deployer not working with known lxc to unit numbers

[John Garrett]

juju deployer failing to find the proper unit here. I get this failure:

2015-08-18 20:03:38 [WARNING] deployer.deploy: Service:percona-cluster, Deploy-with-service:all-in-one, Requested-unit-index=16, Cannot solve, falling back to default placement
2015-08-18 20:03:40 [WARNING] deployer.deploy: Service:percona-cluster, Deploy-with-service:all-in-one, Requested-unit-index=11, Cannot solve, falling back to default placement

unit exists, but there is a gap in numbering due to some stops and starts in this test lab:

cisco@maas:~/charms$ juju status --format short
- all-in-one/0: cimc57.dcnlab.cisco.com (started)
- all-in-one/10: cimc63.dcnlab.cisco.com (started)
- all-in-one/12: ch2blade7.dcnlab.cisco.com (started)
- all-in-one/15: cimc75.dcnlab.cisco.com (started)
- all-in-one/16: cimc61.dcnlab.cisco.com (started)
- all-in-one/19: cimc53.dcnlab.cisco.com (started)

I tried the 11 above to see if it was looking for machine ID instead of unit. machine 11 has unit 10:

cisco@maas:~/charms$ juju status --format tabular

11 started 1.24.3.1 cimc63.dcnlab.cisco.com /MAAS/api/1.0/nodes/node-6a35d714-3b98-11e5-a5cf-0024b400007e/ trusty arch=amd64 cpu-cores=32 mem=49152M tags=controllers

Config:
  services:
    percona-cluster:
      charm: percona-cluster
      local: trusty/percona-cluster
      num_units: 3
      options:
        vip: '10.xxxxx'
        root-password: xxxxx
        sst-password: xxxxx
        max-connections: 10000
      to:
        - 'lxc:all-in-one=15'
        - 'lxc:all-in-one=16'
        - 'lxc:all-in-one=11'

The code in this bug appears to show that it is counting the number of units using the unit number, which is more of an ID than a number. (Although it is generated in order..)
https://code.launchpad.net/~davidpbritton/juju-deployer/lxc-placement-fix/+merge/230924

In that case, I can see why this fails, but I am not a developer so I may be missing something with the way this is supposed to work.

Is there a way to renumber the units to make this work? Or is there workaround to clear "failed" numbers during deployment tries, and redeploy to keep things in order?

My goal is to deploy specific things to containers across specific nodes.

[John Garrett (jogarret)]

This issue may be related to an upgrade that never completed for me. I was at 1.24.3 and upgrade to 1.24.5 never completed. I was running this work in that state. I don't think the units were being seen (Blocked by the upgrade?)

I destroyed the juju environment and will try with 1.24.5 as a new environment and update here.

[John Garrett (jogarret)]

Not a problem on a clean build of juju. The partially finished upgrade may have been root cause of units not being available, even when juju status showed them available and agent started.

Can someone close/delete/junk this bug?