Re: BUG #19879: Blocked COPY FROM reports NULL command and type in pg_stat_progress_copy

From: Raja Sai pranav <rajasaipranav0(at)gmail(dot)com>
To: pgsql-bugs(at)lists(dot)postgresql(dot)org
Cc: imchifan8(at)gmail(dot)com
Subject: Re: BUG #19879: Blocked COPY FROM reports NULL command and type in pg_stat_progress_copy
Date: 2026-11-20 17:07:11
Message-ID: nh-e1412cd3-239b-435b-99a5-b652080ce684@google.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-bugs

Hi,

I reproduced this on master (4a26e36f87) with your script and attached a
patch.

BeginCopyFrom() starts the progress entry with the relation, but sets the
command and type only after AllocateFile() (or OpenPipeStream() for
PROGRAM) returns. Opening a FIFO waits for a writer, so until then the view
shows the relation with NULL command and type. The type is already known
before the open, so the patch reports command and type just before it.

With the patch your script prints "observed=COPY FROM|FILE"; without it,
"observed=|". The core regression tests pass. I didn't add a test, because
it needs a FIFO and a timing loop, which doesn't fit the regression suite
and wouldn't run on Windows.

COPY TO starts its progress entry only after opening the file, so a blocked
COPY TO doesn't show up in the view at all instead of showing a half-filled
row. I left that alone.

Regards,
Sai Pranav

On Fri, Nov 20, 2026 10:36 PM, PG Bug reporting form <noreply(at)postgresql(dot)org>
wrote:

> The following bug has been logged on the website:
>
> Bug reference: 19879
> Logged by: Noah
> Email address: imchifan8(at)gmail(dot)com
> PostgreSQL version: 18.6
> Operating system: Linux/amd64
> Description:
>
> A server-side COPY FROM that blocks while opening a FIFO appears in
> pg_stat_progress_copy with the correct relation OID but NULL command and
> type. The view should identify the running command as COPY FROM and its I/O
> type as FILE, as documented for pg_stat_progress_copy.
>
> How to repeat:
>
> Save this as run.sh:
>
> #!/usr/bin/env bash
> set -euo pipefail
>
> : "${PG_BINDIR:?set PG_BINDIR to the target PostgreSQL bin directory}"
> base="$PWD/work"
> data="$base/data"
> sock="$base/socket"
> fifo="$base/input.fifo"
> mkdir -p "$base" "$sock"
> "$PG_BINDIR/initdb" -D "$data" --no-locale --encoding=UTF8 >/dev/null
> "$PG_BINDIR/pg_ctl" -D "$data" -o "-k $sock -c listen_addresses=''" -w
> start
> >/dev/null
> cleanup() {
> "$PG_BINDIR/pg_ctl" -D "$data" -m immediate -w stop >/dev/null 2>&1 ||
> true
> }
> trap cleanup EXIT
> "$PG_BINDIR/psql" -X -h "$sock" -d postgres -v ON_ERROR_STOP=1 -c 'CREATE
> TABLE copy_progress_target(i int)'
> mkfifo "$fifo"
> "$PG_BINDIR/psql" -X -h "$sock" -d postgres -v ON_ERROR_STOP=1 \
> -c "COPY copy_progress_target FROM '$fifo'" >"$base/copy.out"
> 2>"$base/copy.err" &
> copy_pid=$!
> for _ in $(seq 1 100); do
> row=$("$PG_BINDIR/psql" -XAt -h "$sock" -d postgres -c \
> "SELECT coalesce(command,'<NULL>') || '|' || coalesce(type,'<NULL>')
> FROM pg_stat_progress_copy WHERE relid='copy_progress_target'::regclass")
> test -n "$row" && break
> sleep 0.05
> done
> printf 'observed=%s\n' "$row"
> printf '1\n' >"$fifo"
> wait "$copy_pid"
> test "$row" = 'COPY FROM|FILE'
>
> Run it as a non-root operating-system user from an otherwise empty writable
> directory:
>
> chmod +x run.sh
> PG_BINDIR=/path/to/postgresql/bin ./run.sh
>
> On the affected build, it prints:
>
> observed=<NULL>|<NULL>
>
> The final test then exits nonzero. The progress row exists and has the
> relation OID for copy_progress_target, but its command and type are
> missing.
> Writing to the FIFO allows COPY to finish; the metadata is then populated
> or
> the progress row disappears.
>
> The issue was reproduced on PostgreSQL 18.6 (Debian 18.6-1.pgdg13+2) on
> Linux/amd64. COPY execution and table data are unaffected. This occurs when
> an authorized server-side COPY FROM opens a filesystem object whose open
> blocks while another session monitors pg_stat_progress_copy.
>
>
>
>
>
>
>

Attachment Content-Type Size
v1-0001-Report-COPY-FROM-command-and-type-before-opening-.patch text/x-patch 1.7 KB

In response to

Browse pgsql-bugs by date

  From Date Subject
Next Message Raja Sai pranav 2026-11-20 17:12:34 Re: BUG #19881: CLUSTER uses a disabled sort path when enable_sort is off
Previous Message Raja Sai pranav 2026-11-20 16:58:54 Re: BUG #19891: pg_restore ignores an explicitly empty -F argument