Spark jobs size increasing double for same dataframe for same proration about spark-flight-connector HOT 4 CLOSED

Per your case, can you run the following code?

val df=spark.read.option("host", "[SPARK_IP]").option("port", 32010).option("tls.enabled", false).option("user", "user").option("password", "password").flight(""""dremio_space"."file"""").filter("COLUMN == 22")

Then
df.count

Go to Dremio web console, check the query logs and see what query was submitted to Dremio?
Run df.count again, and see what query submitted to Dremio?

Run the submitted query in Dremio web console multiple times?

Thanks
Wayne

from spark-flight-connector.

Hi Wayne

I have used query without "partition.byColumn" it works fine and fast.
But "partition.byColumn" help to load big data very fast, but it will hit dremio more than a 100 times.
If this is the case, how to manipulate the data in data frame.

When we should use "partition.byColumn" and when we should avoid "partition.byColumn".
Or let me know if there is reference for such scenario.
Please let me know.

Thanks
Nagaraj M M

from spark-flight-connector.

The partition.size and partition.byColumn control how many partitions will be in the dataframe. You get a single dataframe so you manipulate it as normal.

When a dataset being loaded from Dremio is a large one such as millions+ rows, then you need to provide values for this two properties (with current version of Dremio), otherwise you may get "Out of Memory" issue. In your current case (count = 15), you don't need the partitioning.

However df.count should only submit one query to Dremio so there should be only one Dremio job instead of 320 jobs due to the push-down of count.

Hope this helps.

Thanks
Wayne

from spark-flight-connector.

This behavior is by design - "Multiple Jobs submitted to Dremio when custom partitioning is enabled", so closing this ticket.

@nagarajmmu if you still see unusual behavior from the connector, please submit another ticket.

Thanks
Wayne

from spark-flight-connector.