The most interesting part about this talk was not just about tweaking the database with partitioning tricks or other time-oriented extensions, it was actually focusing on normalization and tweaking composite indexes to keep the work on the logarithmic lane, but missing the query plan examination!
Good talk indeed. I really need to test if postgres can run that fast. I do wish we got some stats on impact of normalisation, index, and triggers on insert
Hi, one question. Grafana have some dificult to read data by SQL query on DATETIME format? Using type Timestamp worked the consult, but using datetime not. The Graphic showed hour wrong. I'm very grateful if anyone can help me. Obs.> I need read one table where have on Datetime formate.
Is there any way to ensure that the WHERE clause gets evaluated before the JOIN? Leaving it to postgres to sort this out seems like a very risky trial/error design, that could get broken with an unexpected postgres update.
I think I can answer my own question. You would subquery the metric table with the latter two WHERE clauses and then pass the ID as an IN clause to the query on the value table so that you look at significantly less values. If you want to be even more specific you could make that a subquery to your range clause as well. This seems more stable to me, but maybe creating too many subqueries like this cuts down on postgres' ability to parallelize operations. Can someone provide some detail as to why the presenter chose to leave the query so imperative?
the presenter mentioned the reason is to reduce operational burden: you can use a single db technology for the whole system. also from the video it looks like PostgreSQL can scale well with large query and dataset pretty well. i think 1 scenario that this approach won't fit it if they want much higher write throughput, which LSM based db is definitely better on.
