By "expensive" I mean "seriously?"
On Thu, May 2, 2013 at 2:32 PM, Michael Segel <[email protected]>wrote: > > On May 2, 2013, at 4:18 PM, Andrew Purtell <[email protected]> wrote: > > > Sorry, hit send too soon. I would recommend the following instance types: > > > > hi1.4xlarge: Expensive but it has a comfortable level of resources and > > will perform > > Yeah at a spot price of $3.00 an hour per server? > Its expensive and fast. Note that you will want to up the number of slots > from the default 2 that are set up. ;-) > More tuning is recommended. (Ooops! That's for EMR not just EC2) > > > > hs1.8xlarge: This is what you might see in a typical data center > Hadoop > > deployment, also expensive > > m2.2xlarge/m2.4xlarge: Getting up to the amount of RAM you want for > > caching in "big data" workloads > > m1.xlarge: Less CPU but more RAM than c1.xlarge, so safer > > c1.xlarge: Only if you really know what you are doing and need to be > > cheap > > Anything lesser endowed: Never > > > > You may find that, relative to AWS charges for a hi1.4xlarge, some other > > hosting option for the equivalent would be more economical. > > > > > > On Thu, May 2, 2013 at 2:12 PM, Andrew Purtell <[email protected]> > wrote: > > > >>> OS is Ubuntu 12.04 and instance type is c1.medium > >> > >> Eeek! > >> > >> You shouldn't use less than c1.xlarge for running Hadoop+HBase on EC2. A > >> c1.medium has only 7 GB of RAM in total. > >> > >> > >> On Thu, May 2, 2013 at 1:53 PM, Loic Talon <[email protected]> wrote: > >> > >>> Hi Andrew, > >>> Thanks for those responses. > >>> > >>> The server has been deployed by Cloudera Manager. > >>> OS is Ubuntu 12.04 and instance type is c1.medium. > >>> Instance store are used, not EBS. > >>> > >>> It's possible that this problem is a memory problem ? > >>> Because when region server hab been started I have in stdout.log : > >>> > >>> Thu May 2 17:01:10 UTC 2013 > >>> using /usr/lib/jvm/j2sdk1.6-oracle as JAVA_HOME > >>> using 4 as CDH_VERSION > >>> using as HBASE_HOME > >>> using /run/cloudera-scm-agent/process/381-hbase-REGIONSERVER as > >>> HBASE_CONF_DIR > >>> using /run/cloudera-scm-agent/process/381-hbase-REGIONSERVER as > >>> HADOOP_CONF_DIR > >>> using as HADOOP_HOME > >>> > >>> But when I have the problem, I have in stdout.log : > >>> Thu May 2 17:01:10 UTC 2013 > >>> using /usr/lib/jvm/j2sdk1.6-oracle as JAVA_HOME > >>> using 4 as CDH_VERSION > >>> using as HBASE_HOME > >>> using /run/cloudera-scm-agent/process/381-hbase-REGIONSERVER as > >>> HBASE_CONF_DIR > >>> using /run/cloudera-scm-agent/process/381-hbase-REGIONSERVER as > >>> HADOOP_CONF_DIR > >>> using as HADOOP_HOME > >>> # > >>> # java.lang.OutOfMemoryError: Java heap space > >>> # -XX:OnOutOfMemoryError="kill -9 %p" > >>> # Executing /bin/sh -c "kill -9 20140"... > >>> > >>> Thanks > >>> > >>> Loic > >>> > >>> > >>> > >>> > >>> > >>> > >>> Loïc TALON > >>> > >>> > >>> [email protected] <http://teads.tv/> > >>> Video Ads Solutions > >>> > >>> > >>> > >>> 2013/5/2 Andrew Purtell <[email protected]> > >>> > >>>> Every instance type except t1.micro has a certain number of instance > >>>> storage (locally attached disk) volumes available, 1, 2, or 4 > depending > >>> on > >>>> type. > >>>> > >>>> As you probably know, you can use or create AMIs backed by > >>> instance-store, > >>>> in which the OS image is constructed on locally attached disk by a > >>> parallel > >>>> fetch process from slices of the root volume image stored in S3, or > >>> backed > >>>> by EBS, in which case the OS image is an EBS volume and attached over > >>> the > >>>> network, like a SAN. > >>>> > >>>> If you launch an Amazon Linux instance store backed instance the first > >>>> "ephemeral" local volume will be automatically attached on > >>>> /media/ephemeral0. That's where that term comes from, it's a synonym > for > >>>> instance-store. (You can by the way tell CloudInit via directives sent > >>> over > >>>> instance data to mount all of them.) > >>>> > >>>> If