How do I fix Discovery "duplicating" AIX records when AIX server moves between Hardware frames?

[Thomas Vorster]

We are experiencing an issue with discovery of AIX Server CI with the serial number, CI entry duplication issue when we migrate/move virtual AIX servers between physical hardware and I am hoping that someone has encountered this before and was able to fix it.

Any help will be appreciated.

E.g. moving them from one of our Power 8 servers, to another - this created a duplicate CI. Then when when moving other AIX virtual hosts onto the same first Power 8, they adopted the old serial number, and now we have entries with incorrect hostnames, IPs etc, because of the serial numbers. The server to service relationships are also adopted causing incorrect service maps

Let me break it down into a bit more detail, e.g. using the server called ucmodc2 and twebap1a as an example:

1. Starting point - AIX LPAR (Logical Partition / Virtual server) called ucmodc2 is located on P8 with serial number 7806A88. It’s partition ID is 16. The service now serial number for CI entry ucmodc2 is thus 7806A88_16.

2. Moved AIX LPAR ucmodc2 to a new physical hosting P8 with serial number 7806A68, partition ID 5. A new CI entry is created for ucmodc2. Thus there are now 2 entries for ucmodc2. One with serial number 7806A88_16 (old with service mapping), and a new entry (no service mapping) for serial number 7806A68

3. We complete firmware upgrades on the P8 with serial number 7806A88 (where ucmodc2 was originally), and we need to move other AIX LPARs to this serial number to perform upgrades on the next P8 (7806A78). At this point, twebap1a is located on 7806A78 partition ID of 16, hence giving it the serial number 7806A78

4. I now move the AIX LPAR called twebap1a to P8 7806A88. It lands on 7806A88 with partition ID 16, and thus gets the serial number 7806A88_16, which is the old serial number for ucmodc2. And the CI entry is now confused:

twebap1a now has the CI entry details for the old CI entry of ucmodc2 (due to having the same serial number. The service mapping is incorrect, the IP is incorrect etc:

5. End-state - ucmodc2 is currently running on 7806A68_5(orphan CI entry), twebap1a is running on 7806A88_16 (CI merged with the old entry for ucmodc2), and the old twebap1a entry (on old serial number of 7806A78_16) is now a duplicate entry (but does have the correct mappings and info).

[robertgeen · Accepted solution]

Thomas,

I've seen this behaviour before with another client except it was more overdone then this in that it just had the serial number of the frame it was on with nothing concatenated on the end for the slot. My suggestion would be to modify the identification rule for AIX server and remove the serial number lookup at all from it. In this case what I am hearing is that name is actually what should be looked at for each of these regardless of what frame they happen to be on at that point so you can probably set name to the first order to run. The only reason why I recommend deactivating the serial number is that at some point you will have a new server that hasn't been discovered before that will contain a serial number that already exists on an older item (maybe it's moved frames) and you will end up updating that one instead of creating a new CI.

I hope this helps.