Re: Issue with PGC_BACKEND parameters

Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
> On 12/22/2013 11:30 PM, Amit Kapila wrote:
> - * backend start.
> + * backend start. However for windows, we need to process
> + * config file during backend start for non-default
> parameters,
> + * so we need to allow change of PGC_BACKEND during backend
> + * startup.
> */
> - if (IsUnderPostmaster)
> + if (IsUnderPostmaster && !IsInitProcessingMode())
> return -1;
> }

> I think this change looks OK.

The comment is pretty awful, since this is neither Windows-specific nor
a read of the config file. Perhaps more like "However, in EXEC_BACKEND
builds we load nondefault settings from the CONFIG_EXEC_PARAMS file during
backend start. In that situation we should accept PGC_SIGHUP
settings, so as to have the same value as if we'd forked from the
postmaster."

Also, I think that the extra test should only be made #ifdef EXEC_BACKEND,
so as to minimize the risk of breaking things. Not that this isn't pretty
darn fragile anyway; I think testing IsInitProcessingMode here is a very
random way to detect this case. I wonder if it'd be better to pass down
an explicit flag indicating that we're doing read_nondefault_variables().
If we don't do that, maybe an Assert(IsInitProcessingMode()) in
read_nondefault_variables() would be a good thing.

regards, tom lane