXFYemix Can somebody pinpoint me to logs, etc to further investigate this, I’m feeling kind of blind. Furthermore, do the executors on mesos inherit all configs from the spark application/submit? E.g. I’ve given my executors 20GB of memory through a spark-submit —conf” parameter. Should these settings also be present in the spark-1.2.0-bin-hadoop2.4.tgz distribution’s configs? If, in order to be helped here, I need to present more logs etc, please let me know. Regards, Hans van den Bogert - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: Mesos coarse mode not working (fine grained does)
Hi there, It looks like while trying to launch the executor (or one of the process like the fetcher to fetch the uris) was failing because of the dependencies problem you see. Your mesos-slave shouldn't be able to run though, were you running 0.20.0 slave and upgraded to 0.21.0? We introduced the dependencies for libapr and libsvn for Mesos 0.21.0. What's the stdout for the task like? Tim On Mon, Feb 9, 2015 at 4:10 AM, Hans van den Bogert hansbog...@gmail.com wrote: I wasn’t thorough, the complete stderr includes: g++: /usr/lib64/libaprutil-1.so: No such file or directory g++: /usr/lib64/libapr-1.so: No such file or directoryn (including that trailing ’n') Though I can’t figure out how the process indirection is going from the frontend spark application to mesos executors and where this shared library error comes from. Hope someone can shed some light, Thanks On 08 Feb 2015, at 14:15, Hans van den Bogert hansbog...@gmail.com wrote: Hi, I’m trying to get coarse mode to work under mesos(0.21.0), I thought this would be a trivial change as Mesos was working well in fine-grained mode. However the mesos tasks fail, I can’t pinpoint where things go wrong. This is a mesos stderr log from a slave: Fetching URI 'http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:57:45.415575 25720 fetcher.cpp:126] Downloading ' http://upperpaste.com/spark-1.2.0-bin-hadoop2.4.tgz' to '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' I0208 12:58:09.146960 25720 fetcher.cpp:64] Extracted resource '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151/spark-1.2.0-bin-hadoop2.4.tgz' into '/local/vdbogert/var/lib/mesos//slaves/20150206-110658-16813322-5050-5515-S1/frameworks/20150208-125721-906005770-5050-32371-/executors/0/runs/cb525b32-387c-4698-a27e-8d4213080151’ Mesos slaves' stdout are empty. And I can confirm the spark distro is correctly extracted: $ ls spark-1.2.0-bin-hadoop2.4 spark-1.2.0-bin-hadoop2.4.tgz stderr stdout The spark-submit log is here: http://pastebin.com/ms3uZ2BK Mesos-master http://pastebin.com/QH2Vn1jX Mesos-slave http://pastebin.com/DXFYemix Can somebody pinpoint me to logs, etc to further investigate this, I’m feeling kind of blind. Furthermore, do the executors on mesos inherit all configs from the spark application/submit? E.g. I’ve given my executors 20GB of memory through a spark-submit —conf” parameter. Should these settings also be present in the spark-1.2.0-bin-hadoop2.4.tgz distribution’s configs? If, in order to be helped here, I need to present more logs etc, please let me know. Regards, Hans van den Bogert - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: Spark concurrency question
I think I have this right: You will run one executor per application per worker. Generally there is one worker per machine, and it manages all of the machine's resources. So if you want one app to use this whole machine you need to ask for 48G and 24 cores. That's better than splitting up the resources such that no executor can use more than 4G. (However with big heaps 32G it can make sense to limit the size of an executor, so for example, you could configure to run 3 workers per machine each controlling 8 cores and 16G, and ask for smaller executors. Still I don't think it would make sense to run 12 workers per machine here.) 10 tasks (1 per partition) will execute. They generally get assigned to favor data locality, but here everything's local. If you had 3 executors of 8 cores, I'm not sure if it's guaranteed to balance but it should be using at least 2 executors, since there are 10 tasks and 8*3=24 slots. In your initial scenario, I think it may be waiting because the single worker has all of its cores devoted to your first app's single executor. You can ask for fewer cores in each spark-shell. Not sure what you mean about threads. Yes of course threads are used within one JVM / executor. It's not an executor per partition; it's a task per partition and 1 executor per application per worker (and usually 1 worker per machine but not always). One task executes serially in one thread and as many tasks as slots can run concurrently, and that's 1 slot per core that the executor is using. I suppose in theory you could write a function that starts its own threads too, but that's not generally a good idea or necessary. Did you read the docs on the site? http://spark.apache.org/docs/latest/cluster-overview.html http://spark.apache.org/docs/latest/spark-standalone.html On Sun, Feb 8, 2015 at 7:18 PM, java8964 java8...