On 2010-02-11 00:50 +0200, Marko Tiikkaja wrote:
> On 2010-02-10 23:57 +0200, Tom Lane wrote:
>> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>>> If the executor has buried in it the assumption that the snapshot
>>> can't change after startup, then does that mean that we need to start
>>> up and shut down the executor for each subquery?
>> Yes, I think so. That's the way it's always worked in the past;
>> see for example PortalRunMulti() and ProcessQuery(). I think trying
>> to change that is a high-risk, low-reward activity.
>> This probably means that the planner output for queries involving
>> writeable CTEs has to be a separate PlannedStmt per such CTE.
> I'm looking at this, but I can't think of any good way of associating
> the tuplestores from PortalRunMulti() with the correct CTEs. Any ideas?
Ok, what about the following:
- after planning the original query, standard_planner() goes through
the list of top-level CTEs and assigns a running number for each of
the DML WITHs, in the order they will be executed and returns a
list of PlannedStmts. all necessary statements wi have a flag
signaling that the result should be stored in a tuplestore.
- the portal keeps the list of tuplestores around and passes that
list to every query through PlannedStmt. it keeps on executing
the statements until it finds a PlannedStmt that doesn't want its
results stored anywhere and then frees the list of tuplestores
Does this sound reasonable? And more importantly, am I going to be
wasting my time implementing this? The 15th isn't that far away..
In response to
pgsql-hackers by date
|Next:||From: Joachim Wieland||Date: 2010-02-10 23:52:18|
|Subject: Re: Listen / Notify - what to do when the queue is full|
|Previous:||From: Kurt Harriman||Date: 2010-02-10 23:04:06|
|Subject: Re: Patch: Remove gcc dependency in definition of inline