Can't start LAN failover

[Valentin Likoum]

Hello all,

I'm trying to start LAN failover on my ES40. Network card (accordinly to CLUE CONFIG) is DE602-AA. VMS 7.3-2 with UPDATE-V0600 patch.
DECnet5, TCPIP 5.4 are both configured to EIA interface. So I:
- read the manual
- lancp> define device lla
- lancp> define device lla/failover=(eia0,eib0)
- lancp> define device lla /enable
- lancp> list device lla/char

Value Characteristic
----- --------------
Yes Logical LAN enabled for startup
"EIA0" Failover device
"EIB0" Failover device

- reboot

No LLA device was appeared in volatile database after reboot. There was not any specific error messages. SET DEVICE/FAIL=... obviously isn't working as DECnet,TCPIP and LAT was started. May be DECnet configured for EIA prevented LAN failover to start?
So
- Should I do SET DEVICE/FAIL=... on every reboot? If so, where is the right place for it? Systartup_vms obviously isn't the right place as the DECnet is already started at this point
- Should I reconfigure DECnet, TCPIP for LLA device instead of EIA?
- Or more common: What's the right procedure to setup LAN failover on non-empty system with some running network stacks configured for interface which is supposed to be entered into failover set?

Thank you.

[Jean-François Piéronne]

Hi Valentin,

you need to stop all protocols (except SCS) running on you interfaces before setting LAN failover, this include AMDS, DECNET, IP, etc...
You can check this using
the sda command "sho lan"
or the lanacp "show conf /user"


JF

[Andy Bustamante]

Valentin,

You don't need to stop any protocols with your approach. By defining the device and rebooting the system, the failover lla0 interface should be present. Once the system boots, if LAN protocols are started too quickly, the failover device cannot be created. This may be an issue if your ES40 is in a cluster.

You should comment out network protocols, boot and reconfigure them to use the LLA0 device. In NCP (assume DECnet phase IV)
DEFINE LINE LLA-0 STATE ON
DEFINE CIRCUIT LLA-0 STATE ON COST 4
PURGE LINE EIA-0
PURGE LINE EIA-1
PURGE CIRCUIT EIA-0
PURGE CIRCUIT EIA-1

TCPIP SET CONFIG NOINTERFACE IE0
TCPIP SET CONFIG NOINTERFACE IE1

If that doesn't work, HP recommended the attached command file, called from sylogicals.com. In a cluster SCS traffic might conflict with the creation of the LLA0 device. This was a soon after the release of 7.3-2 and may no longer be required.

As a side note, I use 2 x DE-602 for creating interfaces. EIA and EIC are configured as fail over, connected to unique switching. I use the EIB and EID devices for cluster traffic. Running an LLA device on one NIC is a single point of failure.


Andy

[Jean-François Piéronne]

Stop all protocols or not start them is a subtle difference ;-)

Each time I have do this I have check that all protocol except SCS are stop ant this definitvely work.

Decnet Phase V startup is very different than Phase IV.

Another recommended approch is to boot minimum.

JF

[Michael D. Feeney]

Valetin, the problem is that you have to reconfigure the protocols to use the failover device instead of the physical device. You can't do that while they are running.

If you reboot, you have to keep DECnet and TCP/IP from starting so you can configure them to use the failover device. After that, reboot again and when they start they will be using the failover device. SCA will automatically use it too.

If you can't reboot, you have to stop DECnet and TCP/IP. Then you can use the LANCP SET DEVICE commands to create LLA0 in the volatile db, configure DECnet and TCP/IP to use LLA0, and restart DECnet and TCP/IP.

Use SYS$MANAGER:NET$CONFIGURE ADVANCED to configure DECnet-Plus and TCPIP$CONFIG.COM for HP TCP/IP.


Mike

[Volker Halle · Accepted solution]

Valentin,

the commands you've shown should have worked to activate the LAN Failover device after reboot.

Check SYS$MANAGER:LAN$ACP.LOG and the console for LLDRIVER, EIA and EIB error messages. Are there network cables connected to EIA and EIB at boot time ?

Are there any entries for these devices in the LANACP MOP database (LANCP> SHOW DEV/DLL) ? If so, they must be deleted/disabled !