@hotmail.com wrote: Hi, I have some questions about how the spark run the job concurrently. For example, if I setup the Spark on one standalone test box, which has 24 core and 64G memory. I setup the Worker memory to 48G, and Executor memory to 4G, and using spark-shell to run some jobs. Here is something confusing me: 1) Does the above setting mean that I can have up to 12 Executor running in this box at same time? 2) Let's assume that I want to do a line count of one 1280M HDFS file, which has 10 blocks as 128M per block. In this case, when the Spark program starts to run, will it kick off one executor using 10 threads to read these 10 blocks hdfs data, or 10 executors to read one block each? Or in other way? I read the Apache spark document, so I know that this 1280M HDFS file will be split as 10 partitions. But how the executor run them, I am not clear. 3) In my test case, I started one Spark-shell to run a very expensive job. I saw in the Spark web UI, there are 8 stages generated, with 200 to 400 tasks in each stage, and the tasks started to run. At this time, I started another spark shell to connect to master, and try to run a small spark program. From the spark-shell, it shows my new small program is in a wait status for resource. Why? And what kind of resources it is waiting for? If it is waiting for memory, does this means that there are 12 concurrent tasks running in the first program, took 12 * 4G = 48G memory given to the worker, so no more resource available? If so, in this case, then one running task is one executor? 4) In MapReduce, the count of map and reducer tasks are the resource used by the cluster. My understanding is Spark is using multithread, instead of individual JVM processor. In this case, is the Executor using 4G heap to generate multithreads? My real question is that if each executor corresponding to each RDD partition, or executor could span thread for a RDD partition? On the other hand, how the worker decides how many executors to be created? If there is any online document answering the above questions, please let me know. I searched in the Apache Spark site, but couldn't find it. Thanks Yong - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Spark concurrency question
Hi, I have some questions about how the spark run the job concurrently. For example, if I setup the Spark on one standalone test box, which has 24 core and 64G memory. I setup the Worker memory to 48G, and Executor memory to 4G, and using spark-shell to run some jobs. Here is something confusing me: 1) Does the above setting mean that I can have up to 12 Executor running in this box at same time?2) Let's assume that I want to do a line count of one 1280M HDFS file, which has 10 blocks as 128M per block. In this case, when the Spark program starts to run, will it kick off one executor using 10 threads to read these 10 blocks hdfs data, or 10 executors to read one block each? Or in other way? I read the Apache spark document, so I know that this 1280M HDFS file will be split as 10 partitions. But how the executor run them, I am not clear.3) In my test case, I started one Spark-shell to run a very expensive job. I saw in the Spark web UI, there are 8 stages generated, with 200 to 400 tasks in each stage, and the tasks started to run. At this time, I started another spark shell to connect to master, and try to run a small spark program. From the spark-shell, it shows my new small program is in a wait status for resource. Why? And what kind of resources it is waiting for? If it is waiting for memory, does this means that there are 12 concurrent tasks running in the first program, took 12 * 4G = 48G memory given to the worker, so no more resource available? If so, in this case, then one running task is one executor?4) In MapReduce, the count of map and reducer tasks are the resource used by the cluster. My understanding is Spark is using multithread, instead of individual JVM processor. In this case, is the Executor using 4G heap to generate multithreads? My real question is that if each executor corresponding to each RDD partition, or executor could span thread for a RDD partition? On the other hand, how the worker decides how many executors to be created? If there is any online document answering the above questions, please let me know. I searched in the Apache Spark site, but couldn't find it. Thanks Yong
Re: [GraphX] Excessive value recalculations during aggregateMessages cycles
I changed the curGraph = curGraph.outerJoinVertices(curMessages)( (vid, vertex, message) = vertex.process(message.getOrElse(List[Message]()), ti) ).cache() to curGraph = curGraph.outerJoinVertices(curMessages)( (vid, vertex, message) = (vertex, message.getOrElse(List[Message]())) ).mapVertices( (x,y) = y._1.process( y._2, ti ) ).cache() So the call to the 'process' method was moved out of the outerJoinVertices and into a separate mapVertices call, and the problem went away. Now, 'process' is only called once during the correct cycle. So it would appear that outerJoinVertices caches the closure to be recalculated if needed again while mapVertices actually caches the derived values. Is this a bug or a feature? Kyle On Sat, Feb 7, 2015 at 11:44 PM, Kyle Ellrott kellr...