Spark SQL saveAsTable returns empty result

问题:

I am using the following code for creating / inserting data into a Hive table in Spark SQL:

val sc = SparkSession
  .builder()
  .appName("App")
  .master("local[2]")
  .config("spark.sql.warehouse.dir", "file:///tmp/spark-warehouse")
  .enableHiveSupport()
  .getOrCreate()

// actual code

result.createOrReplaceTempView("result")
result.write.format("parquet").partitionBy("year", "month").mode(SaveMode.Append).saveAsTable("tablename")

Which runs without errors. A result.show(10) confirms this. The input files are csv on the local FS.

It creates parquet files under ./spark-warehouse/tablename/ and also creates the table in hive, using a correct create table statement.

git:(master) ✗ tree
.
└── tablename
    ├── _SUCCESS
    └── year=2017
        └── month=01
            ├── part-r-00013-abaea338-8ed3-4961-8598-cb2623a78ce1.snappy.parquet
            ├── part-r-00013-f42ce8ac-a42c-46c5-b188-598a23699ce8.snappy.parquet
            ├── part-r-00018-abaea338-8ed3-4961-8598-cb2623a78ce1.snappy.parquet
            └── part-r-00018-f42ce8ac-a42c-46c5-b188-598a23699ce8.snappy.parquet

hive:

hive> show create table tablename;
OK
CREATE TABLE `tablename`(
  `col` array<string> COMMENT 'from deserializer')
ROW FORMAT SERDE
  'org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe'
WITH SERDEPROPERTIES (
  'path'='file:/Users/IdeaProjects/project/spark-warehouse/tablename')
STORED AS INPUTFORMAT
  'org.apache.hadoop.mapred.SequenceFileInputFormat'
OUTPUTFORMAT
  'org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat'
LOCATION
  'file:/tmp/spark-warehouse/tablename'
TBLPROPERTIES (
  'EXTERNAL'='FALSE',
  'spark.sql.sources.provider'='parquet',
  'spark.sql.sources.schema.numPartCols'='2',
  'spark.sql.sources.schema.numParts'='1',
  'spark.sql.sources.schema.part.0'='{ 
  // fields
  }',
  'spark.sql.sources.schema.partCol.0'='year',
  'spark.sql.sources.schema.partCol.1'='month',
  'transient_lastDdlTime'='1488157476')

However, the table is empty:

hive> select count(*) from tablename;
...
OK
0
Time taken: 1.89 seconds, Fetched: 1 row(s)

Software used: Spark 2.1.0 with spark-sql and spark-hive_2.10, Hive 2.10 and a mysql metastore, Hadoop 2.70, macOS 10.12.3

回答1:

Spark SQL partitioning is not compatible with Hive. This issue is documented by SPARK-14927.

As a recommended workaround you can create partitioned table with Hive, and only insert from Spark.

回答2:

You should just need to run MSCK REPAIR TABLE when adding new partitions. See Hive docs: https://cwiki.apache.org/confluence/display/Hive/LanguageManual+DDL#LanguageManualDDL-RecoverPartitions(MSCKREPAIRTABLE)