Re: Nicer error when connecting to standby with hot_standby=off

On Tue, Mar 23, 2021 at 1:46 AM Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com> wrote:
>
>
>
> On 2021/03/23 3:59, James Coleman wrote:
> > Are you saying we should only change the message for a single case:
> > the case where we'd otherwise allow connections but EnableHotStandby
> > is false?
>
> No. Let me clarify my opinion.
>
> At PM_STARTUP, "the database system is starting up" should be logged
> whatever the setting of hot_standby is. This is the same as the original
> behavior. During crash recovery, this message is output. Also at archive
> recovery or standby server, until the startup process sends
> PMSIGNAL_RECOVERY_STARTED, this message is logged.
>
> At PM_RECOVERY, originally "the database system is starting up" was logged
> whatever the setting of hot_standby is. My opinion is the same as our
> consensus, i.e., "the database system is not accepting connections" and
> "Hot standby mode is disabled." are logged if hot_standby is disabled.
> "the database system is not accepting connections" and "Consistent
> recovery state has not been yet reached." are logged if hot_standby is
> enabled.
>
> After the consistent recovery state is reached, if hot_standby is disabled,
> the postmaster state is still PM_RECOVERY. So "Hot standby mode is disabled."
> is still logged in this case. This is also different behavior from the original.
> If hot_standby is enabled, read-only connections can be accepted because
> the consistent state is reached. So no message needs to be logged.
>
> Therefore for now what we've not reached the consensus is what message
> should be logged at PM_STARTUP. I'm thinking it's better to log
> "the database system is starting up" in that case because of the reasons
> that I explained upthread.
>
> Thought?

I understand your point now, and I agree, that makes sense.

The attached takes a similar approach to your v5, but I've used
CAC_NOTCONSISTENT instead of CAC_NOCONSISTENT because I think it reads
better (CAC_INCONSISTENT would technically be better English,
but...also it doesn't parallel the code and error message).

Thoughts?

James Coleman