Description
Problem
Currently we run two queries every second (ref1, ref2, ref3), note that one does a DELETE to avoid having the http_request_queue from growing too much. This pollutes pg_stat_statements and causes some users to complain and wrongly think that pg_net is causing a lot of iops.
Solution 1
We can instead add the requests done with net.http_get and others to an in-memory queue, and then read from this queue whenever there are entries available.
This can be done without a breaking change by adding a trigger to the net.http_request_queue table.
Solution 2
A simpler change would be just waking up the worker whenever there's an http_get or other function call and put it to sleep whenever there are 0 rows processed.
This would mean migrating the net.http_<method> functions to use C instead of SQL.
Related
Solving this issue can be a stepping stone towards #62.