How to write to a file name defined at runtime?

I want to write to a gs file but I don’t know the file name at compile time. Its name is based on behavior that is defined at runtime. How can I proceed?

If you're using Beam Java, you can use FileIO.writeDynamic() for this (starting with Beam 2.3 which is currently in the process of being released - but you can already use it via the version 2.3.0-SNAPSHOT), or the older DynamicDestinations API (available in Beam 2.2).

Example of using FileIO.writeDynamic() to write a PCollection of bank transactions to different paths on GCS depending on the transaction's type:

PCollection<BankTransaction> transactions = ...;
transactions.apply(
    FileIO.<BankTransaction, TransactionType>writeDynamic()
      .by(Transaction::getType)
      .via(BankTransaction::toString, TextIO.sink())
      .to("gs://bucket/myfolder/")
      .withNaming(type -> defaultNaming("transactions_", ".txt"));

For an example of DynamicDestinations use, see example code in the TextIO unit tests.

Alternatively, if you want to write each record to its own file, just use the FileSystems API (in particular, FileSystems.create()) from a DoFn.

For the Python crowd:

An experimental write was added to the Beam python SDK in 2.14.0, beam.io.fileio.WriteToFiles:

my_pcollection | beam.io.fileio.WriteToFiles(
      path='/my/file/path',
      destination=lambda record: 'avro' if record['type'] == 'A' else 'csv',
      sink=lambda dest: AvroSink() if dest == 'avro' else CsvSink(),
      file_naming=beam.io.fileio.destination_prefix_naming())

which can be used to write to different files per-record.

If your filename is based on data within your pcollections, you can use the destination and file_naming to create files based on each record's data.