What does SDA> SHOW LAN show after booting ?

Volker.

[Valentin Likoum]

Hmm... Strange enough. LAN$ACP.LOG contains:
--------------
2-JUN-2006 18:40:36.57 Defined LAN$DLL to be SYS$SYSROOT:[MOM$SYSTEM]
22-JUN-2006 18:40:36.72 Found LAN device EIA0, hardware address 00-02-A5-89-98-26
22-JUN-2006 18:40:36.84 Found LAN device EIB0, hardware address 00-02-A5-89-98-27
22-JUN-2006 18:40:37.20 Found LAN device LLA0, hardware address 00-00-00-00-00-00
22-JUN-2006 18:40:37.32 Failed to enable LAN device LLA0 in non-exclusive mode, status no such device available
22-JUN-2006 18:40:37.40 LANACP initialization complete
--------------

The same error ("Failed to enable LAN device LLA0 in non-exclusive mode") is in OPERATOR.LOG. I looked into STARTUP.LOG. LAN$STARTUP was called before DECnet and nothing interesting before it (may be excluding SHADOW_SERVER)

$ mc lancp lis dev/dll

Device Listing, permanent database:
--- Downline Load Service Characteristics ---
Device State Access Mode Clients Data Size
------ ----- ----------- ------- ---------
EIA0 Disabled NoExclusive NoKnownClientsOnly 246 bytes
EIB0 Disabled NoExclusive NoKnownClientsOnly 246 bytes
LLA0 Enabled NoExclusive NoKnownClientsOnly 246 bytes

So DLL seems to be disabled.
The system is standalone. That else can allocate EIx devices?

[Volker Halle]

Valentin,

once the system has booted, you can use SDA> SHOW LAN to see, which protocols have been started on EIA/EIB.

Using SDA> SHOW LAN/FULL/DEV=EIA you can see WHEN the protocols (EIAn) have been started: Last start attempt ...

If that time is BEFORE the LLDRIVER messages on the console (they do have timestamps), you at least know, why the devices could not join the LAN failover set.

If the MOPDL protocol is active, read on:

We had this problem happen on a cluster machine, when booting it with VAXCLUSTER=0 and NISCS_LOAD_PEA0=1. In this case, the EIA/B devices did not join the LAN failover set (MOPDL was active). Booting with VAXCLUSTER=2 worked fine.

It seemed that booting with VAXCLUSTER=2 causes LANACP to start very early and NOT consider the MOP DLL entries. When booting with VAXCLUSTER=0, LANACP will be started later and will also read/use the MOP DLL entries. We were able to solve this problem by disabled the DLL entries for the EIA/EIB device.

Could you (temporarily) create a new LAN$DEVICE_DATABASE.DAT without the EIA/EIB MOP DLL entries ?

Volker.

[Valentin Likoum]

Now it's more interesting. I added line "MC LANCP SHO CONFIG/USER" to the bottom of sys$startup:lan$startup.com to catch the started protocol which I thought prevented failover set to start. Now it's working.
The command didn't show any culprit but somehow switched somewhat in LANACP and failover set started to work.
Now I'm mostly happy. Mostly because I don't like to touch system-provided files which I'm not supposed to touch. Is there a supported place to put the command to? I can be wrong but sylogicals.com doesn't seem to be a good candidate as it starts before lan$startup.

[Volker Halle]

Valentin,

so this then seems to be some kind of timing problem, which you have 'worked around' by using the MC LANCP SHOW command. There is probably no better place to put this command. Let's hope that the workaround is stable and the problem doesn't show up again after a future OpenVMS upgrade - anyway you will then probably remember this workaround ;-)

Volker.

[Valentin Likoum]

Volker, sorry, I made 2 errors:
- made two modifications at once (edit lan$startup.com and disable dll on LLA)
- posted previous message before read your reply
So my theory about magic line in lan$startup.com was totally wrong. And the winner is "DEFINE DEVICE LLA/DLL=DISABLE" as you proposed. Of course, I can enable DLL on LLA somewhere in systartup_vms, no ptoblem at all. Thank you very much for solution and thank others who responded

[Valentin Likoum]

Definitly ENABLEd DLL on LLA device was the culprit.