MIME-Version: 1.0
References: 
 <CAPpHfduMOTnV5Zj2KGJ7zanL_10QvccZHtPUaDfJvBhsh9axnQ@mail.gmail.com>
In-Reply-To: 
 <CAPpHfduMOTnV5Zj2KGJ7zanL_10QvccZHtPUaDfJvBhsh9axnQ@mail.gmail.com>
From: Etsuro Fujita <etsuro.fujita@gmail.com>
Date: Sat, 4 Jul 2026 18:04:59 +0900
Message-ID: 
 <CAPmGK16+4qVT-OQ7i5Lw3a0C1ZO8NJp8JwHW9fqjjLe7wDv4hQ@mail.gmail.com>
Subject: Re: Bug in asynchronous Append
To: Alexander Korotkov <aekorotkov@gmail.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CAPmGK16%2B4qVT-OQ7i5Lw3a0C1ZO8NJp8JwHW9fqjjLe7wDv4hQ%40mail.gmail.com>
Precedence: bulk

Hi Alexander,

On Sat, Jul 4, 2026 at 7:00=E2=80=AFAM Alexander Korotkov <aekorotkov@gmail=
.com> wrote:
> ExecReScanAppend() unconditionally resets callback_pending for all AsyncR=
equests.  The problem is that postgres_fdw keeps its own knowledge for the =
same fact: PgFdwConnState.pendingAreq =E2=80=93 a pointer to "pending async=
 request" for a given connection.  That connection can be shared by several=
 partitions/foreign tables (postgres_fdw caches one connection per server+u=
sermapping pair). The blind reset in nodeAppend.c only touches the local As=
yncRequest.callback_pending; it never touches PgFdwConnState.pendingAreq, w=
hich correctly points to the still-dangling request.
>
> Later, when another partition sharing that same connection gets its own R=
eScan (for instance, its chgParam changed because of the LATERAL parameter,=
 and it already has a cursor open), it sends "CLOSE cursor" via pgfdw_exec_=
query().  Before sending any new command on the connection, that function f=
irst drains whatever request is still outstanding on it:
>
> if (state && state->pendingAreq)
>     process_pending_request(state->pendingAreq);
>
> And process_pending_request() starts with:
>
> Assert(areq->callback_pending);
>
> =E2=80=93 which fails, because the flag was corrupted some rounds earlier=
.
>
> The attached patch contains both the reproduction case and the fix.  The =
fix postpones the reset of the callback_pending flag to ExecAppendAsyncBegi=
n().  ExecAppendAsyncBegin() performs this cleanup along with ExecReScan(),=
 which completes the async fetch.

Interesting!  Thanks for the report and patch!  Will review.

Best regards,
Etsuro Fujita