@soe.ucsc.edu wrote: I'm trying to setup a simple iterative message/update problem in GraphX (spark 1.2.0), but I'm running into issues with the caching and re-calculation of data. I'm trying to follow the example found in the Pregel implementation of materializing and cacheing messages and graphs and then unpersisting them after the next cycle has been done. It doesn't seem to be working, because every cycle gets progressively slower and it seems as if more and more of the values are being re-calculated despite my attempts to cache them. The code: ``` var oldMessages : VertexRDD[List[Message]] = null var oldGraph : Graph[MyVertex, MyEdge ] = null curGraph = curGraph.mapVertices((x, y) = y.init()) for (i - 0 to cycle_count) { val curMessages = curGraph.aggregateMessages[List[Message]](x = { //send messages . }, (x, y) = { //collect messages into lists val out = x ++ y out } ).cache() curMessages.count() val ti = i oldGraph = curGraph curGraph = curGraph.outerJoinVertices(curMessages)( (vid, vertex, message) = vertex.process(message.getOrElse(List[Message]()), ti) ).cache() curGraph.vertices.count() oldGraph.unpersistVertices(blocking = false) oldGraph.edges.unpersist(blocking = false) oldGraph = curGraph if (oldMessages != null ) { oldMessages.unpersist(blocking=false) } oldMessages = curMessages } ``` The MyVertex.process method takes the list of incoming messages, averages them and returns a new MyVertex object. I've also set it up to append the cycle number (the second argument) into a log file named after the vertex. What ends up getting dumped into the log file for every vertex (in the exact same pattern) is ``` Cycle: 0 Cycle: 1 Cycle: 0 Cycle: 2 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 3 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 2 Cycle: 4 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 2 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 0 Cycle: 0 Cycle: 1 Cycle: 2 Cycle: 3 Cycle: 5 ``` Any ideas about what I might be doing wrong for the caching? And how I can avoid re-calculating so many of the values. Kyle
Re: no space left at worker node
Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc. In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. Cheers Gen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow -- Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not. The reply of Micheal is good solution about this type of problem. You can check his site. Cheers Gen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: # LABEL=/ / ext4defaults,noatime 1 1 tmpfs /dev/shmtmpfs defaults0 0 devpts /dev/ptsdevpts gid=5,mode=620 0 0 sysfs /syssysfs defaults0 0 proc/proc procdefaults0 0 /dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0 /dev/sdc/mnt2 auto defaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow -- Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb. If yes, you should 1. stop hdfs 2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful. Cheers Gen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: Spark concurrency question
On Sun, Feb 8, 2015 at 10:26 PM, java8964 java8...@hotmail.com wrote: standalone one box environment, if I want to use all 48G memory allocated to worker for my application, I should ask 48G memory for the executor in the spark shell, right? Because 48G is too big for a JVM heap in normal case, I can and should consider to start multi workers in one box, to lower the executor memory, but still use all 48G memory. Yes. In the spark document, about the -- cores parameter, the default is all available cores, so it means using all available cores in all workers, even in the cluster environment? If so, in default case, if one client submit a huge job, it will use all the available cores from the cluster for all the tasks it generates? Have a look at how cores work in standalone mode: http://spark.apache.org/docs/latest/job-scheduling.html One thing is still not clear is in the given example I have, if 10 tasks (1 per partition) will execute, but there is one executor per application, in this case, I have the following 2 questions, assuming that the worker memory is set to 48G, and executor memory is set to 4G, and I use one spark-shell to connect to the master to submit my application: 1) How many executor will be created on this box (Or even in the cluster it it is running in the cluster)? I don't see any spark configuration related to set number of executor in spark shell. If it is more than one, how this number is calculated? Again from http://spark.apache.org/docs/latest/job-scheduling.html for standalone mode the default should be 1 executor per worker, but you can change that. 2) Do you mean that one partition (or one task for it) will be run by one executor? Is that one executor will run the task sequentially, but job concurrency comes from that multi executors could run synchronous, right? A partition maps to a task, which is computed serially. Tasks are executed in parallel in an executor, which can execute many tasks at once. No, parallelism does not (only) come from running many executors. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
RE: no space left at worker node
