Database Reference
In-Depth Information
Synchronizing between backends
All the preceding functions are designed to run in a single process/backend as if the
other PostgreSQL processes did not exist.
But what if you want to log something to a single file from multiple backends?
Seems easy—just open the file and write what you want. Unfortunately, it is not that
easy if you want to do it from multiple parallel processes and you do not overwrite or
mix up the data with what other processes write.
To have more control over the writing order between backends, you need to have
some kind of inter-process synchronization, and the easiest way to do this in Post-
greSQL is to use
shared memory
and
light-weight locks
(
LWLocks
).
To allocate its own shared memory segment your
.so
file needs to be pre-loaded,
that is, it should be one of the pre-load libraries given in
postgresql.conf
variable
shared_preload_libraries
.
In the
_PG_init()
function of your module, you ask for the address of a name
shared memory segment. If you are the first one asking for the segment, you are also
responsible for initializing the shared structures, including creating and storing any
LWLocks you wish to use in your module.
