[1.9] DNS records flap for a machine with several interfaces (DHCP) in the same network

Bug #1619625 reported by Victor Tapia
14
This bug affects 2 people
Affects Status Importance Assigned to Milestone
MAAS
Won't Fix
Critical
Unassigned

Bug Description

After migrating MAAS from 1.5 to 1.9, I have a group of servers deployed using the dynamic ip pool (with the interfaces set to DHCP). Those servers have multiple interfaces in the same network (for future bonding), and I'm observing the DNS entry change between those interfaces every time I deploy a new node in the cluster. These are the steps that trigger the change:

If a machine has been deployed with IPs 1.1.1.1 and 1.1.1.2 with the DNS entry (A record) pointing to 1.1.1.1, and at a certain point, the 1.1.1.2 interface asks for a renewal, it will create a lease entry with that IP and the same client-hostname of the node in the dhcp.leases file. Then, when a machine is deployed, the DNS zone file will be rebuilt using the latest lease assigned to every client-hostname, triggering a change in the DNS entry making it point to 1.1.1.2 instead of 1.1.1.1.

If the migration from dynamic to static assignments is not possible without redeploying: would it be possible to make the DNS entry point just the PXE boot interface (ignoring the other interfaces leases), or maybe backport the "<interface>.<machine fully-qualified-domain-name>" for non-PXE entries to 1.9?

Tags: sts internal
Revision history for this message
Andres Rodriguez (andreserl) wrote :

Hi Victor,

the <interface>.<fqnd> is a feature that is not backportable to 1.9 and is only available to 2.0+.

That said, can you be more specific as to which record gets created ? Do you have examples? Can you attach config files and log files ?

MAAS should only be creating a DNS record for the PXE interface.

Changed in maas:
status: New → Triaged
status: Triaged → Incomplete
milestone: none → 1.9.5
summary: - DNS records flap for a machine with several interfaces (DHCP) in the
- same network
+ [1.9] DNS records flap for a machine with several interfaces (DHCP) in
+ the same network
Revision history for this message
Yoshi Kadokawa (yoshikadokawa) wrote :

Hi,

I could reproduce the same problem in my environment(v1.9.4) as well.
What I did was,

1. Set only the range for DHCP. So static IP range is not set.[0]
2. Enlist 3 nodes. Those 3 nodes has 3 interfaces, and all 3 interfaces are in the same network(10.12.0.0/24)
3. Commission all 3 nodes
4. Deploy one node

After the deploy is finished, I could confirm that the IP address in the node and DNS registered IP address was different.[1]

I will attach the sosreport later.

[0] http://pastebin.ubuntu.com/23144714/
[1] http://pastebin.ubuntu.com/23144720/

Revision history for this message
Yoshi Kadokawa (yoshikadokawa) wrote :
Changed in maas:
importance: Undecided → High
importance: High → Critical
Revision history for this message
LaMont Jones (lamont) wrote :

Yoshi - your bug sounds like it's completely different ("maas hands out the same IP to multiple hosts", as opposed to "maas changes the IP of the host when there are multiple interfaces on the subnet.")

Could you report yours as a separate bug?

Victor - I'm having trouble reproducing your bug -- could you do this command and send us files.tar.bz2:

    tar cjf files.tar.bz2 -c / etc/bind/maas var/lib/maas var/log

Are there any user reserved addresses in the system?

(One place where I could see it flip-flopping is if we notice lease expiring on the boot interface -- that would cause a non-boot interface's IP to become the new IP until the lease was renewed.)

lamont

Revision history for this message
LaMont Jones (lamont) wrote :

The static range was added to 1.9 specifically because of issues like this. While it's still an issue that we should resolve (probably relating to our handling of discovered addresses as leases expire and the address is given to a new MAC), the best workaround is to create a static range and use "auto assign" or "sticky" instead of having DHCP assign the IPs.

Victor - in the reported situation, is the dhcp server internal to maas, or is an external dhcpd involved?

Revision history for this message
Yoshi Kadokawa (yoshikadokawa) wrote :

lamont - I will answer the question regarding dhcp server is internal or external.
In the reported situation, the customer is using dhcp server internal to maas.

Revision history for this message
Victor Tapia (vtapia) wrote :

Lamont: in this particular scenario, where there's a production Openstack deployment already running, moving all nodes to a different IP range won't be possible. Also, trying to progressively reduce the dynamic subnet size and creating a contiguous static subnet in order to migrate the ips from one pool to the other without changing them is a dangerous option we could consider, I guess.

tags: added: internal
Revision history for this message
Andres Rodriguez (andreserl) wrote :

We believe that this is not longer an issue in the latest releases of MAAS. If you believe this is still an issue, please re-open this bug report and target it accordingly.

Changed in maas:
status: Incomplete → Won't Fix
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.