Hi Gen, Thanks. I save my logs in a file under /var/log. This is the only place to save data. Will the problem go away if I use a better machine? Best regards, Ey-Chih Chow Date: Sun, 8 Feb 2015 23:32:27 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc.In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. CheersGen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not.The reply of Micheal is good solution about this type of problem. You can check his site. CheersGen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
Re: Can't access remote Hive table from spark
Hi Lian, Will the latest 0.14.0 version of Hive,which is installed by ambari 1.7.0 by default, be supported by the next release of Spark? Regards, -- Original -- From: Cheng Lian;lian.cs@gmail.com; Send time: Friday, Feb 6, 2015 9:02 AM To: guxiaobo1...@qq.com; user@spark.apache.orguser@spark.apache.org; Subject: Re: Can't access remote Hive table from spark Please note that Spark 1.2.0 only support Hive 0.13.1 or 0.12.0, none of other versions are supported. Best, Cheng On 1/25/15 12:18 AM, guxiaobo1982 wrote: Hi, I built and started a single node standalone Spark 1.2.0 cluster along with a single node Hive 0.14.0 instance installed by Ambari 1.17.0. On the Spark and Hive node I can create and query tables inside Hive, and on remote machines I can submit the SparkPi example to the Spark master. But I failed to run the following example code : public class SparkTest { public static void main(String[] args) { String appName= This is a test application; String master=spark://lix1.bh.com:7077; SparkConf conf = new SparkConf().setAppName(appName).setMaster(master); JavaSparkContext sc = new JavaSparkContext(conf); JavaHiveContext sqlCtx = new org.apache.spark.sql.hive.api.java.JavaHiveContext(sc); //sqlCtx.sql(CREATE TABLE IF NOT EXISTS src (key INT, value STRING)); //sqlCtx.sql(LOAD DATA LOCAL INPATH '/opt/spark/examples/src/main/resources/kv1.txt' INTO TABLE src); // Queries are expressed in HiveQL. ListRow rows = sqlCtx.sql(FROM src SELECT key, value).collect(); System.out.print(I got + rows.size() + rows \r\n); sc.close();} } Exception in thread main org.apache.hadoop.hive.ql.metadata.InvalidTableException: Table not found src at org.apache.hadoop.hive.ql.metadata.Hive.getTable(Hive.java:980) at org.apache.hadoop.hive.ql.metadata.Hive.getTable(Hive.java:950) at org.apache.spark.sql.hive.HiveMetastoreCatalog.lookupRelation(HiveMetastoreCatalog.scala:70) at org.apache.spark.sql.hive.HiveContext$anon$2.org$apache$spark$sql$catalyst$analysis$OverrideCatalog$super$lookupRelation(HiveContext.scala:253) at org.apache.spark.sql.catalyst.analysis.OverrideCatalog$anonfun$lookupRelation$3.apply(Catalog.scala:141) at org.apache.spark.sql.catalyst.analysis.OverrideCatalog$anonfun$lookupRelation$3.apply(Catalog.scala:141) at scala.Option.getOrElse(Option.scala:120) at org.apache.spark.sql.catalyst.analysis.OverrideCatalog$class.lookupRelation(Catalog.scala:141) at org.apache.spark.sql.hive.HiveContext$anon$2.lookupRelation(HiveContext.scala:253) at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveRelations$anonfun$apply$5.applyOrElse(Analyzer.scala:143) at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveRelations$anonfun$apply$5.applyOrElse(Analyzer.scala:138) at org.apache.spark.sql.catalyst.trees.TreeNode.transformDown(TreeNode.scala:144) at org.apache.spark.sql.catalyst.trees.TreeNode$anonfun$4.apply(TreeNode.scala:162) at scala.collection.Iterator$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157)
Re: Sort Shuffle performance issues about using AppendOnlyMap for large data sets
Hi, Problem still exists. Any experts would take a look at this? Thanks, Sun. fightf...@163.com From: fightf...@163.com Date: 2015-02-06 17:54 To: user; dev Subject: Sort Shuffle performance issues about using AppendOnlyMap for large data sets Hi, all Recently we had caught performance issues when using spark 1.2.0 to read data from hbase and do some summary work. Our scenario means to : read large data sets from hbase (maybe 100G+ file) , form hbaseRDD, transform to schemardd, groupby and aggregate the data while got fewer new summary data sets, loading data into hbase (phoenix). Our major issue lead to : aggregate large datasets to get summary data sets would consume too long time (1 hour +) , while that should be supposed not so bad performance. We got the dump file attached and stacktrace from jstack like the following: From the stacktrace and dump file we can identify that processing large datasets would cause frequent AppendOnlyMap growing, and leading to huge map entrysize. We had referenced the source code of org.apache.spark.util.collection.AppendOnlyMap and found that the map had been initialized with capacity of 64. That would be too small for our use case. So the question is : Does anyone had encounted such issues before? How did that be resolved? I cannot find any jira issues for such problems and if someone had seen, please kindly let us know. More specified solution would goes to : Does any possibility exists for user defining the map capacity releatively in spark? If so, please tell how to achieve that. Best Thanks, Sun. Thread 22432: (state = IN_JAVA) - org.apache.spark.util.collection.AppendOnlyMap.growTable() @bci=87, line=224 (Compiled frame; information may be imprecise) - org.apache.spark.util.collection.SizeTrackingAppendOnlyMap.growTable() @bci=1, line=38 (Interpreted frame) - org.apache.spark.util.collection.AppendOnlyMap.incrementSize() @bci=22, line=198 (Compiled frame) - org.apache.spark.util.collection.AppendOnlyMap.changeValue(java.lang.Object, scala.Function2) @bci=201, line=145 (Compiled frame) - org.apache.spark.util.collection.SizeTrackingAppendOnlyMap.changeValue(java.lang.Object, scala.Function2) @bci=3, line=32 (Compiled frame) - org.apache.spark.util.collection.ExternalSorter.insertAll(scala.collection.Iterator) @bci=141, line=205 (Compiled frame) - org.apache.spark.shuffle.sort.SortShuffleWriter.write(scala.collection.Iterator) @bci=74, line=58 (Interpreted frame) - org.apache.spark.scheduler.ShuffleMapTask.runTask(org.apache.spark.TaskContext) @bci=169, line=68 (Interpreted frame) - org.apache.spark.scheduler.ShuffleMapTask.runTask(org.apache.spark.TaskContext) @bci=2, line=41 (Interpreted frame) - org.apache.spark.scheduler.Task.run(long) @bci=77, line=56 (Interpreted frame) - org.apache.spark.executor.Executor$TaskRunner.run() @bci=310, line=196 (Interpreted frame) - java.util.concurrent.ThreadPoolExecutor.runWorker(java.util.concurrent.ThreadPoolExecutor$Worker) @bci=95, line=1145 (Interpreted frame) - java.util.concurrent.ThreadPoolExecutor$Worker.run() @bci=5, line=615 (Interpreted frame) - java.lang.Thread.run() @bci=11, line=744 (Interpreted frame) Thread 22431: (state = IN_JAVA) - org.apache.spark.util.collection.AppendOnlyMap.growTable() @bci=87, line=224 (Compiled frame; information may be imprecise) - org.apache.spark.util.collection.SizeTrackingAppendOnlyMap.growTable() @bci=1, line=38 (Interpreted frame) - org.apache.spark.util.collection.AppendOnlyMap.incrementSize() @bci=22, line=198 (Compiled frame) - org.apache.spark.util.collection.AppendOnlyMap.changeValue(java.lang.Object, scala.Function2) @bci=201, line=145 (Compiled frame) - org.apache.spark.util.collection.SizeTrackingAppendOnlyMap.changeValue(java.lang.Object, scala.Function2) @bci=3, line=32 (Compiled frame) - org.apache.spark.util.collection.ExternalSorter.insertAll(scala.collection.Iterator) @bci=141, line=205 (Compiled frame) - org.apache.spark.shuffle.sort.SortShuffleWriter.write(scala.collection.Iterator) @bci=74, line=58 (Interpreted frame) - org.apache.spark.scheduler.ShuffleMapTask.runTask(org.apache.spark.TaskContext) @bci=169, line=68 (Interpreted frame) - org.apache.spark.scheduler.ShuffleMapTask.runTask(org.apache.spark.TaskContext) @bci=2, line=41 (Interpreted frame) - org.apache.spark.scheduler.Task.run(long) @bci=77, line=56 (Interpreted frame) - org.apache.spark.executor.Executor$TaskRunner.run() @bci=310, line=196 (Interpreted frame) - java.util.concurrent.ThreadPoolExecutor.runWorker(java.util.concurrent.ThreadPoolExecutor$Worker) @bci=95, line=1145 (Interpreted frame) - java.util.concurrent.ThreadPoolExecutor$Worker.run() @bci=5, line=615 (Interpreted frame) - java.lang.Thread.run() @bci=11, line=744 (Interpreted frame) fightf...@163.com 1 attachments dump.png(42K) download preview
Error when running example (pi.py)
Traceback (most recent call last): File pi.py, line 29, in module sc = SparkContext(appName=PythonPi) File /home/ashish/Downloads/spark-1.1.0-bin-hadoop2.4/python/pyspark/context.py, line 104, in __init__ SparkContext._ensure_initialized(self, gateway=gateway) File /home/ashish/Downloads/spark-1.1.0-bin-hadoop2.4/python/pyspark/context.py, line 211, in _ensure_initialized SparkContext._gateway = gateway or launch_gateway() File /home/ashish/Downloads/spark-1.1.0-bin-hadoop2.4/python/pyspark/java_gateway.py, line 48, in launch_gateway proc = Popen(command, stdout=PIPE, stdin=PIPE, preexec_fn=preexec_func) File /usr/lib/python2.7/subprocess.py, line 710, in __init__ errread, errwrite) File /usr/lib/python2.7/subprocess.py, line 1327, in _execute_child raise child_exception OSError: [Errno 13] Permission denied
Re: WebUI on yarn through ssh tunnel affected by AmIpfilter
Just to add why tunneling is not a good practice sometime: There could be some other ports/apps depeneding on other processes running on different ports. Lets say a web app running on port 8080 pulling info from other processes through rest api which will fail here since you only tunnel for 8080 and hence the ui/data will look ugly. On 9 Feb 2015 11:57, Akhil Das ak...@sigmoidanalytics.com wrote: Just make sure all ports (0-65535) are accessible across your cluster. And you may also want to open these ports for your IP address instead of tunneling: 8080, 8081, 18080, 1, 50030, 50070, 60070, 4040-4045 Thanks Best Regards On Sat, Feb 7, 2015 at 10:38 AM, yangqch davidyang...@gmail.com wrote: Hi folks, I am new to spark. I just get spark 1.2 to run on emr ami 3.3.1 (hadoop 2.4). I ssh to emr master node and submit the job or start the shell. Everything runs well except the webUI. In order to see the UI, I used ssh tunnel which forward my dev machine port to emr master node webUI port. When I open the webUI, at the very beginning of the application (during the spark launch time), the webUI is as nice as shown in many spark docs. However, once the YARN AmIpfilter started to work, the webUI becomes very ugly. No pictures can be displayed, only text can be shown (just like you view it in lynx). Meanwhile, in spark shell, it pops up amfilter.AmIpFilter (AmIpFilter.java:doFilter(157)) - Could not find proxy-user cookie, so user will not be set”. Can anyone give me some help? Thank you! -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/WebUI-on-yarn-through-ssh-tunnel-affected-by-AmIpfilter-tp21540.html Sent from the Apache Spark User List mailing list archive at Nabble.com. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
