On Fri, Aug 7, 2020 at 1:21 PM Justin Pryzby <pryzby@telsasoft.com> wrote:
> On Fri, Aug 07, 2020 at 01:13:51PM +0900, Amit Langote wrote:
> > On Fri, Aug 7, 2020 at 1:05 PM Justin Pryzby <pryzby@telsasoft.com> wrote:
> > > On Fri, Aug 07, 2020 at 12:16:11PM +0900, Amit Langote wrote:
> > > > Curiously, Justin mentioned upthread that the crash occurred during
> > > > BIND of a prepared query, so it better had been that a custom plan was
> > > > being executed, because a generic one based on fewer partitions would
> > > > be thrown away due to A.I.M. invoked during AcquireExecutorLocks().
> > >
> > > Well this statement should only be executed once, and should be using
> > > PQexecParams and not PQexecPrepared (pygresql: pg.DB().query_prepared()).
> > >
> > > (gdb) p portal->name
> > > $30 = 0xf03238 ""
> > >
> > > (gdb) p portal->prepStmtName
> > > $31 = 0x0
> > >
> > > (gdb) p *portal->cplan
> > > $24 = {magic = 953717834, stmt_list = 0x682ec38, is_oneshot = false, is_saved = true, is_valid = true, planRoleId
=16554, dependsOnRole = false, saved_xmin = 0, generation = 1, refcount = 1, context = 0x682dfd0}
> > >
> > > I'm not sure why is_oneshot=false, though...
> >
> > Perhaps printing *unnamed_stmt_psrc (CachedPlanSource for an unnamed
> > statement) would put out more information.
>
> (gdb) p *unnamed_stmt_psrc
> $49 = {... gplan = 0x0, is_oneshot = false, is_complete = true, is_saved = true, is_valid = false, generation = 1,
node= {prev = 0x12fcf28,
> next = 0xdf2c80 <saved_plan_list>}, generic_cost = -1, total_custom_cost = 12187136.696805555, num_custom_plans =
1}
From this part, I think it's clear that a custom plan was used and
that's the only one that this portal seems to know about. Also, I can
see that only SPI ever builds "oneshot" plans, so is_oneshot would
always be false in your use case.
--
Amit Langote
EnterpriseDB: http://www.enterprisedb.com