hive-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ashutosh Chauhan <hashut...@apache.org>
Subject Re: Understanding distributed Hive server and Hive Metastore setup
Date Thu, 25 Aug 2011 23:00:15 GMT
Christian,

Looks like its not possible to do the setup that you are looking for.
Problem arises since HiveServer extends HMSHandler directly instead of
accessing Metastore through HiveMetaStoreClient and because of this
metastore thrift interface is missed entirely. Hiveserver will contact mysql
directly and won't go through external metastore service as you have in your
diagram.  If you consider this as a blocker, please open up a jira for more
discussion.

Hope it helps,
Ashutosh

On Wed, Aug 24, 2011 at 23:21, Christian Kurz <crkurz@gmx.de> wrote:

> **
> Thanks, Edward and Ashutosh
>
> Ashutosh,
> yes, I do not understand why the service "hiveserver" still uses a Derby
> instance even through it should be talking to the service "metastore". Btw,
> if I run the hiveserver without having started the metastore service, the
> hiveserver complains when I try to let it execute a HiveQL command through
> JDBC:
>
> ...
> org.apache.hadoop.hive.ql.metadata.HiveException:
> MetaException(message:Could not connect to meta store using any of the URIs
> provided)
>         at
> org.apache.hadoop.hive.ql.metadata.Hive.getTablesByPattern(Hive.java:919)
> ...
> (full stacktrace at the end of this post)
>
> which is exactly what I expect and which makes me somewhat confident that I
> have configured things correctly.
>
> The entire issue came up, because the hiveserver service did not work, when
> started from the same directory, from which the metastore service had been
> started. It turned out that this was because both services were trying to
> setup a Derby instance in the current dir and therefore ran into a file
> locking situation. I have worked around this by starting the two services
> from different directories, but I am worried that I'd be missing an
> important point in my setup.
>
> When I run "pfiles <pid of hiveserver>" it lists these files for the
> hiveserver service (which should not need a Derby instance, as far as I
> understood):
>       ...tons of jars...
>       /home/hadoop/hive_admin/derby.log
>       /home/hadoop/hive_admin/metastore_db/log/log1.dat
>       /home/hadoop/hive_admin/metastore_db/dbex.lck
>       /home/hadoop/hive_admin/metastore_db/seg0/c191.dat
>       /home/hadoop/hive_admin/metastore_db/seg0/c1a1.dat
>       ...
>       /home/hadoop/hive_admin/metastore_db/seg0/c431.dat
>       /home/hadoop/hive_admin/metastore_db/seg0/c451.dat
>
> Any pointers appreciated. If anybody things this is a bug, I can file one.
>
> Thanks,
> Christian
>
>
> full stacktrace:
>
> Hive history
> file=/tmp/hadoop/hive_job_log_hadoop_201108242305_155100916.txt
> FAILED: Error in semantic analysis: Table not found weblog
> org.apache.hadoop.hive.ql.metadata.HiveException:
> MetaException(message:Could not connect to meta store using any of the URIs
> provided)
>         at
> org.apache.hadoop.hive.ql.metadata.Hive.getTablesByPattern(Hive.java:919)
>         at
> org.apache.hadoop.hive.ql.metadata.Hive.getTablesByPattern(Hive.java:904)
>         at
> org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.analyzeCreateTable(SemanticAnalyzer.java:7074)
>         at
> org.apache.hadoop.hive.ql.parse.SemanticAnalyzer.analyzeInternal(SemanticAnalyzer.java:6573)
>         at
> org.apache.hadoop.hive.ql.parse.BaseSemanticAnalyzer.analyze(BaseSemanticAnalyzer.java:238)
>         at org.apache.hadoop.hive.ql.Driver.compile(Driver.java:340)
>         at org.apache.hadoop.hive.ql.Driver.run(Driver.java:736)
>         at
> org.apache.hadoop.hive.service.HiveServer$HiveServerHandler.execute(HiveServer.java:116)
>         at
> org.apache.hadoop.hive.service.ThriftHive$Processor$execute.process(ThriftHive.java:699)
>         at
> org.apache.hadoop.hive.service.ThriftHive$Processor.process(ThriftHive.java:677)
