Asynchronous Query

Asynchronous query supports users to execute SQL queries asynchronously and provides a more efficient way to export data. For example, if the result set of a SQL query Too large (million results) or SQL execution time is too long, through asynchronous query, the query result set can be exported efficiently to realize self-service data retrieval Various application scenarios.

Currently, asynchronous query only supports calling REST API. For how to use asynchronous query API, please read-Asynchronous Query API.

Configure the retention time for asynchronous query results

Asynchronous query supports the following configuration in kylin.properties:

• kylin.query.async.result-retain-days=7d: The retention time of asynchronous query results on HDFS. The default is 7 days, that is, asynchronous query results and related files older than 7 days will be cleaned up.

Configure a separate cluster queue for asynchronous query

In general, the same cluster queue can be used for asynchronous query and normal query. In some advanced scenarios, if you want to prevent asynchronous queries from affecting ordinary queries, you can deploy a separate queue for asynchronous queries. The specific configuration method is as follows:

1. Enable asynchronous query to deploy a separate cluster queue setting. Set kylin.query.unique-async-query-yarn-queue-enabled to true. Support project-level configuration and system-level configuration. The priority of project-level configuration is higher than system-level configuration. If neither is configured, asynchronous query and normal query use the same cluster queue.

2. Specify the queue used for asynchronous queries. Three levels are supported for designation, the priority from high to low is as follows:

   • Query level, specified by API request parameter spark_queue

   • Project level, specified by setting kylin.query.async-query.spark-conf.spark.yarn.queue

   • System level, specified by setting kylin.query.async-query.spark-conf.spark.yarn.queue in the configuration file /conf/kylin.properties

   Tip: If none of the three are configured, the default queue is default

3. Set configuration: kylin.query.async-query.submit-hadoop-conf-dir=$KYLIN_HOME/async_query_hadoop_conf

4. Put the hadoop configuration of the asynchronous query cluster into the $KYLIN_HOME/async_query_hadoop_conf directory, and put the hive-site.xml for building the cluster into this directory. If Kerberos authentication is enabled, you need to copy the krb5.conf file to the $KYLIN_HOME/async_query_hadoop_conf directory.

5. If Kerberos authentication is enabled between asynchronous query cluster, query cluster, and build cluster, the following additional configuration is required:

    kylin.storage.columnar.spark-conf.spark.yarn.access.hadoopFileSystems=hdfs://readcluster,hdfs://asyncquerycluster,hdfs://writecluster
    kylin.query.async-query.spark-conf.spark.yarn.access.hadoopFileSystems=hdfs://readcluster,hdfs://asyncquerycluster,hdfs://writecluster
    kylin.engine.spark-conf.spark.yarn.access.hadoopFileSystems=hdfs://readcluster,hdfs://asyncquerycluster,hdfs://writecluster

6. In general, the above configuration can meet the requirements. In some more advanced scenarios, you can configure spark related configurations in kylin.properties to achieve more fine-grained control with guidance of Kylin expert.

   The configuration starts with kylin.query.async-query.spark-conf, as shown below:

kylin.query.async-query.spark-conf.spark.yarn.queue=default
kylin.query.async-query.spark-conf.spark.executor.extraJavaOptions=-Dhdp.version=current -Dlog4j.configuration=spark-executor-log4j.properties -Dlog4j.debug -Dkylin.hdfs.working.dir=$ {kylin.env.hdfs-working-dir} -Dkap.metadata.identifier=${kylin.metadata.url.identifier} -Dkap.spark.category=sparder -Dkap.spark.project=${job.project}- Dkap.spark.mountDir=${kylin.tool.mount-spark-log-dir} -XX:MaxDirectMemorySize=896M
kylin.query.async-query.spark-conf.spark.yarn.am.extraJavaOptions=-Dhdp.version=current
kylin.query.async-query.spark-conf.spark.driver.extraJavaOptions=-Dhdp.version=current
kylin.query.async-query.spark-conf.spark.port.maxRetries=128
kylin.query.async-query.spark-conf.spark.driver.memory=4096m
kylin.query.async-query.spark-conf.spark.sql.driver.maxCollectSize=3600m
kylin.query.async-query.spark-conf.spark.executor.memory=12288m
kylin.query.async-query.spark-conf.spark.executor.memoryOverhead=3072m
kylin.query.async-query.spark-conf.spark.yarn.am.memory=1024m
kylin.query.async-query.spark-conf.spark.executor.cores=5
kylin.query.async-query.spark-conf.spark.executor.instances=4
kylin.query.async-query.spark-conf.spark.task.maxFailures=1
kylin.query.async-query.spark-conf.spark.ui.port=4041
kylin.query.async-query.spark-conf.spark.locality.wait=0s
kylin.query.async-query.spark-conf.spark.sql.dialect=hiveql
kylin.query.async-query.spark-conf.spark.sql.constraintPropagation.enabled=false
kylin.query.async-query.spark-conf.spark.ui.retainedStages=300
kylin.query.async-query.spark-conf.spark.hadoop.yarn.timeline-service.enabled=false
kylin.query.async-query.spark-conf.spark.hadoop.hive.exec.scratchdir=${kylin.env.hdfs-working-dir}/hive-scratch
kylin.query.async-query.spark-conf.hive.execution.engine=MR
kylin.query.async-query.spark-conf.spark.sql.crossJoin.enabled=true
kylin.query.async-query.spark-conf.spark.broadcast.autoClean.enabled=true
kylin.query.async-query.spark-conf.spark.sql.objectHashAggregate.sortBased.fallbackThreshold=1
kylin.query.async-query.spark-conf.spark.sql.hive.caseSensitiveInferenceMode=NEVER_INFER
kylin.query.async-query.spark-conf.spark.sql.sources.bucketing.enabled=false
kylin.query.async-query.spark-conf.spark.yarn.stagingDir=${kylin.env.hdfs-working-dir}
kylin.query.async-query.spark-conf.spark.eventLog.enabled=true
kylin.query.async-query.spark-conf.spark.history.fs.logDirectory=${kylin.env.hdfs-working-dir}/sparder-history
kylin.query.async-query.spark-conf.spark.eventLog.dir=${kylin.env.hdfs-working-dir}/sparder-history
kylin.query.async-query.spark-conf.spark.eventLog.rolling.enabled=true
kylin.query.async-query.spark-conf.spark.eventLog.rolling.maxFileSize=100m
kylin.query.async-query.spark-conf.spark.sql.cartesianPartitionNumThreshold=-1
kylin.query.async-query.spark-conf.parquet.filter.columnindex.enabled=false
kylin.query.async-query.spark-conf.spark.master=yarn
kylin.query.async-query.spark-conf.spark.submit.deployMode=client

Limitations

• Asynchronous query does not support query cache.
• When the select column of SQL contains ,;{}()=, these characters will be converted to _ in the result file.