Since nova does not believe in the existence of hostnames, we need
to set them ourselves when we boot new servers in launch-node.
Change-Id: Ib318224a09c1b0b748ab31e1ed507975b3190784
Some clouds may not have a metadata service and need to retrieve key
info via config drive. Add a flag to specifically request that a config
drive is provided to the instance booted by nova to facilitate this
information injection.
Change-Id: Ic41df5b34ea67ad62949244e064db82410077453
Created as a new script so that we aren't hosed if this doesn't
work with current providers.
Change-Id: Ia8d35d0acbfb773ca710c9383d9b746e786766bc
Depends-On: I26bc94408441edf067493b7ffd50eebd9dd95e75
We have a set of hostname patterns which is not a thing that ansible
supports in inventory files. While we can put hostname patterns into
playbooks directly, that does not help us with copying hiera group files
since ansible doesn't know about the groups in site.pp and puppet
doesn't know about the ansible groups.
Instead, do a quick expansion any time the groups.txt file changes and
at the end of launch-node. It will be left to admins to run
expand-groups.sh whenever they delete a node.
Change-Id: I00c60748ddb2d35a3b98f78d828dabebcf065118
With the puppetmaster not there anymore, we should consume inventory
from OpenStack rather than from puppet.
It turns out that because of the way static and dynamic inventories get
merged, the static file needs to stand alone. SO - if you need to
disable a dynamic host from OpenStack (pretty much all of our hosts) you
need to not only add it to dynamic:children, you need to add an emtpy
group into the static file too, otherwise you'll get an error like:
root@puppetmaster:~# ansible -i newinv '!disabled' --list-hosts
ERROR: newinv/static:4: child group is not defined: (jenkins-dev.openstack.org)
Change-Id: Ic6809ed0b7014d7aebd414bf3a342e3a37eb10b6
This is a quick hack to handle public ip addrs on OVH. A better
fix is porting to shade or ansible.
Change-Id: I62bba7215249521307be5b42af1c6f8fc886d982
This parameter is usually not necessary, and over-specifying it
can be problematic because the value is not consistent.
Change-Id: I0a90631499294e7a6eb287f24739cf4884a8db7b
New systemd based distros reboot so quickly that the ssh connection
errors returning 255 (or -1 in python because signed integers). Ignore
return codes of -1 when rebooting over ssh as a result. All other return
codes will be propogated properly.
Change-Id: I272f00e9e07f1ed04f2b97d0e1609c6e8d49caf3
When adding a new server now, Ansible needs the root ssh key
fingerprint on the new server to be verified on the puppetmaster,
add this to instructions for setting up a new server.
Change-Id: I7345c22c47120c946b64e72aa6bf7a7a8f590132
In order to run launch-node.py for creation of new infra servers,
you need to be in the launch/ directory of the system-config
repo on disk, documenting this in the launch/README file.
Change-Id: I8499a7b8a43522cd365054d8f8c279ed15abe696
Because, you know, we should have to do this.
Rackspace publishes two things called "compute" into their catalog for
customers who still have the old compute service.
Change-Id: I27d5a78fc743057a270acd4bf7c85fd998f9074d
* launch/launch-node.py: Now that we're using --detailed-exit-codes
with puppet, we should not abort when the ssh call exits nonzero.
Puppet explicitly returns nonzero codes which indicate success
conditions, and we already have a function in place to interpret
them.
Change-Id: If66d345902d05fa0898d202f5f71c087df9f4ce7
HP Cloud centos7 images are currently failing with 'Please login as
the user "centos" rather than the user "root"'.
This adds the 'centos' user as one of the logins to try, and adds a
little documentation to the subsequent fixup of the authorized_keys
file
Change-Id: I46b62c8c4aed7a1c9522d60db7f726ebcb30281c
We've got the logic elsewhere to properly deal with puppet bizarre exit
code strategy. Put it in launch_node too.
Change-Id: I4a414a27b58b2d4d890020f4b2aaed50f7c8f26b
Currently launch-node.py doesn't allow to specify network label
and floating IP pool name, using the defaults from nova-network.
This patch allows to specify network label to attach instance to,
and floating IP pool from which floating IPs could be assigned.
Change-Id: I46024bc1c217f99f19717aeb04d424a01440203a
The new code to create volumes incorrectly affects the non-volume
case due to an indentation bug. This corrects that.
Change-Id: I18eb1222447393e769fbbbf550407bf103b11e99
On HP, when we launch nodes, we need them to have a floating IP for us
to be able to connect to them.