>         at
> org.apache.thrift.server.TThreadPoolServer$WorkerProcess.run(TThreadPoolServer.java:253)
>         at
> java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
>         at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
>         at java.lang.Thread.run(Thread.java:619)
> Caused by: MetaException(message:Could not connect to meta store using any
> of the URIs provided)
>         at
> org.apache.hadoop.hive.metastore.HiveMetaStoreClient.open(HiveMetaStoreClient.java:183)
>         at
> org.apache.hadoop.hive.metastore.HiveMetaStoreClient.<init>(HiveMetaStoreClient.java:151)
>         at
> org.apache.hadoop.hive.ql.metadata.Hive.createMetaStoreClient(Hive.java:1855)
>         at org.apache.hadoop.hive.ql.metadata.Hive.getMSC(Hive.java:1865)
>         at
> org.apache.hadoop.hive.ql.metadata.Hive.getTablesByPattern(Hive.java:917)
>         ... 13 more
> FAILED: Error in metadata: MetaException(message:Could not connect to meta
> store using any of the URIs provided)
> FAILED: Execution Error, return code 1 from
> org.apache.hadoop.hive.ql.exec.DDLTask
>
>
>
>
> On 25.08.2011 01:29, Ashutosh Chauhan wrote:
>
> Edward,
>
>  Apart from recommended best practices what Christian is asking for is why
> HiveServer is still trying to interact with local db instance even after
> setting the config variables. AFAIK it should not. Christian, you found that
> out by looking at files opened by HiveServer jvm. Can you provide more info
> there like how did you find that out and which these files are?
>
>  Ashutosh
>
> On Wed, Aug 24, 2011 at 14:20, Edward Capriolo <edlinuxguru@gmail.com>wrote:
>
>>
>>
>> On Wed, Aug 24, 2011 at 3:02 PM, Christian Kurz <crkurz@gmx.de> wrote:
>>
>>>
>>> Thanks for the quick reply, Edward
>>>
>>>  I am not sure I got you: My HiveService has been started with hive.metastore.local=false.
>>> So shouldn't it use thrift instead of its own local Derby instance?
>>>
>>>  Thanks,
>>> Christian
>>>
>>> Am 24.08.2011 um 19:33 schrieb Edward Capriolo <edlinuxguru@gmail.com>:
>>>
>>>
>>>
>>> On Wed, Aug 24, 2011 at 10:53 AM, Christian Kurz <crkurz@gmx.de> wrote:
>>>
>>>>  Greetings,
>>>>
>>>> could somebody confirm/correct my understanding of a fully distributed
>>>> Hive setup, please?
>>>>
>>>> My setup is as follows
>>>>
>>>>    - *Java application using Hive JDBC driver *connects to
>>>>     - *hive --service hiveserver*, which connects to
>>>>     - *hive --service metastore*, which uses an embedded Derby database
>>>>    for metadata storage
>>>>
>>>> Please find more details in the image attached.
>>>>
>>>> The thing I find confusing is that JVM2 (Hive Server) starts up a Derby
>>>> database instance. I can see that from the files the JVM has opened.
>>>>
>>>> Does anybody know, why the Hive Server needs a Derby instance even
>>>> though hive-site.xml says: hive.metastore.local=false ?
>>>>
>>>> Any hints are much appreciated.
>>>>
>>>> Thanks,
>>>> Christian
>>>>
>>>> btw,
>>>> I have not been able to access the picture on the wiki<https://cwiki.apache.org/Hive/adminmanual-metastoreadmin.html#AdminManualMetastoreAdmin-MetastoreDeploymentOptionsinPictures>.
>>>> ("Not permitted"; even though I have registered on the wiki)
>>>>
>>>>
>>>>
>>>  hive.metastore.local is really misnamed.
>>>
>>>  local=true means communicate using datanucleus/JPOX and talking
>>> directly to the metastore.
>>>
>>>  local=false means use thrift which is essentially a level of
>>> indirection.
>>>
>>>    Talking about HiveService can confuse things because HiveService is a
>> different thrift interface.
>>
>>  You could be setup like this:
>>  HiveServiceClient->HiveService->metastore.local=true->derby
>> or
>>
>> HiveServiceClient->HiveService->metastore.local=false>thrift->hive_metastore
>>
>>  most people are setup like this:
>>
>>  HiveServiceClient->HiveService->metastore.local=true->mysql
>>  cli->metastore.local=true->mysql
>>
>>
>>
>

Mime
View raw message