You know I can’t resist… 😉

Sure thing:

graph.pregel.withVertexColumn(
colName="path",
initialExpr=F.when(F.col("parent_id").isNull(), F.col("curr_id")).otherwise(F.lit(None)),
updateAfterAggMsgsExpr=F.when(F.col("path").isNull(), F.concat(F.col("curr_id"), F.lit(","), Pregel.msg())).otherwise(
F.col("path")
),
)
.sendMsgToSrc(Pregel.dst("path"))
.aggMsgs(F.any_value(Pregel.msg()))
.setMaxIter(25)
.run()

I have a bunch of disjoint trees with each node having only one parent. I'm constructing an upwards path of comma-separated vertex IDs from each vertex to its highest level parent.

I'm encountering the following error when trying to reproduce the issue:

Exception in thread "main" org.apache.spark.SparkRuntimeException: [UNSUPPORTED_FEATURE.LITERAL_TYPE] The feature is not supported: Literal for '()' of class scala.runtime.BoxedUnit._

It appears this is caused by using F.lit(None), which creates a NullType literal without an explicit type, leading to an unsupported literal type in Spark.

Overall:

• Check that there is no NULL / duplicates in GraphFrame.ID column;
• Try to disable AQE;

I confirmed that there are no null or duplicate values in the ID column. I'll give disabling AQE a shot and if this persists, look into the feasibility of migrating to 0.9.2.