just one comment: we need to be a bit carefull here. If you expect lifetime data, it should not be marker as sporadic.
Otherwise, if data arrives from different sources, DPL will not wait for all data to arrive, but once all timeframe data is there (i.e. immediately when everything is sporadic), start processing even if some of the sporadic inputs from other sources are missing. You will also not get any warning / error if expected sporadic inputs are missing.

This might be improved in the future, but right now, sporadic must not be used if inputs coming from different channels for the same tf are to be merged, as in that case only the input that arrives first is taken.

Anyway, it applies only to consumeWhenAll completion policy, for consumeWhenAny it doesn't matter anyway. Not sure what the mergers use.

Mergers is consumeWhenAny, so it does not wait for a complete slice. It cannot really count on it, since different QC tasks publish at different times and the timeslice id is the timestamp of the timer (AFAIK).

So I assume we are safe here.