Re: Assertion failure from plan cache invalidation

Heikki Linnakangas <heikki(dot)linnakangas(at)enterprisedb(dot)com> writes:
> On 13/08/10 18:18, Tom Lane wrote:
>> Heikki Linnakangas<heikki(dot)linnakangas(at)enterprisedb(dot)com> writes:
>>> PushOverrideSearchPath() assumes that if the temporary namespace existed
>>> when an override search path was memorized with GetOverrideSearchPath(),
>>> it must still exist. That's not true in the above example, rolling back
>>> the transaction that the temporary namespace was created in drops it.
>>
>> Hm ... seems like there are two possibilities here. We could forcibly
>> recreate the temp schema, or we could just ignore the useTemp flag.

> Yeah, I was undecided on that too.

>> The former would more nearly approximate the situation that prevailed
>> at GetOverrideSearchPath() time, but on the other hand it's not clear
>> that it's a good idea for PushOverrideSearchPath() to have side-effects
>> like that. I *think* that it'd be safe, at least for the two existing
>> callers, but ...
>>
>> In the plancache case it could be argued that there's no real reason
>> to recreate the temp schema: it would necessarily be empty, so it
>> couldn't affect the results of planning anyhow. So the second solution
>> would work just fine for the current usage.
>>
>> Thoughts?

> Let's do the latter, add a comment noting that, and extend it later if
> necessary.

That's the way I was leaning, too. Will take care of it.

regards, tom lane