On 23.06.2020 10:15, James Sewell wrote:
Using multithreading in bgworker is possible if you do not use any
Postgres runtime inside thread procedures or do it in exclusive critical
section.
It is not so convenient but possible. The most difficult thing from my
point of view is error reporting.
Happy to be proved wrong, but I don't think this is correct.
PostgreSQL can call sigprocmask() in your BGWorker whenever it wants, and "The use of sigprocmask() is unspecified in a multithreaded process" [1]
Sorry, may be I missed something.
But in my bgworker I am not using Postgres runtime at all (except initial bgworker startup code).
So I am not using latches (which are based on signals), snapshots,...
In my case bgworker has no connection to Postgres at all.
Yes, it can still receives signals from Postmaster (SIGTERM, SIGHUP). But their handler are trivial and do not need to mask any signals.
So may be in general case combination of signals and threads may cause some problems,
but it doesn't mean that you can't create multithreaded bgworker.