Hi,

Cathy Zhou wrote:
> Hi Dan/Team,
> 
> I tried to test bfu from snv_74 to the clearview-uv bits today with the 
> following configuration:
> 
> a. aggr1 created over bge1
> b. /etc/hostname.aggr1
> 
> The system successfully booted but it fails to plumb the aggr1 
> interface. The problem is that the data-link configuration migration 
> (aggregation creation in this case) happens in the 
> /var/svc/profile/upgrade script, which is run by the 
> system/manifest-import service, and is after the network/physical 
> service (which plumbs the aggr1 interface).
> 
> I am wondering while you tested system upgrade, whether the same problem 
> showed up.
> 

I did not see the same problem, but I think that might be because I did 
not test the same way you tested.

I created a set of aggregations using bge[1-3].  I'd have one over bge1, 
and another over bge2 and bge3.  Once I created the aggregation, I'd 
make sure the configuration in /etc/aggregation.conf (or 
/etc/dladm/aggregation.conf, depending on the OS version) was correct, 
and then I'd upgrade the sytem to my test bits.  After the system 
rebooted, I'd check two things:

- the contents of /etc/dladm/datalink.conf had the correct information 
for both aggregations
- that dladm show-aggr showed the correct information for both aggregations.

If both were correct, I considered that the test passed and I went on to 
the next OS version.

I didn't try any tests involving /etc/hostname.aggrN, that was an 
oversight on my part.  Sorry.

> I can think of two options to solve this problem:
> 
> 1. Instead of write those "dladm" commands to /var/svc/profile/upgrade, 
> we write it to another script, and run that script at the very beginning 
> of network/physical service.
> 
> 2. Change the dependency so that instead of system/manifest-import 
> depending on network/physical, we make network/physical depends on 
> system/manifest-import.
> 
> Option 2 requires to break the current dependency between 
> manifest-import and network/physical. I don't know it is possible. 
> Today, system/manifest-import indirectly depends on 
> network/iscsi_initiator and system/identity:node, which both depends on 
> network/physical.
>

I think Seb's option three is the best way to go.

Dan


Reply via email to