Re: our checks for read-only queries are not great

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Wed, Jan 8, 2020 at 2:57 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> * I find COMMAND_IS_WEAKLY_READ_ONLY to be a more confusing concept
>> than what it replaces. The test for LockStmt is an example --- the
>> comment talks about restricting locks during recovery, which is fine and
>> understandable, but then it's completely unobvious that the actual code
>> implements that behavior rather than some other one.

> Uh, suggestions?

COMMAND_NOT_IN_RECOVERY, maybe?

>> * ALTER SYSTEM SET is readonly? Say what?

> It would be extremely lame and a huge usability regression to
> arbitrary restrict ALTER SYSTEM SET on standby nodes for no reason.

I didn't say that it shouldn't be allowed on standby nodes. I said
it shouldn't be allowed in transactions that have explicitly declared
themselves to be read-only. Maybe we need to disaggregate those
concepts a bit more --- a refactoring such as this would be a fine
time to do that.

regards, tom lane