RussellSpitzer commented on a change in pull request #3862:
URL: https://github.com/apache/iceberg/pull/3862#discussion_r806300891
##########
File path:
spark/v3.2/spark/src/main/java/org/apache/iceberg/spark/source/SparkTable.java
##########
@@ -272,6 +283,65 @@ public void deleteWhere(Filter[] filters) {
}
}
+ @Override
+ public StructType partitionSchema() {
+ return (StructType)
SparkSchemaUtil.convert(Partitioning.partitionType(table()));
+ }
+
+ @Override
+ public void createPartition(InternalRow ident, Map<String, String>
properties) throws UnsupportedOperationException {
+ throw new UnsupportedOperationException("Cannot explicitly create
partitions in Iceberg tables");
+ }
+
+ @Override
+ public boolean dropPartition(InternalRow ident) {
+ throw new UnsupportedOperationException("Cannot explicitly drop partitions
in Iceberg tables");
+ }
+
+ @Override
+ public void replacePartitionMetadata(InternalRow ident, Map<String, String>
properties)
+ throws UnsupportedOperationException {
+ throw new UnsupportedOperationException("Iceberg partitions do not support
metadata");
+ }
+
+ @Override
+ public Map<String, String> loadPartitionMetadata(InternalRow ident) throws
UnsupportedOperationException {
+ throw new UnsupportedOperationException("Iceberg partitions do not support
metadata");
+ }
+
+ @Override
+ public InternalRow[] listPartitionIdentifiers(String[] names, InternalRow
ident) {
+ // support show partitions
+ List<InternalRow> rows = Lists.newArrayList();
+ Dataset<Row> df = SparkTableUtil.loadMetadataTable(sparkSession(),
icebergTable, MetadataTableType.PARTITIONS);
+ if (names.length > 0) {
+ StructType schema = partitionSchema();
+ df.collectAsList().forEach(row -> {
+ GenericRowWithSchema genericRow = (GenericRowWithSchema) row.apply(0);
+ boolean exits = true;
+ int index = 0;
+ while (index < names.length) {
+ DataType dataType = schema.apply(names[index]).dataType();
Review comment:
It looks like we are trying to align the metadata table schema with the
current table schema. I think we should still just be displaying metadata table
partition values as is but if we choose to go this route I think we have an
issue here still.
Consider a table
```
Add Partition Column Identity (a)
Remove Partition Column identity (a)
Drop Column a
Add Column a
Add partition Column Identity (a)
```
This should result in a row which has multiple "a"'s in the partition spec
(at least I believe this is the current behavior). We should make sure we are
correctly projecting columns in those cases. I think it is also ok for this
just to be a light wrapper around the Metadata Table for Partitions and just
list the partitions in the extended schema it provides.
##########
File path:
spark/v3.2/spark/src/main/java/org/apache/iceberg/spark/source/SparkTable.java
##########
@@ -272,6 +283,65 @@ public void deleteWhere(Filter[] filters) {
}
}
+ @Override
+ public StructType partitionSchema() {
+ return (StructType)
SparkSchemaUtil.convert(Partitioning.partitionType(table()));
+ }
+
+ @Override
+ public void createPartition(InternalRow ident, Map<String, String>
properties) throws UnsupportedOperationException {
+ throw new UnsupportedOperationException("Cannot explicitly create
partitions in Iceberg tables");
+ }
+
+ @Override
+ public boolean dropPartition(InternalRow ident) {
+ throw new UnsupportedOperationException("Cannot explicitly drop partitions
in Iceberg tables");
+ }
+
+ @Override
+ public void replacePartitionMetadata(InternalRow ident, Map<String, String>
properties)
+ throws UnsupportedOperationException {
+ throw new UnsupportedOperationException("Iceberg partitions do not support
metadata");
+ }
+
+ @Override
+ public Map<String, String> loadPartitionMetadata(InternalRow ident) throws
UnsupportedOperationException {
+ throw new UnsupportedOperationException("Iceberg partitions do not support
metadata");
+ }
+
+ @Override
+ public InternalRow[] listPartitionIdentifiers(String[] names, InternalRow
ident) {
+ // support show partitions
+ List<InternalRow> rows = Lists.newArrayList();
+ Dataset<Row> df = SparkTableUtil.loadMetadataTable(sparkSession(),
icebergTable, MetadataTableType.PARTITIONS);
+ if (names.length > 0) {
+ StructType schema = partitionSchema();
+ df.collectAsList().forEach(row -> {
+ GenericRowWithSchema genericRow = (GenericRowWithSchema) row.apply(0);
+ boolean exits = true;
+ int index = 0;
+ while (index < names.length) {
+ DataType dataType = schema.apply(names[index]).dataType();
Review comment:
It looks like we are trying to align the metadata table schema with the
current table schema. I think we should still just be displaying metadata table
partition values as is but if we choose to go this route I think we have an
issue here still.
Consider a table
```
Add Partition Column Identity (a)
Remove Partition Column identity (a)
Drop Column a
Add Column a
Add partition Column Identity (a)
```
This should result in a row which has multiple "a"'s in the partition spec
(at least I believe this is the current behavior). We should make sure we are
correctly projecting columns in those cases. I think it is also ok for this
just to be a light wrapper around the Metadata Table for Partitions and just
list the partitions in the extended schema it provides.
I guess this may be a little odd for unpartitioned tables since they may
still show that partitions do exist but this is probably more accurate ...
@jackye1995 + @szehon-ho Any thoughts?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
