Skip to content

[SPARK-29448][SQL] Support the INTERVAL type by Parquet datasource #26102

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
MaxGekk wants to merge 19 commits into from
Closed

[SPARK-29448][SQL] Support the INTERVAL type by Parquet datasource #26102

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
1c64fb1
Support the interval type in schemas
MaxGekk Oct 10, 2019
0769585
Support the interval type in ColumnarBatchRow
MaxGekk Oct 10, 2019
cfcecf1
Fix imports
MaxGekk Oct 10, 2019
c7889ed
Support writing and reading intervals
MaxGekk Oct 12, 2019
4563a4e
Remove unused variable
MaxGekk Oct 12, 2019
83b4bf2
Fix coding style
MaxGekk Oct 12, 2019
81ef7be
Exclude parquet from checking
MaxGekk Oct 12, 2019
de0872e
Exclude parquet from checks of unsupported interval
MaxGekk Oct 12, 2019
281f62b
Combine AtomicType and CalendarIntervalType in supportDataType()
MaxGekk Oct 14, 2019
232b1fb
Replace timestampBuffer and intervalBuffer by reusableBuffer
MaxGekk Oct 14, 2019
3f6ad61
Move unsupported type check to checkUnsupportedTypes()
MaxGekk Oct 14, 2019
c260622
Add more tests to ParquetQuerySuite
MaxGekk Oct 14, 2019
8fd037b
Merge remote-tracking branch 'origin/master' into parquet-interval
MaxGekk Oct 25, 2019
032a2ea
CalendarInterval.fromString -> IntervalUtils.fromString
MaxGekk Oct 25, 2019
6834f1e
Merge remote-tracking branch 'remotes/origin/master' into parquet-int…
MaxGekk Nov 1, 2019
712378c
Rebase on new field - days
MaxGekk Nov 1, 2019
cf8023a
Add one more test
MaxGekk Nov 1, 2019
8fa4a0d
Merge remote-tracking branch 'remotes/origin/master' into parquet-int…
MaxGekk Dec 10, 2019
9041110
Rebase on the master
MaxGekk Dec 10, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Rebase on new field - days
  • Loading branch information
@MaxGekk
MaxGekk committed Nov 1, 2019
commit 712378c6001c352d76c28c6cdf6176b694ea5355
3 changes: 2 additions & 1 deletion sql/core/src/main/java/org/apache/spark/sql/execution/vectorized/ColumnVectorUtils.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,7 +89,8 @@ public static void populate(WritableColumnVector col, InternalRow row, int field
} else if (t instanceof CalendarIntervalType) {
CalendarInterval c = (CalendarInterval)row.get(fieldIdx, t);
col.getChild(0).putInts(0, capacity, c.months);
col.getChild(1).putLongs(0, capacity, c.microseconds);
col.getChild(1).putInts(0, capacity, c.days);
col.getChild(2).putLongs(0, capacity, c.microseconds);
} else if (t instanceof DateType) {
col.putInts(0, capacity, row.getInt(fieldIdx));
} else if (t instanceof TimestampType) {
Expand Down
7 changes: 2 additions & 5 deletions ...c/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRowConverter.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -333,13 +333,10 @@ private[parquet] class ParquetRowConverter(
s"but got a ${value.length()}-byte array.")

val buf = value.toByteBuffer.order(ByteOrder.LITTLE_ENDIAN)
val milliseconds = buf.getInt
var microseconds = milliseconds * DateTimeUtils.MICROS_PER_MILLIS
val microseconds = buf.getInt * DateTimeUtils.MICROS_PER_MILLIS
val days = buf.getInt
val daysInUs = Math.multiplyExact(days, DateTimeUtils.MICROS_PER_DAY)
microseconds = Math.addExact(microseconds, daysInUs)
val months = buf.getInt
updater.set(new CalendarInterval(months, microseconds))
updater.set(new CalendarInterval(months, days, microseconds))
}
}

Expand Down
7 changes: 2 additions & 5 deletions ...c/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetWriteSupport.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,13 +211,10 @@ class ParquetWriteSupport extends WriteSupport[InternalRow] with Logging {
case CalendarIntervalType =>
(row: SpecializedGetters, ordinal: Int) =>
val interval = row.getInterval(ordinal)
val microseconds = interval.microseconds % DateTimeUtils.MICROS_PER_DAY
val milliseconds: Int = (microseconds / DateTimeUtils.MICROS_PER_MILLIS).toInt
val days: Int = Math.toIntExact(interval.microseconds / DateTimeUtils.MICROS_PER_DAY)
val buf = ByteBuffer.wrap(reusableBuffer)
buf.order(ByteOrder.LITTLE_ENDIAN)
.putInt(milliseconds)
.putInt(days)
.putInt((interval.milliseconds()).toInt)
Copy link
Member

@HyukjinKwon HyukjinKwon Nov 15, 2019 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@MaxGekk the doc(https://github.com/apache/parquet-format/blob/master/LogicalTypes.md#interval) says:

three little-endian unsigned integers

what happens if we set negative values for some parts in the interval and negative values are written here?

Copy link
Member Author

@MaxGekk MaxGekk Nov 15, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Spark will read them back as negative values: https://github.com/apache/spark/pull/26102/files#diff-35a70bb270f17ea3a1d964c4bec0e0a2R912 . I don't know about other systems.

Copy link
Member

@HyukjinKwon HyukjinKwon Nov 18, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, shouldn't we maybe add an assert to reject negative parts for now? Seems it doesn't comply parquet format. I'm just worried about the case we have to explain this multiple times to users later like https://issues.apache.org/jira/browse/SPARK-20937 and https://issues.apache.org/jira/browse/SPARK-20297

Do you think it is common to use negative parts? If not, let's just disallow.

Copy link
Contributor

@cloud-fan cloud-fan Nov 18, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we should use strict toInt so that we can fail earlier if out of range.

Copy link
Member Author

@MaxGekk MaxGekk Dec 10, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

fixed

.putInt(interval.days)
.putInt(interval.months)
recordConsumer.addBinary(Binary.fromReusedByteArray(reusableBuffer))

Expand Down