rui-mo commented on code in PR #12967:
URL: https://github.com/apache/gluten/pull/12967#discussion_r4026631740
##########
backends-velox/src/main/scala/org/apache/gluten/config/VeloxConfig.scala:
##########
@@ -981,9 +981,12 @@ object VeloxConfig extends ConfigRegistry {
val ENABLE_TIMESTAMP_NTZ_VALIDATION =
buildConf("spark.gluten.sql.columnar.backend.velox.enableTimestampNtzValidation")
.doc(
- "Enable validation fallback for TimestampNTZ type. When true, any plan
" +
- "containing TimestampNTZ will fall back to Spark execution. When
false, " +
- "allows native execution for TimestampNTZ scan.")
+ "Enable validation fallback for TimestampNTZ type. When true, plans
with " +
+ "TimestampNTZ in their input or output schemas fall back to Spark
execution. " +
+ "When false, supported TimestampNTZ scans, aggregates, shuffles and
projections " +
+ "are eligible for native execution. Supported projections include
direct columns " +
+ "and struct fields, casts, hour/minute/second extraction, timestamp
addition " +
+ "and null checks. Other operator, expression and native validation
rules still apply.")
Review Comment:
> Supported projections include direct columns and struct fields, casts,
hour/minute/second extraction, timestamp addition and null checks.
Support for `timestamp_ntz` is evolving quickly and may change frequently,
so could we avoid documenting those implementation details for now? It would be
better to keep the configuration documentation at a high level.
##########
docs/velox-configuration.md:
##########
@@ -29,7 +29,7 @@ nav_order: 16
| spark.gluten.sql.columnar.backend.velox.cudf.shuffleMaxPrefetchBytes
| 🔄 Dynamic | 1028MB | Maximum bytes to prefetch in CPU
memory during GPU shuffle read while waiting for GPU available.
|
| spark.gluten.sql.columnar.backend.velox.directorySizeGuess
| âš“ Static | 32KB | Deprecated, rename to
spark.gluten.sql.columnar.backend.velox.footerEstimatedSize
|
| spark.gluten.sql.columnar.backend.velox.driverSideBroadcastHashTableBuild
| 🔄 Dynamic | false | Enable driver-side broadcast hash
table build. When enabled, the hash table is built and serialized on the
driver, then broadcast to executors. When disabled, each executor builds its
own hash table from the broadcast data.
|
-| spark.gluten.sql.columnar.backend.velox.enableTimestampNtzValidation
| 🔄 Dynamic | false | Enable validation fallback for
TimestampNTZ type. When true, any plan containing TimestampNTZ will fall back
to Spark execution. When false, allows native execution for TimestampNTZ scan.
|
+| spark.gluten.sql.columnar.backend.velox.enableTimestampNtzValidation
| 🔄 Dynamic | false | Enable validation fallback for
TimestampNTZ type. When true, plans with TimestampNTZ in their input or output
schemas fall back to Spark execution. When false, supported TimestampNTZ scans,
aggregates, shuffles and projections are eligible for native execution.
Supported projections include direct columns and struct fields, casts,
hour/minute/second extraction, timestamp addition and null checks. Other
operator, expression and native validation rules still apply.
|
Review Comment:
ditto
##########
gluten-substrait/src/main/scala/org/apache/gluten/extension/columnar/validator/Validators.scala:
##########
@@ -273,17 +279,21 @@ object Validators {
case _ => false
}
val isSupportedNtz = plan match {
+ case _: HashAggregateExec | _: ObjectHashAggregateExec | _:
SortAggregateExec => true
+ case _: ShuffleExchangeExec => true
case p: ProjectExec =>
p.projectList.forall {
expr =>
(!containsNTZ(expr.dataType) &&
!expr.references.exists(a => containsNTZ(a.dataType))) ||
+ isDirectNtzProjection(expr) ||
expr.exists {
case Hour(child, _) => containsNTZ(child.dataType)
case Minute(child, _) => containsNTZ(child.dataType)
case Second(child, _) => containsNTZ(child.dataType)
case TimestampAdd(_, _, child, _) =>
containsNTZ(child.dataType)
case c: Cast if isNTZ(c.dataType) || isNTZ(c.child.dataType)
=> true
+ case IsNull(child) => containsNTZ(child.dataType)
Review Comment:
`isnull` is also unverified for TIMESTAMP_UTC type in Velox.
--
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]
