E-mail on my main system (ITS) has stopped working (for no obvious
reason). Apparently, the main SMTP queue has vanished.

ITS $ tcpip show version

HP TCP/IP Services for OpenVMS Industry Standard 64 Version V5.7 - ECO 5
on an HP rx2600 (1.50GHz/6.0MB) running OpenVMS V8.4-2L3

ITS $ tcpip start mail
Starting execution queue TCPIP$SMTP got status %JBC-E-NOSUCHQUE, no such queue
%JBC-E-NOSUCHQUE, no such queue

"SHOW QUEUE *smtp*" finds queues for other (stopped) cluster members,
but none for ITS. For example:

ITS $ show queue *smtp*it*
Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, stopped, on IT::, mounted form DEFAULT

Note that IT and ITS are different systems.

I'm guessing that the "no such queue" complaint refers to the missing
"Generic server queue TCPIP$SMTP_ITS_0". And that the individual
TCPIP$SMTP_ITS_* "service queues" might get created if that generic
queue existed.

Who creates this "Generic server queue" in the first place?

Steven,

the 'Generic server queue TCPIP$SMTP' is being defined and started by @SYS$STARTUP:TCPIP$SMTP_STARTUP
In a cluster with a common queue database, those generic TCPIP$SMTP_node_nn queues may be created for each node, depending on the setting of Number-Of-Queues-Per-Node : ... in TCPIP$SMTP_COMMON:TCPIP$SMTP.CONF

Volker.

> the 'Generic server queue TCPIP$SMTP' is being defined and started by
> @SYS$STARTUP:TCPIP$SMTP_STARTUP

Ok. Thanks.

> In a cluster with a common queue database, those generic
> TCPIP$SMTP_node_nn queues may be created for each node, depending on the
> setting of Number-Of-Queues-Per-Node : ... in
> TCPIP$SMTP_COMMON:TCPIP$SMTP.CONF

Ah. My cluster members have separate system disks, but there is a
common queue database (on ITS). All I see in TCPIP$SMTP.CONF related to
queues is:
Queue-Name :
which looks pretty harmless.

Further poking around found this in SYSTARTUP_VMS.COM (ahead of
TCPIP$STARTUP.COM):

[...]
$ THIS_NODE == F$EDIT( F$GETSYI( "SCSNODE"), "TRIM")
[...]
$! Keep SMTP mail delivery local. Use the specific, local queue
$! (TCPIP$SMTP_node_1), not the generic, cluster-wide queue
$! (TCPIP$SMTP).
$!
$ if (this_node .nes. "")
$ then
$ define /system TCPIP$SMTP TCPIP$SMTP_'this_node'_1
$ endif
[...]

which changes the interpretation of the error message:
Starting execution queue TCPIP$SMTP got status %JBC-E-NOSUCHQUE, no such queue
%JBC-E-NOSUCHQUE, no such queue

So:

ITS $ deassign /system TCPIP$SMTP

After that, restarting the SMTP service seems to have gotten "Server
queue TCPIP$SMTP_ITS_1" recreated, which (along with a "tcpip analyze
mail /repair) seems to have restored normal e-mail operation.

There's still no "Generic server queue TCPIP$SMTP_ITS_00", but ITS is
the only active cluster member at the moment.

Node IT had been running V8.3-1H1, but i just updated it to V8.4-2L3,
so IT has been restarted repeatedly, which might explain some or all of
this.

I'd guess that that definition of TCPIP$SMTP is less of a good idea
than it seemed to be years ago. (Before "!SMTP rollover to ASCII
configuration in V5.7:"? My SYSTARTUP_VMS.COM was dated 22-DEC-2017.)

Ok, I stopped defining TCPIP$SMTP (in SYSTARTUP_VMS.COM) on IT and
ITS, and things looked good for a while. Until I restarted IT:

IT $ show queue *smtp*it*
Server queue TCPIP$SMTP_ITS_1, idle, on ITS::, mounted form DEFAULT

Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT

[Looks the same on ITS, as expected.]

IT $ tcpip start mail

IT $ show queue *smtp*it*
Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT

it's not immediately obvious to me why starting the SMTP queues on IT
should delete the one on ITS. This way, every time SMTP on IT starts,
e-mail on ITS dies.

"TCPIP START MAIL" on ITS recreates the lost queue (without
disturbing the ones on IT):

ITS $ tcpip start mail

ITS $ show queue *smtp*it*
Server queue TCPIP$SMTP_ITS_1, idle, on ITS::, mounted form DEFAULT

Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT

Knowing nothing, I can't help but wonder if someone on IT is doing
something unpleasant with names like "TCPIP$SMTP_IT*" where it should be
doing that unpleasant thing with names like "TCPIP$SMTP_IT_*".

