What happened
In the early stages of developing a background worker on Postgres 17.6, I built a worker that logged a few messages:
…
BackgroundWorkerBuilder::new("worker")
.set_start_time(BgWorkerStartTime::PostmasterStart)
.set_function("worker_main")
.set_library("worker")
.load();
…
#[pg_guard]
#[unsafe(no_mangle)]
pub extern "C-unwind" fn worker_main(_arg: pg_sys::Datum) {
log!("{} is starting", BackgroundWorker::get_name());
log!("{} stopped", BackgroundWorker::get_name());
}Postgres started successfully, but instead of my messages, Postgres logged:
[2025-10-22 00:21:52.679 CDT] [25769] [68f869f0.64a9]: LOG: background worker "worker": background workers without shared memory access are not supported
I had to call enable_shmem_access(None) for Postgres to start the worker.
BackgroundWorkerBuilder::new("worker")
.set_start_time(BgWorkerStartTime::PostmasterStart)
.set_function("worker_main")
.set_library("worker")
.enable_shmem_access(None)
.load();[2025-10-22 00:23:00.205 CDT] [26443] [68f86a34.674b]: LOG: worker is starting
What I expected
- I expected my initial set of builder calls to produce a valid background worker.
- I expected an error when I produced a background worker that Postgres didn't like. The
LOG level from Postgres honestly surprised me.
Other information
That message led straight to this discussion about why shmem access is necessary now: https://postgr.es/m/20210802065116.j763tz3vz4egqy3w@alap3.anarazel.de
What happened
In the early stages of developing a background worker on Postgres 17.6, I built a worker that logged a few messages:
… BackgroundWorkerBuilder::new("worker") .set_start_time(BgWorkerStartTime::PostmasterStart) .set_function("worker_main") .set_library("worker") .load(); … #[pg_guard] #[unsafe(no_mangle)] pub extern "C-unwind" fn worker_main(_arg: pg_sys::Datum) { log!("{} is starting", BackgroundWorker::get_name()); log!("{} stopped", BackgroundWorker::get_name()); }Postgres started successfully, but instead of my messages, Postgres logged:
I had to call
enable_shmem_access(None)for Postgres to start the worker.What I expected
LOGlevel from Postgres honestly surprised me.Other information
That message led straight to this discussion about why shmem access is necessary now: https://postgr.es/m/20210802065116.j763tz3vz4egqy3w@alap3.anarazel.de