| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1887.1 | /var/ase/lib/ni_status_awk script. | BACHUS::DEVOS | Manu Devos DEC/SI Brussels 856-7539 | Tue Feb 18 1997 07:19 | 18 |
| Hi Ron,
If the system running the service is detecting that the card itself is broken,
then the control is given to the ni_status_awk script, otherwyse ASE is simply
saying that there is a partitioned network, in which case it kill the
asedirector so no subsequent failover is possible for any cause.
ASE is able to check if the card is down because it checks the "SEND" hardware
counter everytime it sends a packet. If this counter is not changing, then the
card is down, otherwyse it is the network or the other system.
When the ni_status_awk script is invoked on one system, its default behaviour
is to "DISCONNECT" the system from ASE when ALL the "monitored" network of the
system are DOWN, so a failover of the services is done.
You can now change this default behaviour to do what you want.
Hope it helps, Manu.
|
| 1887.2 | | LEXSS1::GINGER | Ron Ginger | Tue Feb 18 1997 11:47 | 23 |
| Thanks, but Im still confused. Lets try a diagram.
___________ ------------
| | | |
| host A | | host B |
----------- -----------
| |
-----------| |----------|
| |
---------
| net hub |
----------
|------------> to "the world"
suppose A is serving NFS, and someone trips over the wire between Host A
and the hub, or the hub card for host A fails. ASE will not be able to
ping between A and B.
I would expect ASE to failover the service to B, is this correct?
If it was B's wire or hub card that had broken, then this would be bad,
since A is still connected to the hub and running fine. How does ASE
handle this?
|
| 1887.3 | 2 paths... | TROOA::MSCHNEIDER | martin.schneider@tro.mts.dec.com | Wed Feb 19 1997 12:07 | 4 |
| ASE uses both the network interface and the shared SCSI bus to ping the
other host. So the failure of one interface should not preclude the
hosts from determining the state of the other host via the alternate
2nd bus.
|
| 1887.4 | | LEXSS1::GINGER | Ron Ginger | Wed Feb 19 1997 12:26 | 7 |
| OK, in my example both hosts are fine, but something in the net is broken
that prevents B from getting to the net. Net pings will fail. Will ASE
decide to stop the service on A and fail it over to B, even though A is
still reaching the rest of the net just fine?
Has it been considered to have ASE ping some other 'independent' party
out on the net to decide if it can reach real net clients?
|
| 1887.5 | NO FAILOVER in this case !!! | BRSDVP::DEVOS | Manu Devos DEC/SI Brussels 856-7539 | Wed Feb 19 1997 19:44 | 15 |
| No Ron,
As I explained in .1, the Network card on A (and on B) is OK as proved
by the "send" hardware counter changing with packet. SO the ASEDIRECTOR
is killed and NO failover will occur either NOW or LATER.
In this particular case, ASE is NOT able to know if the ping is not
working because a problem is affecting SYSTEM A cable or system A or B
HUB card or SYSTEM B cable or SYSTEM B network card. So, ASE is simply
preventing itself to fail any service over to the other host as long as
this problem is persisting.
Is it clear ?
Manu.
|
| 1887.6 | | USCTR1::ASCHER | Dave Ascher | Thu Feb 20 1997 10:41 | 26 |
|
I think it would be useful to directly address Ron's last inquiry.
There was some ambiguity in the .5 response.
<<< Note 1887.4 by LEXSS1::GINGER "Ron Ginger" >>>
OK, in my example both hosts are fine, but something in the net is broken
that prevents B from getting to the net. Net pings will fail. Will ASE
decide to stop the service on A and fail it over to B, even though A is
still reaching the rest of the net just fine?
To clarify - you are saying that "A" has NO net problems:
good net card, good cable, can talk to other nodes but not
B; B cannot reach the net because either it's net card is
broken or it cable is out.
It would be a definite design bug if ASE decided to failover
under this scenario. In the tests that I have run at customer
sites (only pulling cables out - not failing the net interface
card) ASE does the right thing. It does NOT failover the
service due to a net problem on a node NOT running the service.
Has it been considered to have ASE ping some other 'independent' party
out on the net to decide if it can reach real net clients?
|