[MLlib] Performance issues when building GBM models
Hi All, I wonder if anyone else has some experience building a Gradient Boosted Trees model using spark/mllib? I have noticed when building decent-size models that the process slows down over time. We observe that the time to build tree n is approximately a constant time longer than the time to build tree n-1 i.e. t(n) = t(n-1) + const. The implication is that the total build time goes as something like N^2, where N is the total number of trees. I would expect that the algorithm should be approximately linear in total time (i.e. each boosting iteration takes roughly the same time to complete). So I have a couple of questions: 1. Is this behaviour expected, or consistent with what others are seeing? 2. Does anyone know if there a tuning parameters (e.g. in the boosting strategy, or tree stategy) that may be impacting this? All aspects of the build seem to slow down as I go. Here's a random example culled from the logs, from the beginning and end of the model build: 15/02/09 17:22:11 INFO scheduler.DAGScheduler: Job 42 finished: count at DecisionTreeMetadata.scala:111, took 0.077957 s 15/02/09 19:44:01 INFO scheduler.DAGScheduler: Job 7954 finished: count at DecisionTreeMetadata.scala:111, took 5.495166 s Any thoughts or advice, or even suggestions on where to dig for more info would be welcome. thanks chris Christopher Thom QUANTIUM Level 25, 8 Chifley, 8-12 Chifley Square Sydney NSW 2000 T: +61 2 8222 3577 F: +61 2 9292 6444 W: quantium.com.auwww.quantium.com.au linkedin.com/company/quantiumwww.linkedin.com/company/quantium facebook.com/QuantiumAustraliawww.facebook.com/QuantiumAustralia twitter.com/QuantiumAUwww.twitter.com/QuantiumAU The contents of this email, including attachments, may be confidential information. If you are not the intended recipient, any use, disclosure or copying of the information is unauthorised. If you have received this email in error, we would be grateful if you would notify us immediately by email reply, phone (+ 61 2 9292 6400) or fax (+ 61 2 9292 6444) and delete the message from your system. - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
RE: no space left at worker node
Is there any way we can disable Spark copying the jar file to the corresponding directory. I have a fat jar and is already copied to worker nodes using the command copydir. Why Spark needs to save the jar to ./spark/work/appid each time a job get started? Ey-Chih Chow Date: Sun, 8 Feb 2015 20:09:32 -0800 Subject: Re: no space left at worker node From: 2dot7kel...@gmail.com To: eyc...@hotmail.com CC: gen.tan...@gmail.com; user@spark.apache.org I guess you may set the parameters below to clean the directories: spark.worker.cleanup.enabledspark.worker.cleanup.intervalspark.worker.cleanup.appDataTtl They are described here: http://spark.apache.org/docs/1.2.0/spark-standalone.html Kelvin On Sun, Feb 8, 2015 at 5:15 PM, ey-chih chow eyc...@hotmail.com wrote: I found the problem is, for each application, the Spark worker node saves the corresponding std output and std err under ./spark/work/appid, where appid is the id of the application. If I ran several applications in a row, it will out of space. In my case, the disk usage under ./spark/work/ is as follows: 1689784 ./app-20150208203033-0002/01689788 ./app-20150208203033-000240324 ./driver-20150208180505-00011691400 ./app-20150208180509-0001/01691404 ./app-20150208180509-000140316 ./driver-20150208203030-000240320 ./driver-20150208173156-1649876 ./app-20150208173200-/01649880 ./app-20150208173200-5152036. Any suggestion how to resolve it? Thanks. Ey-Chih ChowFrom: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 15:25:43 -0800 By this way, the input and output paths of the job are all in s3. I did not use paths of hdfs as input or output. Best regards, Ey-Chih Chow From: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 14:57:15 -0800 Hi Gen, Thanks. I save my logs in a file under /var/log. This is the only place to save data. Will the problem go away if I use a better machine? Best regards, Ey-Chih Chow Date: Sun, 8 Feb 2015 23:32:27 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc.In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. CheersGen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not.The reply of Micheal is good solution about this type of problem. You can check his site. CheersGen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception
Re: Installing a python library along with ec2 cluster