After awakening my (experimental) rx2660 ("ITX"), and straightening
out some unrelated problems (DECnet node registration, proxies, et c.,
because IT had never seen ITX before):

ITX $ tcpip show version

HP TCP/IP Services for OpenVMS Industry Standard 64 Version V5.7 - ECO 5
on an HP rx2660 (1.59GHz/9.0MB) running OpenVMS V8.4-2L3

IT $ show queue *smtp*it*
Server queue TCPIP$SMTP_ITS_1, idle, on ITS::, mounted form DEFAULT

Server queue TCPIP$SMTP_ITX_1, idle, on ITX::, mounted form DEFAULT

Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT

IT $ tcpip start mail ! Now it kills two with one blow:

IT $ show queue *smtp*it*
Generic server queue TCPIP$SMTP_IT_00

Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT

Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT

Again, "TCPIP START MAIL" on ITS or ITX recreates the lost queue
(without disturbing the ones on IT).

With a differently named (loud and close-to-my-head) rx2600 (REX)
(and ITX shut down):

REX $ tcpip show version

HP TCP/IP Services for OpenVMS Industry Standard 64 Version V5.7 - ECO 5
on an HP rx2600 (1.50GHz/6.0MB) running OpenVMS V8.4-2L3

IT $ pipe show queue *smtp* | search sys$input it, rex
Server queue TCPIP$SMTP_ITS_1, idle, on ITS::, mounted form DEFAULT
Server queue TCPIP$SMTP_ITX_1, stopped, on ITX::, mounted form DEFAULT
Generic server queue TCPIP$SMTP_IT_00
Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT
Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT
Generic server queue TCPIP$SMTP_REX_00
Server queue TCPIP$SMTP_REX_01, stopped, on REX::, mounted form DEFAULT
Server queue TCPIP$SMTP_REX_1, idle, on REX::, mounted form DEFAULT

IT $ tcpip start mail ! It still kills (only) two with one blow:

IT $ pipe show queue *smtp* | search sys$input it, rex
Generic server queue TCPIP$SMTP_IT_00
Server queue TCPIP$SMTP_IT_01, stopped, on IT::, mounted form DEFAULT
Server queue TCPIP$SMTP_IT_1, idle, on IT::, mounted form DEFAULT
Generic server queue TCPIP$SMTP_REX_00
Server queue TCPIP$SMTP_REX_01, stopped, on REX::, mounted form DEFAULT
Server queue TCPIP$SMTP_REX_1, idle, on REX::, mounted form DEFAULT

Thus, "TCPIP START MAIL" on IT whacks the queues on ITS and ITX, but
not those on REX. Bug? I like my sloppy-wildcard/locate hypothesis.

Steven,

there are some obscure warnings regarding the SET CONFIG SMTP commands in the VSI TCPIP V5.7 VSI TCP/IP Services for OpenVMS Management Command Reference guide:

"Restrictions
For clusters, issue this command only on the nodes where the SMTP queues reside — that is, on
nodes that are not using clusterwide queues and are not managing clusterwide queues for other nodes."

Also note that the queue names seem to have changed in TCPIP V5.7 (from TCPIP$SMTP_node_01 to ..._1)

Volker.

> there are some obscure warnings regarding the SET CONFIG SMTP commands
> [...]

I believe that the only "SET CONFIG SMTP commands" on any of these
systems came from TCPIP$CONFIG.COM. I can check, but I suspect that
their configurations are all very similar. Only my main system, ITS,
gets any SMTP access from the outside world. ITS normally runs
continuously; other nodes are transients.

> Also note that the queue names seem to have changed in TCPIP V5.7
> (from TCPIP$SMTP_node_01 to ..._1)

I assumed that the names like "_01" were old/obsolete, and I'm
willing to delete them, but I didn't want to make more changes than
necessary while investigating the main problem, namely, that "TCPIP
START MAIL" on node "IT" causes problems on other nodes with
similar-and-longer names, "IT*".

Note that years ago, when IT (a zx2000) got some actual use (and was
running VMS V8.3-1H1 and TCPIP V5.6), I never saw any such trouble.

Knowing nothing (except the observed behavior), my guess would be
that "TCPIP START MAIL" first deletes all the old stuff it can find, and
then creates the desired new stuff. And the problem is that the
difference between "_<node>*" and "_<node>_*" is causing it to find and
delete too much old stuff, _if_ you use an unlucky set of node names.

Still looks to me like a bug.