Usman Masood

Yes, that was an internal implementation we had. This one is built into PostgreSQL core and they use it for parallel aggregation. But we'll obviously not sacrifice performance or stability...

Upcoming Redis Streams could be a potential candidate here as well. It's API is closely modeled around Kafka, but will likely me much faster.

You'll probably have to create a different view since you're now aggregating over `id` and a time window. ``` CREATE CONTINUOUS VIEW online_counts AS SELECT id, minute(ts), sum(status) AS count...

Not sure if I follow. Do you mean you want to keep track of what's the least number of times you've seen a device online per minute vs what's the...

Could you describe the system you're trying to build at a higher level? That might be helpful for me to understand what you're trying to accomplish with your queries.

@derekjn: Is this needed for 0.8.0?

Postgres 10 is going to ship with full-fledged support for logical replication. Unless there is a big reason, this should probably happen post refactor? https://wiki.postgresql.org/wiki/New_in_postgres_10

Side-note: composite indexes will likely slow things down, especially if the values are wide. Maybe use a function to transform whatever columns you want into a value that you can...

I think the word `combine` might not be appropriate here since we use `combine` to denote combining two transition states. Maybe `merge` instead? The hard requirements I think are: -...

Postgres' trigger functions don't really allow batching because all they expose is a single `NEW` and `OLD` tuple. We have an optimized path for `pipeline_stream_insert` that processes the entire batch...