What you are looking for is a BucketingSink that works on event time (the
timestamp is encoded in your data).
AFAIK, Flink's BucketingSink has been designed to work in processing time,
but you can implement a Bucketer that creates buckets based on a timestamp
in the data.
You might need to play around with the parameters for closing open buckets
for a good behavior (similar to watermark tuning).
2018-02-14 22:18 GMT+01:00 robert <tel...@yahoo.com>:
> I need to grab avro data from a kafka topic and write to the local file
> Inside the avro record there is a date time field. From that field I need
> name the file accordingly. (20180103) as an example
> I was thinking of using flink to read, unpack this generic record then put
> to a sink that will sort to make sure it goes into the right file.
> Does anyhow have a high-level approach for this ?
> The bucketing sink look promising. Any examples of this type of problem for
> flink to solve ?
> Sent from: http://apache-flink-user-mailing-list-archive.2336050.