Hi I am very new both in spark and aws stuff.. Say, I want to install pandas on ec2.. (pip install pandas) How do I create the image and the above library which would be used from pyspark. Thanks On Sun, Feb 8, 2015 at 3:03 AM, gen tang gen.tan...@gmail.com wrote: Hi, You can make a image of ec2 with all the python libraries installed and create a bash script to export python_path in the /etc/init.d/ directory. Then you can launch the cluster with this image and ec2.py Hope this can be helpful Cheers Gen On Sun, Feb 8, 2015 at 9:46 AM, Chengi Liu chengi.liu...@gmail.com wrote: Hi, I want to install couple of python libraries (pip install python_library) which I want to use on pyspark cluster which are developed using the ec2 scripts. Is there a way to specify these libraries when I am building those ec2 clusters? Whats the best way to install these libraries on each ec2 node? Thanks
Re: Installing a python library along with ec2 cluster
You can basically add one function call to install the stuffs you want. If you look at the spark-ec2 script, there's a function which does all the setup named: setup_cluster(..) https://github.com/apache/spark/blob/master/ec2/spark_ec2.py#L625. Now, if you want to install a python library ( assuming pip is already installed), you can add one more line in the above function like: ssh(master, opts, pip install pandas) This will install it on the master node, you have slave_nodes variable which has all info of slave machines . You can iterate through it and do the same. Thanks Best Regards On Sun, Feb 8, 2015 at 2:16 PM, Chengi Liu chengi.liu...@gmail.com wrote: Hi, I want to install couple of python libraries (pip install python_library) which I want to use on pyspark cluster which are developed using the ec2 scripts. Is there a way to specify these libraries when I am building those ec2 clusters? Whats the best way to install these libraries on each ec2 node? Thanks
Re: no space left at worker node
Maybe, try with local: under the heading of Advanced Dependency Management here: https://spark.apache.org/docs/1.1.0/submitting-applications.html It seems this is what you want. Hope this help. Kelvin On Sun, Feb 8, 2015 at 9:13 PM, ey-chih chow eyc...@hotmail.com wrote: Is there any way we can disable Spark copying the jar file to the corresponding directory. I have a fat jar and is already copied to worker nodes using the command copydir. Why Spark needs to save the jar to ./spark/work/appid each time a job get started? Ey-Chih Chow -- Date: Sun, 8 Feb 2015 20:09:32 -0800 Subject: Re: no space left at worker node From: 2dot7kel...@gmail.com To: eyc...@hotmail.com CC: gen.tan...@gmail.com; user@spark.apache.org I guess you may set the parameters below to clean the directories: spark.worker.cleanup.enabled spark.worker.cleanup.interval spark.worker.cleanup.appDataTtl They are described here: http://spark.apache.org/docs/1.2.0/spark-standalone.html Kelvin On Sun, Feb 8, 2015 at 5:15 PM, ey-chih chow eyc...@hotmail.com wrote: I found the problem is, for each application, the Spark worker node saves the corresponding std output and std err under ./spark/work/appid, where appid is the id of the application. If I ran several applications in a row, it will out of space. In my case, the disk usage under ./spark/work/ is as follows: 1689784 ./app-20150208203033-0002/0 1689788 ./app-20150208203033-0002 40324 ./driver-20150208180505-0001 1691400 ./app-20150208180509-0001/0 1691404 ./app-20150208180509-0001 40316 ./driver-20150208203030-0002 40320 ./driver-20150208173156- 1649876 ./app-20150208173200-/0 1649880 ./app-20150208173200- 5152036 . Any suggestion how to resolve it? Thanks. Ey-Chih Chow -- From: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 15:25:43 -0800 By this way, the input and output paths of the job are all in s3. I did not use paths of hdfs as input or output. Best regards, Ey-Chih Chow -- From: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 14:57:15 -0800 Hi Gen, Thanks. I save my logs in a file under /var/log. This is the only place to save data. Will the problem go away if I use a better machine? Best regards, Ey-Chih Chow -- Date: Sun, 8 Feb 2015 23:32:27 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc. In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. Cheers Gen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow -- Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not. The reply of Micheal is good solution about this type of problem. You can check his site. Cheers Gen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: # LABEL=/ / ext4defaults,noatime 1 1 tmpfs /dev/shmtmpfs defaults0 0 devpts /dev/ptsdevpts gid=5,mode=620 0 0 sysfs /syssysfs defaults0 0 proc/proc procdefaults0 0 /dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0 /dev/sdc/mnt2 auto defaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow -- Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From:
RE: no space left at worker node
