MAAS

  1. Bug #1519527
  2. Comment #23

Comment 23 for bug 1519527

Revision history for this message
Mike Pontillo (mpontillo) wrote : Re: 1.25.1 proposed: lxc units all have the same IP address

One issue we were seeing here with regard to MAAS was, when errors would occur in the middle of an API call, they would silently fail, and we wouldn't know why.

I worked with Andreas to triage this by having him apply this patch (via "cat patch | sudo patch -p2 -d /usr/lib/python2.7/dist-packages/maasserver"):

http://paste.ubuntu.com/13505469/

(as an aside, I want to land that as well - for support reasons, until we have proper observability via the API.)

We then saw this in the log:

maas.exception: [ERROR] The IP address 10.245.202.8 is already in use.

This points to _attempt_allocation() in staticipaddress.py. When this occurs, the expected behavior is that the transaction will be marked as a serialization failure, and retried (for example, if two containers are trying to reserve the same IP address in parallel, you might expect that to happen). It's a bit of a mystery why (in this setup) the serialization failures are reproducible, even when you run claim_sticky_ip_address from the command line.