Обсуждение: Transaction control in a C language function

Поиск
Список
Период
Сортировка

Transaction control in a C language function

От
Flaris Feller
Дата:
Hello all,

I would like to implement transaction control in a function implemented in C language. Currently I'm using SPI to do the job.
When calling SPI_commit() it returns this error message:
SQL Error [XX000]: ERROR: portal snapshots (0) did not account for all active snapshots (1)
Below, there is an example.

PG_FUNCTION_INFO_V1(my_commit);

Datum my_commit(PG_FUNCTION_ARGS)
{
    SPIExecuteOptions execopts;

    SPI_connect_ext(SPI_OPT_NONATOMIC);

    memset(&execopts, 0, sizeof(execopts));
    execopts.params = NULL;
    execopts.read_only = false;
    execopts.allow_nonatomic = true;
    execopts.must_return_tuples = false;
    execopts.tcount = 0;
    execopts.dest = NULL;
    execopts.owner = NULL;

    /* Execute some SQL commands */
    SPI_execute_extended("UPDATE mytable SET name = 'done'", &execopts);

    /* Commit the transaction */
    SPI_commit();

    SPI_finish();

    PG_RETURN_NULL();
}

Thanks in advance,
Flaris Feller

Re: Transaction control in a C language function

От
Tom Lane
Дата:
Flaris Feller <flaris.feller@gmail.com> writes:
> I would like to implement transaction control in a function implemented in
> C language.

That will certainly not work unless you make it a procedure.
It'd be a good idea also to check that the CALL came from a
non-atomic context.

            regards, tom lane



Re: Transaction control in a C language function

От
Flaris Feller
Дата:
Hello Tom,
You are absolutely right, I was calling it from a function instead of a procedure.
The documentation should mention that.
Thanks for your clarification.
Regards,
Flaris.

Em seg., 6 de mar. de 2023 às 12:21, Tom Lane <tgl@sss.pgh.pa.us> escreveu:
Flaris Feller <flaris.feller@gmail.com> writes:
> I would like to implement transaction control in a function implemented in
> C language.

That will certainly not work unless you make it a procedure.
It'd be a good idea also to check that the CALL came from a
non-atomic context.

                        regards, tom lane


--