Change-Id: I5c7ad3183d0b6329e2019a2c95968fcf3984618b
We are using cinder block devices more and more. Update launch node so
that it can attach a preexisting cinder volume against a new nova server
on first boot. This will allow puppet or other config management to
format and mount that block device into the VMs filesystem.
Change-Id: Ic121cdc06dcbea0e38e8d0ff8946e999af3d727e
It is deemed useful by some to pass IDs rather than names to
launch-node.py for flavors and images; mostly due to the fact that
names are not unique, and the "first-matched" behavior on names
means that it can end up doing the wrong thing in certain cases.
Change-Id: Ief89174fedac7921f85bbfe3a456059ecc0f19bf
We don't really use the salt infrastructure that we set up, which means
it's cruft. Go ahead and admit that we don't use it and remove it from
our systems.
Change-Id: Ic43695719cbad30aded16ac480deb3dfd9b2a110
* launch/README: Correct outdated references to salt and puppet
certs. Modernize the recommended image name.
* launch/launch-node.py: When run as root, keep the puppet group on
private keys so that they're readable by puppet group members. That
way it won't break subsequent runs as non-root following the current
instructions in the README file. Modernize the default image name.
Change-Id: Ic5034121d4d0df3c096c2e474a9d3d41ec361021
According to https://docs.python.org/3/howto/pyporting.html the
syntax changed in Python 3.x. The new syntax is usable with
Python >= 2.6 and should be preferred to be compatible with Python3.
Enabled hacking check H231.
Change-Id: I4c20a04bc7732efc2d4bbcbc3d285107b244e5fa
Added the abillity to use launch_node.py to connect to and create
infrastructure instances in openstack setups that are not setup to use
the ssl secured clients.
This matches the functionality of the nova cli to accept NOVACLIENT_INSECURE.
Change-Id: Ibfb5bb4665f36a155ec488c475e03f9cd901e9d6
Signed-off-by: Philip Schwartz <philip.schwartz@rackspace.com>
Add a script to handle cloud servers with no swap. This can be
used before install_puppet when developers run it by hand, and
is called directly by launch_node. If the host has
no swap, create a swap partition from the ephemeral disk and use
the remainder on /opt.
Change-Id: Iac8bfe4924d8607ddbddc45312a51fe52eb4cdbc
We'd like to be able to control sequencing of how and when puppet
runs across our machines. Currently, it's just a set of agents
that run kinda whenever they run. At times they hang and we don't
know about it. Also, cross-server sequencing is impossible to
achieve.
Change the operation away from agents running on the machine as
daemons, and instead ssh from the master to each machine.
Change-Id: I76e41e63c6d0825e8735c484ba4580d545515e43
In anticipation of swtiching to flake8 as part of the removal of
pyflakes jobs, clean up a few flake8 errors so that we don't have to
ignore as many things.
Change-Id: Idb8466bb3194904d84e17dd3f2a7c32aff13788c
Commit 5ae5e6cc added puppetry for /root/ci-launch on the Puppet
master server, but set permissions on it too restrictive for users
launching new servers to be able to read the files within it.
* launch/README: Note that the user following these directions
should also be in the admin group.
* modules/openstack_project/manifests/puppetmaster.pp: Set group
ownership of /root/ci-launch to admin so members of that group will
be able to read the files within it.
Change-Id: I6c657eb4311b27ce329f249df3e60c2b902677ae
This is a minor step towards PEP-8 as well, but mainly makes it easier
to document what third party modules are needed.
Change-Id: Id901ef437fcdeb11c8dd555e8db7c13daa86c4b2
Infrastructures derived from the OpenStack CI infrastructure currently
need to fork launch-node.py, better if it just takes an option ;).
Change-Id: Iad94451ed2b4da425f059595ae727dcf2847733a
* launch/launch-node.py: Make salt key management when launching nodes
optional as we may not always be running salt or choose to have a node
run a salt daemon. Default to disabling management.
Change-Id: If1bf137e91d250d5c94a957ff00d19401685c730
* launch/README: On servers where the puppet agent daemon needs to
enabled, don't run the initscript directly under sudo. Passing it
through su avoids permission issues with the puppet mysql module.
* launch/launch-node.py: Only test IPv6 connectivity when providers
support it.
Change-Id: Ic37305331ba57ba3de2fb349e1ab503cabcc697d
* launch/launch-node.py: Sometimes we get a broken server from a
provider, probably actually a broken compute node behind the scenes,
which fails to send IPv6 packets with a DSCP > 3. Test that we can
send DSCP 0x10 packets to at least one IPv6-connected host and get a
response, or else discard the server.
Change-Id: I379589d74f9e5892e3fda85bd267567038b6c2c9
