S3 output: Is there a way to have every event in its own file?

[rickfish (Rick A Fishman)]

I have an unusual requirement to put every event in a separate file in an s3 bucket. I have tried to use this:

rotation_strategy => "size"
size_file => 1

but that is just an estimate and it still puts multiple events into the same file. I have also tried to use this:

tags => "%{[event_unique_id]}"

but that is not dynamic per-event, it just puts that string into every file name.

The only way I could get it to work is to set the prefix, i.e.

prefix => "%{[event_unique_id]}"

but that creates separate subfolders for each file and it decreases performance significantly.

Is there a way I can accomplish this?

[Badger]

@rickfish

Is there a way I can accomplish this?

No, there is not. The multi_receive_encoded method in the filter receives a batch of events. It processes them all and then calls the rotate_if_needed function. It deliberately only calls fstat once for the entire batch as an optimisation.

[rickfish (Rick A Fishman)]

Ok, thanks for the quick response

[Badger]

Come to think of it, if you set pipeline.batch.size to 1 then you might get the result you want. However tiny batches are less efficient that regular batches (125 events). Not sure how much less.

[rickfish (Rick A Fishman)]

That works @Badger but, as you suggested, performance is really bad as there could be tons of these events. I actually question the validity of this requirement and will go back to the users and try to change it. Thanks so much for giving me the solution!