I'm trying to use Azure Synapse to get the schema of some data that is stored in Azure storage accounts. A complication is that the data is sharded across 4 storage accounts, so I need to read each account into a dataframe separately and union the dataframes into a single one for querying. I tried using
val data1 = spark.read.parquet("abfss://PATH1")
val data2 = spark.read.parquet("abfss://PATH2")
val data3 = spark.read.parquet("abfss://PATH3")
val data4 = spark.read.parquet("abfss://PATH4")
val data = data1
.unionByName(data2, allowMissingColumns=true)
.unionByName(data3, allowMissingColumns=true)
.unionByName(data4, allowMissingColumns=true)
data.printSchema()
The problem is that I have a column that is a struct, let's call it ABC, which among other things has a column called #XYZ# (# to denote the value is hashed).
This doesn't seem to cause a problem for Spark on it's own, as this works:
data1.select("ABC.#XYZ#").where(col("#XYZ#").isNotNull).show(5, truncate = false).
But if I try the unionByName steps, I get this error:
org.apache.spark.sql.catalyst.parser.ParseException:
extraneous input '#' expecting {'ADD', 'AFTER', 'ALL', 'ALTER', 'ANALYZE', 'AND', 'ANTI', 'ANY', 'ARCHIVE', 'ARRAY', 'AS', 'ASC', 'AT', 'AUTHORIZATION', 'BETWEEN', 'BOTH', 'BUCKET', 'BUCKETS', 'BY', 'CACHE', 'CASCADE', 'CASE', 'CAST', 'CHANGE', 'CHECK', 'CLEAR', 'CLUSTER', 'CLUSTERED', 'CODEGEN', 'COLLATE', 'COLLECTION', 'COLUMN', 'COLUMNS', 'COMMENT', 'COMMIT', 'COMPACT', 'COMPACTIONS', 'COMPUTE', 'CONCATENATE', 'CONSTRAINT', 'COST', 'CREATE', 'CROSS', 'CUBE', 'CURRENT', 'CURRENT_DATE', 'CURRENT_TIME', 'CURRENT_TIMESTAMP', 'CURRENT_USER', 'DATA', 'DATABASE', DATABASES, 'DBPROPERTIES', 'DEFINED', 'DELETE', 'DELIMITED', 'DESC', 'DESCRIBE', 'DFS', 'DIRECTORIES', 'DIRECTORY', 'DISTINCT', 'DISTRIBUTE', 'DIV', 'DROP', 'ELSE', 'END', 'ESCAPE', 'ESCAPED', 'EXCEPT', 'EXCHANGE', 'EXISTS', 'EXPLAIN', 'EXPORT', 'EXTENDED', 'EXTERNAL', 'EXTRACT', 'FALSE', 'FETCH', 'FIELDS', 'FILTER', 'FILEFORMAT', 'FIRST', 'FOLLOWING', 'FOR', 'FOREIGN', 'FORMAT', 'FORMATTED', 'FROM', 'FULL', 'FUNCTION', 'FUNCTIONS', 'GLOBAL', 'GRANT', 'GROUP', 'GROUPING', 'HAVING', 'IF', 'IGNORE', 'IMPORT', 'IN', 'INDEX', 'INDEXES', 'INNER', 'INPATH', 'INPUTFORMAT', 'INSERT', 'INTERSECT', 'INTERVAL', 'INTO', 'IS', 'ITEMS', 'JOIN', 'KEYS', 'LAST', 'LATERAL', 'LAZY', 'LEADING', 'LEFT', 'LIKE', 'LIMIT', 'LINES', 'LIST', 'LOAD', 'LOCAL', 'LOCATION', 'LOCK', 'LOCKS', 'LOGICAL', 'MACRO', 'MAP', 'MATCHED', 'MERGE', 'MSCK', 'NAMESPACE', 'NAMESPACES', 'NATURAL', 'NO', NOT, 'NULL', 'NULLS', 'OF', 'ON', 'ONLY', 'OPTION', 'OPTIONS', 'OR', 'ORDER', 'OUT', 'OUTER', 'OUTPUTFORMAT', 'OVER', 'OVERLAPS', 'OVERLAY', 'OVERWRITE', 'PARTITION', 'PARTITIONED', 'PARTITIONS', 'PERCENT', 'PIVOT', 'PLACING', 'POSITION', 'PRECEDING', 'PRIMARY', 'PRINCIPALS', 'PROPERTIES', 'PURGE', 'QUERY', 'RANGE', 'RECORDREADER', 'RECORDWRITER', 'RECOVER', 'REDUCE', 'REFERENCES', 'REFRESH', 'RENAME', 'REPAIR', 'REPLACE', 'RESET', 'RESTRICT', 'REVOKE', 'RIGHT', RLIKE, 'ROLE', 'ROLES', 'ROLLBACK', 'ROLLUP', 'ROW', 'ROWS', 'SCHEMA', 'SELECT', 'SEMI', 'SEPARATED', 'SERDE', 'SERDEPROPERTIES', 'SESSION_USER', 'SET', 'MINUS', 'SETS', 'SHOW', 'SKEWED', 'SOME', 'SORT', 'SORTED', 'START', 'STATISTICS', 'STORED', 'STRATIFY', 'STRUCT', 'SUBSTR', 'SUBSTRING', 'TABLE', 'TABLES', 'TABLESAMPLE', 'TBLPROPERTIES', TEMPORARY, 'TERMINATED', 'THEN', 'TIME', 'TO', 'TOUCH', 'TRAILING', 'TRANSACTION', 'TRANSACTIONS', 'TRANSFORM', 'TRIM', 'TRUE', 'TRUNCATE', 'TYPE', 'UNARCHIVE', 'UNBOUNDED', 'UNCACHE', 'UNION', 'UNIQUE', 'UNKNOWN', 'UNLOCK', 'UNSET', 'UPDATE', 'USE', 'USER', 'USING', 'VALUES', 'VIEW', 'VIEWS', 'WHEN', 'WHERE', 'WINDOW', 'WITH', 'ZONE', IDENTIFIER, BACKQUOTED_IDENTIFIER}(line 1, pos 0)
== SQL ==
#XYZ#
^^^
Does this indicate a bug with Spark's union code? Or is there something I'm missing?
As alluded to by @Gabio's comment, the issue seems to be with this version of Spark, 3.1.2. I switched to using the latest version of Spark on Synapse, which is 3.3.1, and the issue went away.