VIP failovers do not update the IPs named is listening on
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| MAAS |
Triaged
|
Low
|
Unassigned | ||
| 3.3 |
Triaged
|
Low
|
Unassigned | ||
Bug Description
I am experiencing intermittent DNS resolution failures throughout a cloud. This appears to be due to:
* deployed machines are told to use as DNSs all three physical maas addresses, plus the maas and pgsql VIPs (this can be overridden in the subnet configuration, but this appears to be the default)
* maas generates a listen-on option in /etc/bind/
* if a failover causes one or more VIPs to move, named.conf.options is not rewritten
I'm not sure what the rationale is behind listening on VIPs, but I think this could be solved by either having all maas nodes listen on all VIPs, regardless of whether the addresses are currently present or not on a given node, or by having deployed machines only use physical maas addresses by default.
I have worked around this by manually editing /etc/bind/
| Alberto Donato (ack) wrote | #1 |
| Changed in maas: | |
| status: | New → Triaged |
| importance: | Undecided → Medium |
| milestone: | none → next |
| Jerzy Husakowski (jhusakowski) wrote | #2 |
| Changed in maas: | |
| importance: | Medium → Low |
| milestone: | next → 3.3.0 |
| Changed in maas: | |
| milestone: | 3.3.0 → 3.4.0 |
| Changed in maas: | |
| milestone: | 3.4.0 → 3.4.x |
MAAS currently configures bind to listen on all interfaces.
It should probably only have it listen on interfaces for which dns is active in maas, although that's a bit trickier in deb installs, as there could be external additional config.