Apache Spark 2.3.1 with Hive metastore 3.1.0

Question

We have upgraded HDP cluster to 3.1.1.3.0.1.0-187 and have discovered:

1. Hive has a new metastore location
2. Spark can't see Hive databases

In fact we see:

org.apache.spark.sql.catalyst.analysis.NoSuchDatabaseException: Database ... not found

Could you help me understanding what has happened and how to solve this?

Update:

Configuration:

(spark.sql.warehouse.dir,/warehouse/tablespace/external/hive/) (spark.admin.acls,) (spark.yarn.dist.files,file:///opt/folder/config.yml,file:///opt/jdk1.8.0_172/jre/lib/security/cacerts) (spark.history.kerberos.keytab,/etc/security/keytabs/spark.service.keytab) (spark.io.compression.lz4.blockSize,128kb) (spark.executor.extraJavaOptions,-Djavax.net.ssl.trustStore=cacerts) (spark.history.fs.logDirectory,hdfs:///spark2-history/) (spark.io.encryption.keygen.algorithm,HmacSHA1) (spark.sql.autoBroadcastJoinThreshold,26214400) (spark.eventLog.enabled,true) (spark.shuffle.service.enabled,true) (spark.driver.extraLibraryPath,/usr/hdp/current/hadoop-client/lib/native:/usr/hdp/current/hadoop-client/lib/native/Linux-amd64-64) (spark.ssl.keyStore,/etc/security/serverKeys/server-keystore.jks) (spark.yarn.queue,default) (spark.jars,file:/opt/folder/component-assembly-0.1.0-SNAPSHOT.jar) (spark.ssl.enabled,true) (spark.sql.orc.filterPushdown,true) (spark.shuffle.unsafe.file.output.buffer,5m) (spark.yarn.historyServer.address,master2.env.project:18481) (spark.ssl.trustStore,/etc/security/clientKeys/all.jks) (spark.app.name,com.company.env.component.MyClass) (spark.sql.hive.metastore.jars,/usr/hdp/current/spark2-client/standalone-metastore/*) (spark.io.encryption.keySizeBits,128) (spark.driver.memory,2g) (spark.executor.instances,10) (spark.history.kerberos.principal,spark/[email protected]) (spark.unsafe.sorter.spill.reader.buffer.size,1m) (spark.ssl.keyPassword,*********(redacted)) (spark.ssl.keyStorePassword,*********(redacted)) (spark.history.fs.cleaner.enabled,true) (spark.shuffle.io.serverThreads,128) (spark.sql.hive.convertMetastoreOrc,true) (spark.submit.deployMode,client) (spark.sql.orc.char.enabled,true) (spark.master,yarn) (spark.authenticate.enableSaslEncryption,true) (spark.history.fs.cleaner.interval,7d) (spark.authenticate,true) (spark.history.fs.cleaner.maxAge,90d) (spark.history.ui.acls.enable,true) (spark.acls.enable,true) (spark.history.provider,org.apache.spark.deploy.history.FsHistoryProvider) (spark.executor.extraLibraryPath,/usr/hdp/current/hadoop-client/lib/native:/usr/hdp/current/hadoop-client/lib/native/Linux-amd64-64) (spark.executor.memory,2g) (spark.io.encryption.enabled,true) (spark.shuffle.file.buffer,1m) (spark.eventLog.dir,hdfs:///spark2-history/) (spark.ssl.protocol,TLS) (spark.dynamicAllocation.enabled,true) (spark.executor.cores,3) (spark.history.ui.port,18081) (spark.sql.statistics.fallBackToHdfs,true) (spark.repl.local.jars,file:///opt/folder/postgresql-42.2.2.jar,file:///opt/folder/ojdbc6.jar) (spark.ssl.trustStorePassword,*********(redacted)) (spark.history.ui.admin.acls,) (spark.history.kerberos.enabled,true) (spark.shuffle.io.backLog,8192) (spark.sql.orc.impl,native) (spark.ssl.enabledAlgorithms,TLS_RSA_WITH_AES_128_CBC_SHA,TLS_RSA_WITH_AES_256_CBC_SHA) (spark.sql.orc.enabled,true) (spark.yarn.dist.jars,file:///opt/folder/postgresql-42.2.2.jar,file:///opt/folder/ojdbc6.jar) (spark.sql.hive.metastore.version,3.0)

And from hive-site.xml:

<property>
  <name>hive.metastore.warehouse.dir</name>
  <value>/warehouse/tablespace/managed/hive</value>
</property>

Code looks like:

val spark = SparkSession
  .builder()
  .appName(getClass.getSimpleName)
  .enableHiveSupport()
  .getOrCreate()
...
dataFrame.write
  .format("orc")
  .options(Map("spark.sql.hive.convertMetastoreOrc" -> true.toString))
  .mode(SaveMode.Append)
  .saveAsTable("name")

Spark-submit:

    --master yarn \
    --deploy-mode client \
    --driver-memory 2g \
    --driver-cores 4 \
    --executor-memory 2g \
    --num-executors 10 \
    --executor-cores 3 \
    --conf "spark.dynamicAllocation.enabled=true" \
    --conf "spark.shuffle.service.enabled=true" \
    --conf "spark.executor.extraJavaOptions=-Djavax.net.ssl.trustStore=cacerts" \
    --conf "spark.sql.warehouse.dir=/warehouse/tablespace/external/hive/" \
    --jars postgresql-42.2.2.jar,ojdbc6.jar \
    --files config.yml,/opt/jdk1.8.0_172/jre/lib/security/cacerts \
    --verbose \
    component-assembly-0.1.0-SNAPSHOT.jar \

Ranice

Could you try passing the hive.xml location in the spark-submit as --file command ? – Balcke

Can you check the value of spark.sql.warehouse.dir and perhaps hive.metastore.warehouse.dir? Could you include the Environment tab from web UI in the question? You can always use hive-site.xml on CLASSPATH to point to the directory. – Preestablish

BTW I can't seem to find the version of HDP at docs.hortonworks.com. The latest seems HDP-3.0.1. I'm a bit confused. – Preestablish

Thank you for quick responce, guys. Jacek, this build: repo.hortonworks.com/content/repositories/releases/org/apache/… – Ranice

How do you access a Hive table? Can you show the exact query (e.g. spark.read...)? What's the directory of the Hive warehouse? Can you check all the HADOOP_-, YARN_- or HIVE_-related environment variables? – Preestablish

Jacek, what Hadoop, YARN and Hive variables I should check? – Ranice

Avishek, manual adding of hive-site.xml doesn't help. – Ranice

Any ideas, guys? – Ranice

Answer 1

Looks like this is a not implemented Spark feature. But the only one way to use Spark and Hive since 3.0 that I found is to use HiveWarehouseConnector from Horton. Documentation here. And good guide from Horton Community here. I leave the question unanswered until Spark developers have prepared an own solution.

Ranice

Answer 2

I've got a bit of a throwback trick for this one although disclaimer, it bypasses the ranger permissions (don't blame me if you incur the wrath of an admin).

To use with the spark-shell

export HIVE_CONF_DIR=/usr/hdp/current/hive-client/conf
spark-shell --conf "spark.driver.extraClassPath=/usr/hdp/current/hive-client/conf"

To use with sparklyR

Sys.setenv(HIVE_CONF_DIR="/usr/hdp/current/hive-client/conf")
conf = spark_config()
conf$'sparklyr.shell.driver-class-path' = '/usr/hdp/current/hive-client/conf'

It should work for the thriftserver too but I have not tested.

Niggling