About darouwan

darouwan · ‎05-08-2019

I want to rebalance the hdfs datanode. However it throws following error Failed to move blk**** with size=**** from *** to ***through *** Get access token error, status message, block move is failed.

darouwan · ‎05-08-2019

One of datanodes in my cluster is down due to the storage being full. Then I clean some space, however, the datanode throws following error: Caught exception while adding replicas from /sata/hadoop/hdfs/data/current. Will throw later block pool BP-***** is not found at org.apache.hadoop.hdfs.server.datanode.fsdataset.imp.FsVolumeImpl.getBlockPoolSlice(FsVolumeImpl.java:423) .... BlockPoolSliceStorage.recoverTransitionRead:attempt to load an used block storage: /sata/hadoop/hdfs/data/current/BP-***** Then even I delete the whole hdfs data dir, and datanode make a new one automatically, it still throw the same error message. How to fix it?

darouwan · ‎04-26-2019

I have met the similar problem. I have move the block pool to another place, but the problem occurs as well. How to fix it? su -l hdfs -c "/usr/hdp/current/hadoop-hdfs-datanode/../hadoop/sbin/hadoop-daemon.sh start datanode" output following message: Java Hotshpot 64bit server VM warning: cannot open file /var/log/hadoop/hdfs/gc.log-2019... due to No such file or directory

darouwan · ‎03-12-2019

Thanks @Jordan Moore As 'Kafka is very memory and disk sensitive. ', do you recommend to install kafka brokers on a virtual machine, as I cannot have more dedicated machines for kafka?

darouwan · ‎03-11-2019

I meet a significant performance problem recently. I have about 30 spark streaming applications, which read data from kafka and write the data to hdfs. But recently the writing progress on some spark executor become very slow. The data amount for each spark tasks are similar, but the time cost of tasks are in great difference, where the slowest one is about 4 times of the fastest one. I have checked the disk usage, where disk use time on some hosts are about 80% to 90%. So I guess if it is caused by slow hdfs writing speed, because of my kafka broker, hdfs data node, yarn nodemanager locating on same hosts. So will it actually affect the performance?

darouwan · ‎01-10-2019

I need to merge some small parquet files into one larger file, as too many files on hdfs cost too large name node memory. There small parquet files have similar structures, but not same. when I use parquet-tools to merge them, it throw: could not merge metadata key org.apache.spark.sql.parquet.row.metadata has conflicting values Spark is a good way, but it's to slow comparing to parquet-tools Is there any good solutions? Thanks!

darouwan · ‎12-07-2018

Hi @Jagadeesan A S my current save mode is append. My sparking streaming apps will run every 5 min, it is not convenient to delete manually....So I think the better solution is customize the temp location. Or Can I set offset of the scheduled running time? For example, my current 2 apps every 5 minutes, that's run at 0, 5, 10, 15, 20 Can I set a schedule, make one still runs at 0, 5, 10 , 15, and another runs at 2.5, 7.5, 10.5?

darouwan · ‎12-06-2018

Thanks @Jagadeesan A S _temporary is a temp directory under path of the df.write.parquet(path) on hdfs. However spark.local.dir default value is /tmp, and in document, Directory to use for "scratch" space in Spark, including map output files and RDDs that get stored on disk. This should be on a fast, local disk in your system. So it should be a directory on local file system. I am not sure spark.local.dir refers to the temp directory of spark writing ...

darouwan · ‎12-06-2018

I have two spark applications writing data to one directory on HDFS, which cause the faster completed app will delete the working directory _temporary containing some temp file belonging to another app. So can I specify a _temporary directory for each Spark application？

darouwan · ‎12-04-2018

I have enabled HDFS name node ha(nn1, nn2), and the ranger is installed on original name node(nn1). Now I launch a hdfs namenode failover operation, the active namenode switches to the nn2 successfully. But the user sync plugin of Ranger is installed on nn1, cause the auth problem. Although I solve the problem by copy user and group info from nn1 to nn2, but, I think there should be another easier way to cover the hdfs ha switch situation. Is there any better solution for this? Thanks!

darouwan · ‎11-30-2018

I use kafka with default value(PLAINTEXT://localhost:9092), however, it broadcasts it's hostname instead of ip address. The result is, if the producer runs a host without the kafka host info, it cannot send messages to kafka. How to resolve it?

