Implement GetIndexedField for map-typed columns #7825
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
logical-expr
swgillespie
force-pushed
the
swgillespie/maps-3
branch
from
22aa582
to
545548d
Compare
alamb
approved these changes
Oct 15, 2023
There was a problem hiding this comment.
Thank you @swgillespie - this looks very nice.
I also played around with the file locally with datafusion-cli and it worked great.
Thank you 🙏
❯ describe 'parquet_map.parquet';
+-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------+
| column_name | data_type | is_nullable |
+-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------+
| ints | Map(Field { name: "entries", data_type: Struct([Field { name: "key", data_type: Utf8, nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }, Field { name: "value", data_type: Int64, nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }]), nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }, false) | NO |
| strings | Map(Field { name: "entries", data_type: Struct([Field { name: "key", data_type: Utf8, nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }, Field { name: "value", data_type: Utf8, nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }]), nullable: false, dict_id: 0, dict_is_ordered: false, metadata: {} }, false) | NO |
| timestamp | Utf8 | NO |
+-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------+
3 rows in set. Query took 0.018 seconds.| (DataType::Map(fields, _), _) => { | ||
| match fields.data_type() { | ||
| DataType::Struct(fields) if fields.len() == 2 => { | ||
| // Arrow's MapArray is essentially a ListArray of structs with two columns. They are |
| (DataType::Map(_, _), ScalarValue::Utf8(Some(k))) => { | ||
| let map_array = as_map_array(array.as_ref())?; | ||
| let key_scalar = Scalar::new(StringArray::from(vec![k.clone()])); | ||
| let keys = arrow_ord::cmp::eq(&key_scalar, map_array.keys())?; |
There was a problem hiding this comment.
I think arrow::compute::eq is probably the more standard way to compare two arrays (though they call the same underlying kernels). That would let you avoid having to add the (newly explict) arrow-ord dependency.
|
@swgillespie I think you could avoid adding the arrow-ord dependency, but I don't think that is critical (as arrow-ord is already a transitive dependency via arrow anyways). Let me know if you are willing to make this change |
|
@alamb no problem - done! |
alamb
reviewed
Oct 16, 2023
| @@ -183,6 +186,14 @@ impl PhysicalExpr for GetIndexedFieldExpr { | |||
| let array = self.arg.evaluate(batch)?.into_array(batch.num_rows()); | |||
| match &self.field { | |||
| GetFieldAccessExpr::NamedStructField{name} => match (array.data_type(), name) { | |||
| (DataType::Map(_, _), ScalarValue::Utf8(Some(k))) => { | |||
alamb
approved these changes
Oct 16, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Which issue does this PR close?
Closes #7824.
Rationale for this change
It's impossible to write a logical plan to query a column from a Parquet data source whose type is
Map. TheMaptype is not explicitly supported byGetIndexedField. Maps are a useful column, are already supported by both Arrow and Parquet, and it makes sense to support it here.What changes are included in this PR?
This commit extends the
NamedStructFieldFieldAccess type to understand theMapdata type. I chose this because the DataFusion SQL frontend parses the expressionx['y']into aNamedStructField, which is a reasonable thing to do if we require that the argument toxbe a constant scalar (which it is, in this implementation).The Arrow Map array is essentially a list of structs, where each struct is a two-field struct. The first field of the struct is the key, and the second field of the struct is the value. Arrow traditionally names these
keyandvalue, though this implementation does not assume what they are named and instead assumes that the second column is thevaluecolumn and the first is thekeycolumn, which is the same assumption made by the Arrow implementation we use.To execute a mapped index access, we first scan the key column to identify entries that match the key that we are indexing, and again to gather the values corresponding to the keys that were selected.
Are these changes tested?
This PR adds a new test,
map.slt, which includes a Parquet file with twoMapcolumns (one mapping strings to strings, the other mapping strings to ints) and writes some queries that use them.Are there any user-facing changes?
This change allows for the
GetIndexedFieldtype to now be usable with columns of typeMap, which was not possible before.