Feedback-ID: ia2694551:Fastmail
Date: Tue, 24 Feb 2026 19:29:17 +0100
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Antonin Houska <ah@cybertec.at>
Cc: Mihail Nikalayeu <mihailnikalayeu@gmail.com>, 
	Pg Hackers <pgsql-hackers@lists.postgresql.org>, Robert Treat <rob@xzilla.net>
Subject: Re: Adding REPACK [concurrently]
Message-ID: <202602241757.6ac3iss2u4vo@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <202602231821.eqzziask6rjj@alvherre.pgsql>
Archived-At: <https://www.postgresql.org/message-id/202602241757.6ac3iss2u4vo%40alvherre.pgsql>
Precedence: bulk

On 2026-Feb-23, Alvaro Herrera wrote:

Looking at this function in pgoutput_repack.c:

> +/* Store concurrent data change. */
> +static void
> +store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
> +			 HeapTuple tuple)
> +{

[...] we have this:

> +	size = VARHDRSZ + SizeOfConcurrentChange;
> +
> +	/*
> +	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
> +	 * context and frees them after having called apply_change().  Therefore
> +	 * we need flat copy (including TOAST) that we eventually copy into the
> +	 * memory context which is available to decode_concurrent_changes().
> +	 */
> +	if (HeapTupleHasExternal(tuple))
> +	{
> +		/*
> +		 * toast_flatten_tuple_to_datum() might be more convenient but we
> +		 * don't want the decompression it does.
> +		 */
> +		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
> +		flattened = true;
> +	}
> +
> +	size += tuple->t_len;
> +	if (size >= MaxAllocSize)
> +		elog(ERROR, "Change is too big.");
> +
> +	/* Construct the change. */
> +	change_raw = (char *) palloc0(size);
> +	SET_VARSIZE(change_raw, size);

I wonder if this isn't problematic with large tuples.  If a row has some
very wide columns, each of which individually is less than 1 GB, then it
might happen that the sum of their sizes exceeds 1 GB, causing palloc()
to complain and abort the whole repack operation.  This wouldn't be very
nice, so I think we need to address it somehow.

Another thing I'm not very keen on, is the fact that we have to memcpy()
the tuple contents a few lines below:

> +	/*
> +	 * Copy the tuple.
> +	 *
> +	 * Note: change->tup_data.t_data must be fixed on retrieval!
> +	 */
> +	memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
> +	memcpy(dst, &change, SizeOfConcurrentChange);
> +	dst += SizeOfConcurrentChange;
> +	memcpy(dst, tuple->t_data, tuple->t_len);

> +	/* Store as tuple of 1 bytea column. */
> +	values[0] = PointerGetDatum(change_raw);
> +	isnull[0] = false;
> +	tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
> +						 values, isnull);

To make matters worse, tuplestore_putvalues does a
heap_form_minimal_tuple() on this and copies the data again.  This seems
pretty wasteful.

I think we need some new APIs to avoid all this copying.  It appears
that it all starts with reorderbuffer doing something unhelpful with the
memory context of the TOAST chunks.  Maybe we should address this by
"fixing" reorderbuffer so that it doesn't do this, instead of playing so
many games to cope.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/