darouwan · ‎11-20-2018

OK @Jay Kumar SenSharma So can I upgrade the ambari metrics service individually , besides to the whole Ambari?

darouwan · ‎11-20-2018

for netstat -anlp | grep :6188 | grep CLOSE_WAIT | wc -l I get 0. May my restarting the service solve this problem temporarily. You mean the CLOSE_WAITING problem relates to Failed to invoke 'unbuffer' method inclassclass org.apache.hadoop.fs.FSDataInputStream ?

darouwan · ‎11-20-2018

I am using Ambari 2.6.1.0 The ambari metrics collector log is stored on production environment. I cannot export it. And I can give you the error log by hand typing copy: now In embedded mode, ambari metrics log: MetaDataProtos$MetaDataService for row \x00\x00METRIC_RECORD .... Caused by java.lang.InterruptedException ams hbase log: FSDataInputStreamWrapper: Failed to invoke 'unbuffer' method in class class org.apache.hadoop.fs.FSDataInputStream So there may be a TCP socket connection left open in CLOSE_WAIT state .... caused by java.lang.UnsupportedOperationException: this stream does not support ubbuffering All settings for ams are in default. There are 6 nodes in total in cluster. The host running ams has 64 cores and 256GB. Currently it has 53GB free memory and 233GB free memory in cache

darouwan · ‎11-19-2018

my ambari metric collector always doesn't work. In log, I found Failed to get result with timeout, timeout = 300000ms row 'METRIC_AGGREGATE' on table 'SYSTEM.CATALOG' at region=SYSTEM.CATALOG, ***** host name=[hostname], [port] *** seNum=6....causes by ....IOException: Failed to get result with timeout, timeout=300000ms then stuck at org.apache.hadoop.hbase.client.AsynProcess:#1 waiting for 28379 actions to finish It seems relate to hbase, but my hbase runs well without any error If I restart ambari metric collector, it recovers immediately, and become unavailable again after several hours. How to fix it? Thanks

darouwan · ‎11-14-2018

My ambari metrics collector always goes down after running one or two days. From the ambari metrics collector log, I find following message: java.sql.SQLException: Encountered exception in a sub plan[0] execution ..... caused by: java.sql.SQLException:Encountered exception in sub plan[0] execution ... caused by java.sql.SQLException: java.util.concurrent.TimeoutException at orig.apache.phoenix.cache.ServerCacheClient.addServerCache... It seems be related to phoenix. But from hbase page on ambari, I have no phoenix server. The mode is distributed.

darouwan · ‎11-13-2018

Thanks @KB And another question: When my spark application writing massive of data to hdfs, it always throws error message like following: No lease on /user/xx/sample_2016/_temporary/0/_temporary/attempt_201604141035_0058_m_019029_0/part-r-19029-1b93e1fa-9284-4f2c-821a-c83795ad27c1.gz.parquet:File does not exist.HolderDFSClient_NONMAPREDUCE_1239207978_115 does not have any open files. How to solve this problem? I search online and others said it is related to dfs.datanode.max.xcievers

darouwan · ‎11-13-2018

I increase the ipc max length according to this https://community.hortonworks.com/questions/101841/issue-requested-data-length-146629817-is-longer-th.html The hdfs service seems back to work.

darouwan · ‎11-13-2018

Thanks @KB I have reset the dfs.datanode.max.xcievers to 32768, is it still too high? I increase it to avoid "No lease on file (inode 5425306)" error. So what's the proper value for this property? If I set the value to a proper value, will the missing block be recovered automatically?

darouwan · ‎11-13-2018

Yesterday I add three more data nodes to my hdfs cluster with hdp 2.6.4. Few hours later, because of sparking writing error(No lease on...), I increase dfs.datanode.max.xcievers to 65536 and increase the heap size of name node and data node from 5G to 12G. And then restart it. However, the hdfs restart progress pauses in name node stage. It shows it is always in safe mode, and continues for 10 minutes. I force to leave the safe mode manually, and then hdfs reports a lot blocks are missing(about more than 90%). I checked the log of datanode and namenode, there are two kinds of error log: 1. In name node: Requested data length ** is longer than maximum configured RPC length ** 2. In data node: End of file exception between local host is "***", destination host is "**:8020" So how can I recovery my missing file? and what's the actual cause of this problem?

darouwan · ‎10-15-2018

Solve by using HttpFs. It set a gateway where no need to access data node.