I found the problem is, for each application, the Spark worker node saves the corresponding std output and std err under ./spark/work/appid, where appid is the id of the application. If I ran several applications in a row, it will out of space. In my case, the disk usage under ./spark/work/ is as follows: 1689784 ./app-20150208203033-0002/01689788 ./app-20150208203033-000240324 ./driver-20150208180505-00011691400 ./app-20150208180509-0001/01691404 ./app-20150208180509-000140316 ./driver-20150208203030-000240320 ./driver-20150208173156-1649876 ./app-20150208173200-/01649880 ./app-20150208173200-5152036. Any suggestion how to resolve it? Thanks. Ey-Chih ChowFrom: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 15:25:43 -0800 By this way, the input and output paths of the job are all in s3. I did not use paths of hdfs as input or output. Best regards, Ey-Chih Chow From: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 14:57:15 -0800 Hi Gen, Thanks. I save my logs in a file under /var/log. This is the only place to save data. Will the problem go away if I use a better machine? Best regards, Ey-Chih Chow Date: Sun, 8 Feb 2015 23:32:27 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc.In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. CheersGen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not.The reply of Micheal is good solution about this type of problem. You can check his site. CheersGen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at
RE: no space left at worker node
By this way, the input and output paths of the job are all in s3. I did not use paths of hdfs as input or output. Best regards, Ey-Chih Chow From: eyc...@hotmail.com To: gen.tan...@gmail.com CC: user@spark.apache.org Subject: RE: no space left at worker node Date: Sun, 8 Feb 2015 14:57:15 -0800 Hi Gen, Thanks. I save my logs in a file under /var/log. This is the only place to save data. Will the problem go away if I use a better machine? Best regards, Ey-Chih Chow Date: Sun, 8 Feb 2015 23:32:27 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I am sorry that I made a mistake. r3.large has only one SSD which has been mounted in /mnt. Therefore this is no /dev/sdc.In fact, the problem is that there is no space in the under / directory. So you should check whether your application write data under this directory(for instance, save file in file:///). If not, you can use watch du -sh to during the running time to figure out which directory is expanding. Normally, only /mnt directory which is supported by SSD is expanding significantly, because the data of hdfs is saved here. Then you can find the directory which caused no space problem and find out the specific reason. CheersGen On Sun, Feb 8, 2015 at 10:45 PM, ey-chih chow eyc...@hotmail.com wrote: Thanks Gen. How can I check if /dev/sdc is well mounted or not? In general, the problem shows up when I submit the second or third job. The first job I submit most likely will succeed. Ey-Chih Chow Date: Sun, 8 Feb 2015 18:18:03 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, In fact, /dev/sdb is /dev/xvdb. It seems that there is no problem about double mount. However, there is no information about /mnt2. You should check whether /dev/sdc is well mounted or not.The reply of Micheal is good solution about this type of problem. You can check his site. CheersGen On Sun, Feb 8, 2015 at 5:53 PM, ey-chih chow eyc...@hotmail.com wrote: Gen, Thanks for your information. The content of /etc/fstab at the worker node (r3.large) is: #LABEL=/ / ext4defaults,noatime 1 1tmpfs /dev/shm tmpfs defaults0 0devpts /dev/ptsdevpts gid=5,mode=620 0 0sysfs /syssysfs defaults0 0proc/proc procdefaults0 0/dev/sdb/mntauto defaults,noatime,nodiratime,comment=cloudconfig 0 0/dev/sdc/mnt2 autodefaults,noatime,nodiratime,comment=cloudconfig 0 0 There is no entry of /dev/xvdb. Ey-Chih Chow Date: Sun, 8 Feb 2015 12:09:37 +0100 Subject: Re: no space left at worker node From: gen.tan...@gmail.com To: eyc...@hotmail.com CC: user@spark.apache.org Hi, I fact, I met this problem before. it is a bug of AWS. Which type of machine do you use? If I guess well, you can check the file /etc/fstab. There would be a double mount of /dev/xvdb.If yes, you should1. stop hdfs2. umount /dev/xvdb at / 3. restart hdfs Hope this could be helpful.CheersGen On Sun, Feb 8, 2015 at 8:16 AM, ey-chih chow eyc...@hotmail.com wrote: Hi, I submitted a spark job to an ec2 cluster, using spark-submit. At a worker node, there is an exception of 'no space left on device' as follows. == 15/02/08 01:53:38 ERROR logging.FileAppender: Error writing stream to file /root/spark/work/app-20150208014557-0003/0/stdout java.io.IOException: No space left on device at java.io.FileOutputStream.writeBytes(Native Method) at java.io.FileOutputStream.write(FileOutputStream.java:345) at org.apache.spark.util.logging.FileAppender.appendToFile(FileAppender.scala:92) at org.apache.spark.util.logging.FileAppender.appendStreamToFile(FileAppender.scala:72) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply$mcV$sp(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.logging.FileAppender$$anon$1$$anonfun$run$1.apply(FileAppender.scala:39) at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1311) at org.apache.spark.util.logging.FileAppender$$anon$1.run(FileAppender.scala:38) === The command df showed the following information at the worker node: Filesystem 1K-blocks Used Available Use% Mounted on /dev/xvda1 8256920 8256456 0 100% / tmpfs 7752012 0 7752012 0% /dev/shm /dev/xvdb 30963708 1729652 27661192 6% /mnt Does anybody know how to fix this? Thanks. Ey-Chih Chow -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/no-space-left-at-worker-node-tp21545.html