Hey Nick,
Thanks! Yep that’s an alternative (together with a uniqueness constraint and retry mechanism)
I guess what I’m trying to get my head around is whether and why this would be better than using advisory locks…
Cheers
Perryn
Ok, so you want to allow _other_ updates to a customer while this process is happening? In that case, advisory locks will probably work. The only consideration is that the 'id' is a bigint. If your customer id maps to that, great. If not (for example we use UUID's), you will need some way to convert that id to a bigint.
Alternatively, create a new table that records the start timestamp of the most recent run of your code block for each customer, and update that as the first action in your transaction. Then row locks on that table will protect you from concurrent runs.