Problem: Completed print jobs remain visible in Samba print queues after upgrading to UCS 5.2-6

[fdavid]

Problem

After upgrading from UCS 5.2-5 to UCS 5.2-6 with Samba 4.24.2, completed print jobs may remain visible in Windows/Samba print queues although CUPS has already processed them successfully.

For an affected printer, CUPS reports no active jobs:

lpstat -W not-completed -o <PRINTER>

while Samba still returns completed jobs:

rpcclient localhost -U Administrator -c 'enumjobs <PRINTER>'

/var/log/samba/log.rpcd_spoolss contains messages such as:

fetch_share_cache_time: Error fetching timing data for [<PRINTER>]. Err:Success
print_cache_expired: Unable to get last scan timing for <PRINTER>

Investigation

Check the relevant Samba printing database:

python3 - <<'PY'
import os
import tdb

printer = "<PRINTER>"
path = f"/var/cache/samba/printing/{printer}.tdb"
db = tdb.Tdb(path, flags=os.O_RDONLY)

for key in (
    b"INFO/version\x00",
    f"CACHE/{printer}\x00".encode(),
    f"MSG_PENDING/{printer}\x00".encode(),
):
    try:
        value = db[key]
        print(key.rstrip(b"\0").decode(), len(value), value.hex())
    except KeyError:
        print(key.rstrip(b"\0").decode(), "missing")

db.close()
PY

The affected state looks like this:

INFO/version 4 09000000
CACHE/<PRINTER> 4 ...

The database is already version 9, but the existing CACHE/<PRINTER> value still uses the old 4-byte format.

Root Cause

Samba 4.21 uses print database version 8 with 32-bit timing values.

After upgrading to Samba 4.24.2, the first access to an existing print database changes INFO/version from 8 to 9, but an existing 4-byte CACHE/<PRINTER> record is not migrated or removed.

Samba 4.24.2 expects this value in the new 8-byte format. Reading the old record fails, preventing the print cache from being refreshed correctly. Completed CUPS jobs can therefore remain in the Samba queue.

See here for the bugreport.

Solution

As a workaround, remove only the obsolete CACHE/<PRINTER> record. Samba recreates it in the correct 8-byte format.

First verify that CUPS has no active jobs for the printer:

lpstat -W not-completed -o <PRINTER>

If the output is empty, stop Samba and create a backup:

systemctl stop samba-ad-dc
cp -a /var/cache/samba/printing/<PRINTER>.tdb /root/<PRINTER>.tdb.backup

Remove only the affected cache record:

python3 - <<'PY'
import tdb

printer = "<PRINTER>"
path = f"/var/cache/samba/printing/{printer}.tdb"
db = tdb.Tdb(path)

key = f"CACHE/{printer}\x00".encode()
value = db[key]
print("removing:", len(value), value.hex())
del db[key]

db.close()
PY

Start Samba again:

systemctl start samba-ad-dc

Verify that stale jobs disappear:

rpcclient localhost -U Administrator -c 'enumjobs <PRINTER>'

The recreated cache record should now use 8 bytes:

python3 - <<'PY'
import os
import tdb

printer = "<PRINTER>"
path = f"/var/cache/samba/printing/{printer}.tdb"
db = tdb.Tdb(path, flags=os.O_RDONLY)

key = f"CACHE/{printer}\x00".encode()
value = db[key]
print(key.rstrip(b"\0").decode(), len(value), value.hex())

db.close()
PY

Expected result:

CACHE/<PRINTER> 8 ...

This is a workaround for the affected print database state, not a permanent fix for the upgrade issue.