I think it makes sense to allow specifying more than one, if desired. This is equivalent to just stacking multiple Unnests. (Possibly one could even have a special syntax like "*" for all array fields.)
On Thu, Jan 14, 2021 at 10:05 AM Reuven Lax <re...@google.com> wrote: > Should Unnest be allowed to specify multiple array fields, or just one? > > On Wed, Jan 13, 2021 at 11:59 PM Manninger, Matyas < > matyas.mannin...@veolia.com> wrote: > >> I would also not unnest arrays nested in arrays just the top-level array >> of the specified fields. >> >> On Wed, 13 Jan 2021 at 20:58, Reuven Lax <re...@google.com> wrote: >> >>> Nested fields are not part of standard SQL AFAIK. Beam goes further and >>> supports array of array, etc. >>> >>> On Wed, Jan 13, 2021 at 11:42 AM Kenneth Knowles <k...@apache.org> >>> wrote: >>> >>>> Just the fields specified, IMO. When in doubt, copy SQL. (and I mean >>>> SQL generally, not just Beam SQL) >>>> >>>> Kenn >>>> >>>> On Wed, Jan 13, 2021 at 11:17 AM Reuven Lax <re...@google.com> wrote: >>>> >>>>> Definitely could be a top-level transform. Should it automatically >>>>> unnest all arrays, or just the fields specified? >>>>> >>>>> We do have to define the semantics for nested arrays as well. >>>>> >>>>> On Wed, Jan 13, 2021 at 10:57 AM Robert Bradshaw <rober...@google.com> >>>>> wrote: >>>>> >>>>>> Ah, thanks for the clarification. UNNEST does sound like what you >>>>>> want here, and would likely make sense as a top-level relational >>>>>> transform >>>>>> as well as being supported by SQL. >>>>>> >>>>>> On Wed, Jan 13, 2021 at 10:53 AM Tao Li <t...@zillow.com> wrote: >>>>>> >>>>>>> @Kyle Weaver <kcwea...@google.com> sure thing! So the input/output >>>>>>> definition for the Flatten.Iterables >>>>>>> <https://beam.apache.org/releases/javadoc/2.25.0/org/apache/beam/sdk/transforms/Flatten.Iterables.html> >>>>>>> is: >>>>>>> >>>>>>> >>>>>>> >>>>>>> Input: PCollection<Iterable<T> >>>>>>> >>>>>>> Output: PCollection<T> >>>>>>> >>>>>>> >>>>>>> >>>>>>> The input/output for a explode transform would look like this: >>>>>>> >>>>>>> Input: PCollection<Row> The row schema has a field which is an >>>>>>> array of T >>>>>>> >>>>>>> Output: PCollection<Row> The array type field from input schema is >>>>>>> replaced with a new field of type T. The elements from the array type >>>>>>> field >>>>>>> are flattened into multiple rows in the new table (other fields of input >>>>>>> table are just duplicated. >>>>>>> >>>>>>> >>>>>>> >>>>>>> Hope this clarification helps! >>>>>>> >>>>>>> >>>>>>> >>>>>>> *From: *Kyle Weaver <kcwea...@google.com> >>>>>>> *Reply-To: *"user@beam.apache.org" <user@beam.apache.org> >>>>>>> *Date: *Tuesday, January 12, 2021 at 4:58 PM >>>>>>> *To: *"user@beam.apache.org" <user@beam.apache.org> >>>>>>> *Cc: *Reuven Lax <re...@google.com> >>>>>>> *Subject: *Re: Is there an array explode function/transform? >>>>>>> >>>>>>> >>>>>>> >>>>>>> @Reuven Lax <re...@google.com> yes I am aware of that transform, >>>>>>> but that’s different from the explode operation I was referring to: >>>>>>> https://spark.apache.org/docs/latest/api/sql/index.html#explode >>>>>>> <https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fspark.apache.org%2Fdocs%2Flatest%2Fapi%2Fsql%2Findex.html%23explode&data=04%7C01%7Ctaol%40zillow.com%7C1226a5d9efee43fc7d5508d8b75e5bfd%7C033464830d1840e7a5883784ac50e16f%7C0%7C0%7C637460963191408293%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=IjXWhmHTGsbpgbxa1gJ5LcOFI%2BoiGIDYBwXPnukQfxk%3D&reserved=0> >>>>>>> >>>>>>> >>>>>>> >>>>>>> How is it different? It'd help if you could provide the signature >>>>>>> (input and output PCollection types) of the transform you have in mind. >>>>>>> >>>>>>> >>>>>>> >>>>>>> On Tue, Jan 12, 2021 at 4:49 PM Tao Li <t...@zillow.com> wrote: >>>>>>> >>>>>>> @Reuven Lax <re...@google.com> yes I am aware of that transform, >>>>>>> but that’s different from the explode operation I was referring to: >>>>>>> https://spark.apache.org/docs/latest/api/sql/index.html#explode >>>>>>> <https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fspark.apache.org%2Fdocs%2Flatest%2Fapi%2Fsql%2Findex.html%23explode&data=04%7C01%7Ctaol%40zillow.com%7C1226a5d9efee43fc7d5508d8b75e5bfd%7C033464830d1840e7a5883784ac50e16f%7C0%7C0%7C637460963191418249%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=XuUUmNB3fgBasjDj0Dq1Z2g6%2Bc5fbvluf%2BnAp2m8cuE%3D&reserved=0> >>>>>>> >>>>>>> >>>>>>> >>>>>>> *From: *Reuven Lax <re...@google.com> >>>>>>> *Reply-To: *"user@beam.apache.org" <user@beam.apache.org> >>>>>>> *Date: *Tuesday, January 12, 2021 at 2:04 PM >>>>>>> *To: *user <user@beam.apache.org> >>>>>>> *Subject: *Re: Is there an array explode function/transform? >>>>>>> >>>>>>> >>>>>>> >>>>>>> Have you tried Flatten.iterables >>>>>>> >>>>>>> >>>>>>> >>>>>>> On Tue, Jan 12, 2021, 2:02 PM Tao Li <t...@zillow.com> wrote: >>>>>>> >>>>>>> Hi community, >>>>>>> >>>>>>> >>>>>>> >>>>>>> Is there a beam function to explode an array (similarly to spark >>>>>>> sql’s explode())? I did some research but did not find anything. >>>>>>> >>>>>>> >>>>>>> >>>>>>> BTW I think we can potentially use FlatMap to implement the explode >>>>>>> functionality, but a Beam provided function would be very handy. >>>>>>> >>>>>>> >>>>>>> >>>>>>> Thanks a lot! >>>>>>> >>>>>>>