you have an EBS backed instance the default is to NOT attach any of > >>>> these volumes. > >>>> > >>>> If you are launching your instance with the Amazon Web console, in the > >>>> volume configuration part you can set up instance-store aka > "ephemeral" > >>>> mounts whether it is instance-store backed or EBS backed. > >>>> > >>>> Sorry I can't get into more background on this. Hope it helps. > >>>> > >>>> > >>>> > >>>> On Thu, May 2, 2013 at 1:23 PM, Jean-Marc Spaggiari < > >>>> [email protected] > >>>>> wrote: > >>>> > >>>>> Hi Andrew, > >>>>> > >>>>> No, this AWS instance is configured with instance stores too. > >>>>> > >>>>> What do you mean by "ephemeral"? > >>>>> > >>>>> JM > >>>>> > >>>>> 2013/5/2 Andrew Purtell <[email protected]> > >>>>> > >>>>>> Oh, I have faced issues with Hadoop on AWS personally. :-) But not > >>> this > >>>>>> one. I use instance-store aka "ephemeral" volumes for DataNode block > >>>>>> storage. Are you by chance using EBS? > >>>>>> > >>>>>> > >>>>>> On Thu, May 2, 2013 at 1:10 PM, Jean-Marc Spaggiari < > >>>>>> [email protected] > >>>>>>> wrote: > >>>>>> > >>>>>>> But that's wierld. This instance is running on AWS. If there > >>> issues > >>>>> with > >>>>>>> Hadoop and AWS I think some other people will have faced it before > >>>> me. > >>>>>>> > >>>>>>> Ok. I will move the discussion on the Hadoop mailing list since it > >>>>> seems > >>>>>> to > >>>>>>> be more related to hadoop vs OS. > >>>>>>> > >>>>>>> Thank, > >>>>>>> > >>>>>>> JM > >>>>>>> > >>>>>>> 2013/5/2 Andrew Purtell <[email protected]> > >>>>>>> > >>>>>>>>> 2013-05-02 14:02:41,063 INFO org.apache.hadoop.hdfs.DFSClient: > >>>>>>> Exception > >>>>>>>> in > >>>>>>>> createBlockOutputStream java.io.EOFException: Premature EOF: no > >>>>> length > >>>>>>>> prefix available > >>>>>>>> > >>>>>>>> The DataNode aborted the block transfer. > >>>>>>>> > >>>>>>>>> 2013-05-02 14:02:41,063 ERROR org.apache.hadoop.hdfs.server. > >>>>>>>> datanode.DataNode: > >>>>>>>> ip-10-238-38-193.eu-west-1.compute.internal:50010:DataXceiver > >>>>>>>> error processing WRITE_BLOCK operation src: / > >>> 10.238.38.193:39831 > >>>>> dest: > >>>>>>> / > >>>>>>>> 10.238.38.193:50010 java.io.FileNotFoundException: > >>>>>>> /mnt/dfs/dn/current/BP- > >>>>>>>> 1179773663-10.238.38.193-1363960970263/current/rbw/blk_ > >>>>>>>> 7082931589039745816_1955950.meta (Invalid argument) > >>>>>>>>> at java.io.RandomAccessFile.open(Native Method) > >>>>>>>>> at > >>>>> java.io.RandomAccessFile.<init>(RandomAccessFile.java:216) > >>>>>>>> > >>>>>>>> This looks like the native (OS level) side of RAF got EINVAL > >>> back > >>>>> from > >>>>>>>> create() or open(). Go from there. > >>>>>>>> > >>>>>>>> > >>>>>>>> > >>>>>>>> On Thu, May 2, 2013 at 12:27 PM, Jean-Marc Spaggiari < > >>>>>>>> [email protected]> wrote: > >>>>>>>> > >>>>>>>>> Hi, > >>>>>>>>> > >>>>>>>>> Any idea what can be the cause of a "Premature EOF: no length > >>>>> prefix > >>>>>>>>> available" error? > >>>>>>>>> > >>>>>>>>> 2013-05-02 14:02:41,063 INFO org.apache.hadoop.hdfs.DFSClient: > >>>>>>> Exception > >>>>>>>> in > >>>>>>>>> createBlockOutputStream > >>>>>>>>> java.io.EOFException: Premature EOF: no length prefix > >>> available > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.protocol.HdfsProtoUtil.vintPrefixed(HdfsProtoUtil.java:171) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.DFSOutputStream$DataStreamer.createBlockOutputStream(DFSOutputStream.java:1105) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.DFSOutputStream$DataStreamer.nextBlockOutputStream(DFSOutputStream.java:1039) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.DFSOutputStream$DataStreamer.run(DFSOutputStream.java:487) > >>>>>>>>> 2013-05-02 14:02:41,064 INFO org.apache.hadoop.hdfs.DFSClient: > >>>>>>> Abandoning > >>>>>>>>> > >>>>>>> > >>>>> > >>> > BP-1179773663-10.238.38.193-1363960970263:blk_7082931589039745816_1955950 > >>>>>>>>> 2013-05-02 14:02:41,068 INFO org.apache.hadoop.hdfs.DFSClient: > >>>>>>> Excluding > >>>>>>>>> datanode 10.238.38.193:50010 > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> I'm getting that on a server start. Logs are splitted > >>> correctly, > >>>>>>>>> coprocessors deployed corretly, and then I'm getting this > >>>>> exception. > >>>>>>> It's > >>>>>>>>> excluding the datanode, and because of that almost everything > >>>>>> remaining > >>>>>>>> is > >>>>>>>>> failing. > >>>>>>>>> > >>>>>>>>> There is only one server in this "cluster"... But even so, it > >>>>> should > >>>>>> be > >>>>>>>>> working. There is one master, one RS, one NN and one DN. On a > >>> AWS > >>>>>> host. > >>>>>>>>> > >>>>>>>>> At the same time on the hadoop datanode side I'm getting that: > >>>>>>>>> > >>>>>>>>> 2013-05-02 14:02:41,063 INFO > >>>>>>>>> org.apache.hadoop.hdfs.server.datanode.DataNode: opWriteBlock > >>>>>>>>> > >>>>>>> > >>>>> > >>> > BP-1179773663-10.238.38.193-1363960970263:blk_7082931589039745816_1955950 > >>>>>>>>> received exception java.io.FileNotFoundException: > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > /mnt/dfs/dn/current/BP-1179773663-10.238.38.193-1363960970263/current/rbw/blk_7082931589039745816_1955950.meta > >>>>>>>>> (Invalid argument) > >>>>>>>>> 2013-05-02 14:02:41,063 ERROR > >>>>>>>>> org.apache.hadoop.hdfs.server.datanode.DataNode: > >>>>>>>>> ip-10-238-38-193.eu-west-1.compute.internal:50010:DataXceiver > >>>> error > >>>>>>>>> processing WRITE_BLOCK operation src: /10.238.38.193:39831 > >>> dest: > >>>>> / > >>>>>>>>> 10.238.38.193:50010 > >>>>>>>>> java.io.FileNotFoundException: > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > /mnt/dfs/dn/current/BP-1179773663-10.238.38.193-1363960970263/current/rbw/blk_7082931589039745816_1955950.meta > >>>>>>>>> (Invalid argument) > >>>>>>>>> at java.io.RandomAccessFile.open(Native Method) > >>>>>>>>> at > >>>>> java.io.RandomAccessFile.<init>(RandomAccessFile.java:216) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.server.datanode.ReplicaInPipeline.createStreams(ReplicaInPipeline.java:187) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.server.datanode.BlockReceiver.<init>(BlockReceiver.java:199) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.server.datanode.DataXceiver.writeBlock(DataXceiver.java:457) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.protocol.datatransfer.Receiver.opWriteBlock(Receiver.java:103) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.protocol.datatransfer.Receiver.processOp(Receiver.java:67) > >>>>>>>>> at > >>>>>>>>> > >>>>>>>>> > >>>>>>>> > >>>>>>> > >>>>>> > >>>>> > >>>> > >>> > org.apache.hadoop.hdfs.server.datanode.DataXceiver.run(DataXceiver.java:221) > >>>>>>>>> at java.lang.Thread.run(Thread.java:662) > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> Does is sound more an hadoop issue than an HBase one? > >>>>>>>>> > >>>>>>>>> JM > >>>>>>>>> > >>>>>>>> > >>>>>>>> > >>>>>>>> > >>>>>>>> -- > >>>>>>>> Best regards, > >>>>>>>> > >>>>>>>> - Andy > >>>>>>>> > >>>>>>>> Problems worthy of attack prove their worth by hitting back. - > >>> Piet > >>>>>> Hein > >>>>>>>> (via Tom White) > >>>>>>>> > >>>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>> -- > >>>>>> Best regards, > >>>>>> > >>>>>> - Andy > >>>>>> > >>>>>> Problems worthy of attack prove their worth by hitting back. - Piet > >>>> Hein > >>>>>> (via Tom White) > >>>>>> > >>>>> > >>>> > >>>> > >>>> > >>>> -- > >>>> Best regards, > >>>> > >>>> - Andy > >>>> > >>>> Problems worthy of attack prove their worth by hitting back. - Piet > Hein > >>>> (via Tom White) > >>>> > >>> > >> > >> > >> > >> -- > >> Best regards, > >> > >> - Andy > >> > >> Problems worthy of attack prove their worth by hitting back. - Piet Hein > >> (via Tom White) > >> > > > > > > > > -- > > Best regards, > > > > - Andy > > > > Problems worthy of attack prove their worth by hitting back. - Piet Hein > > (via Tom White) > > -- Best regards, - Andy Problems worthy of attack prove their worth by hitting back. - Piet Hein (via Tom White)
