gengliangwang commented on code in PR #56124:
URL: https://github.com/apache/spark/pull/56124#discussion_r3307403195
##########
sql/connect/server/src/main/scala/org/apache/spark/sql/connect/pipelines/PipelinesHandler.scala:
##########
@@ -388,13 +400,107 @@ private[connect] object PipelinesHandler extends Logging
{
objectName = Option(flowIdentifier.unquotedString),
language = Some(Python()))))
case proto.PipelineCommand.DefineFlow.DetailsCase.AUTO_CDC_FLOW_DETAILS
=>
- throw new UnsupportedOperationException("AutoCdcFlowDetails is not yet
implemented.")
+ val autoCdcDetails = flow.getAutoCdcFlowDetails
+ graphElementRegistry.registerFlow(
+ buildAutoCdcFlow(
+ autoCdcDetails = autoCdcDetails,
+ flow = flow,
+ flowIdentifier = flowIdentifier,
+ destinationIdentifier = destinationIdentifier,
+ defaultCatalog = defaultCatalog,
+ defaultDatabase = defaultDatabase,
+ transformExpressionFunc = transformExpressionFunc))
case other =>
throw new UnsupportedOperationException(s"Unsupported DefineFlow
details case: $other")
}
flowIdentifier
}
+ /**
+ * Build an [[AutoCdcFlow]] from the proto-supplied AutoCDC flow details.
+ *
+ * The flow's source expression is encoded by the Python client as a
streaming-table name; we
+ * model that on the server side as a streaming [[UnresolvedRelation]] so
that pipelines flow
+ * analysis (which already handles `STREAM(t)` references) can resolve it
against the rest of
+ * the dataflow graph.
+ */
+ private def buildAutoCdcFlow(
+ autoCdcDetails: proto.PipelineCommand.DefineFlow.AutoCdcFlowDetails,
+ flow: proto.PipelineCommand.DefineFlow,
+ flowIdentifier: TableIdentifier,
+ destinationIdentifier: TableIdentifier,
+ defaultCatalog: String,
+ defaultDatabase: String,
+ transformExpressionFunc: proto.Expression => Expression): AutoCdcFlow = {
+ val sourcePlan: LogicalPlan = UnresolvedRelation(
+ multipartIdentifier =
scala.collection.immutable.Seq(autoCdcDetails.getSource),
+ isStreaming = true
+ )
+
+ val toColumn: proto.Expression => Column = expr =>
Column(transformExpressionFunc(expr))
+
+ val asUnqualifiedColumnName: proto.Expression => UnqualifiedColumnName =
+ expr => UnqualifiedColumnName(transformExpressionFunc(expr).sql)
+
+ val keys =
autoCdcDetails.getKeysList.asScala.toSeq.map(asUnqualifiedColumnName)
+
+ val columnSelection: Option[ColumnSelection] = {
+ val included = autoCdcDetails.getColumnListList.asScala.toSeq
+ val excluded = autoCdcDetails.getExceptColumnListList.asScala.toSeq
+ if (included.nonEmpty && excluded.nonEmpty) {
+ // The Python API enforces the "at most one" contract; we don't expect
both to be set
Review Comment:
The "The Python API enforces the 'at most one' contract" claim isn't quite
right: `python/pyspark/pipelines/api.py:create_auto_cdc_flow` documents the
rule in its docstring but doesn't validate it — both `column_list` and
`except_column_list` flow unchanged through
`SparkConnectGraphElementRegistry.register_auto_cdc_flow` into the proto, and
the downstream `ColumnSelection` model is one-or-the-other by construction so
it never sees both. This server-side throw is therefore the only place in the
full stack where the rule is actually enforced, not a defensive backstop.
Worth rewording — and, if you take @szehon-ho's adjacent suggestion to use
`AnalysisException` with a structured error class, dropping the misleading
framing here as well.
--
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]