darouwan · ‎10-15-2018

I am willing to open a file on hdfs via WebHDFS. As I cannot access the redirect location of datanode, I set a proxy to access it. However, I find it sometimes return the correct file content, but sometimes not. Instead, it returns 404 error for "Problem accessing {Path}", and I cannot find any rule, just like in random. How to resolve it? Thank you.

darouwan · ‎10-11-2018

I need to read files on hdfs which is distributed on another network, where I can access it via proxy. Now I can access the directories, files status via WebHDFS by specific port successfully. However when I try to read the file content, it redirects me to another data node url, while I only have access to name node. So how can I read the file in a convenient way?

darouwan · ‎10-10-2018

@Aditya Sirna So in default , there are up to 1000 lines of results stored on hdfs for each query? If I increase the limit, will it have some negative effects? Such as slow http transferring? Or result receiving failed?

darouwan · ‎10-10-2018

@Aditya Sirna Thanks Aditya So what about paging? Since the whole results are saved on hdfs in JSON format, if I need to load part of whole result, just load the whole json file and cut out part of it by given page size and page number in memory ? In practice for zeppelin, will it have out of memory problem if the size is too huge?

darouwan · ‎10-10-2018

I am working on designing a hdfs query system based on spark, which containing a paging function, and zeppelin seems be a good sample for me. Now I have a problem. I see spark or spark sql query results are existed even I refresh or reopen the notebook. So the results must be saved on some place. So I am wondering where these result data is saved on? If the data is saved on database, what if the result data size is pretty huge so that causing the database performance problem?

darouwan · ‎08-16-2018

Hi @Jonathan Sneep Fine. thanks. I have added the user and group info to my namenode. So the typical way to adding the new user or group is creating the user and group on namenode, and waiting for usersync to sync the user info to Ranger? So if I don't care the group policy, creating internal user in ranger and specifying them in allow conditions also works? At least it seems work in practice..

darouwan · ‎08-16-2018

Hi @Jonathan Sneep Not yet. So I need to add the user the related group on my namenode host manually?

darouwan · ‎08-15-2018

Hi, @Jonathan Sneep Thanks for your response. Actually both user and group are created in Ranger, which are internal for ranger

darouwan · ‎08-15-2018

I meet some problem in ranger authentication. Here is my step to represent it: 1. I create one account in ranger, where the username is test01 2. I set it belong to a group test_group01 3. In the ranger hdfs policy, I set test_group01 has access to the directory /data/ If it runs normally, the test01 user should have the access to /data/ from the privilege inheriting from the group "test_group01". But in practice, it cannot access the directory /data. However if I specify the 'select user' with the test01, it works well. So it seems that specifying the group in policy doesn't work, and specifying the permitted user is fine. How to solve it? Thanks!

Online	Offline
Last Visited	‎04-02-2018 12:02 AM

Date Registered	‎12-21-2017 12:43 AM
Last Visited	‎04-02-2018 12:02 AM
Total Messages Posted	67
Total Kudos Received	3

Re: Access file on hdfs via proxy

Re: NullPointerException when running spark on Zep...

Get access token error when re-balance the hdfs da...

Block pool is not found when start the data node

Re: datanode not starting

Re: Can kafka broker, yarn nodemanager, hdfs data ...

Can kafka broker, yarn nodemanager, hdfs data node...

How to merge parquet files with different structur...

Re: How to change Spark _temporary directory when ...

Re: How to change Spark _temporary directory when ...

How to change Spark _temporary directory when writ...

Ranger user sync problem when switch name node on ...

How to set kafka to broadcast the ip address rathe...

Re: No data available for Ambari Metric Collector

Re: No data available for Ambari Metric Collector

Re: No data available for Ambari Metric Collector

No data available for Ambari Metric Collector

Ambari metrics collector down with "Encounter exce...

Re: A lot of blocks missing in HDFS

Re: A lot of blocks missing in HDFS

Re: A lot of blocks missing in HDFS

A lot of blocks missing in HDFS

Re: Access file on hdfs via proxy

WebHDFS return error 404 in random when I open the...

Access file on hdfs via proxy

Re: How does zeppelin storage query result?

Re: How does zeppelin storage query result?

How does zeppelin storage query result?

Re: Ranger auth error in hdfs

Re: Ranger auth error in hdfs

Re: Ranger auth error in hdfs

Ranger auth error in hdfs