| Age | Commit message (Collapse) | Author |
|---|
|
|
|
This is a fix for the failing test_fresh_install_with_config.
|
|
Add test for not displaying HA hints for 3.0 installs
|
|
Small update to remove the hint for multiple masters if we are doing
a 3.0 installation.
|
|
This addresses the stack trace that has been plaguing recent demos. In the
case of an error with callback_facts.yaml the program output is much clearer
and a course of action is suggested.
|
|
|
|
working with preconfigured load balancer
|
|
Fedora changes:
|
|
- ansible bootstrap playbook for Fedora 23+
- add conditionals to handle yum vs dnf
- add Fedora OpenShift COPR
- update BYO host README for repo configs and fedora bootstrap
Fix typo in etcd README, remove unnecessary parens in openshift_node main.yml
rebase on master, update package cache refresh handler for yum vs dnf
Fix typo in etcd README, remove unnecessary parens in openshift_node main.yml
|
|
|
|
Displays each host and the roles it will play based on the current
configuration. As the configuration grows the summary will adapt to indicate
embedded vs separte etcd, scheduled vs unscheduled nodes, etc.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
If only 2 masters are specified, consider this a configuration error if running
an unattended install, and prevent it completely if running an attended
install. (continues to prompt for hosts until you have at least 3)
Because this condition cannot be entered in the interactive install, we can't
really write a test for this negative case.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
We're asserting the same things in loading facts over and over, which is not
what these tests are really intended to catch. This behavior is tested
elsewhere.
|
|
Checking behavior when there is no LB specified, and when the user attempts
to re-use a master or node as their LB.
|
|
|
|
Code was present to catch this in unattended installs but was looking for a
host record with both master/node and master_lb set to true, but in the
attended installs we were adding a separate host record with the same
connect_to.
Attended tests can now optionally specify multiple "attempted" strings for the
master_lb specification, we'll try to input each if multiple are specified.
Cleanup some empty defaults and error messages as well.
|
|
When the masters are the only nodes in play, we need to explicitly set
schedulable to True due to logic in openshift_facts.py which assumes that if
the node is also a master, schedulable should be false.
|
|
Improve scaleup playbook
|
|
HA support for atomic-openshift-installer
|
|
harness
|
|
If the only Nodes we have are also on Masters we set the scheduleable.
|
|
|
|
configuration file is missing.
|
|
|
|
Also:
* minor rewording of the text that informs the admin about scheduleable
masters.
|
|
|
|
|
|
The preconfigured load balancers, previously denoted by having 'run_on' set to
false, cannot have their facts gathered which results in a stack trace. Later
when we write out the inventory we have to fake out the hostname and just use
'connect_to'.
We're likely going to have the concept of other types of "plug-in" hosts where
we don't run ansible. We should make sure we abstract this properly so it's
easy to add additional types of hosts.
Also in the commit:
- Renamed 'run_on' to 'preconfigured' and inverted the logic as needed
- Output tally of Masters and Nodes as well as remaining Masters required for
HA
- Minor rewording in a few places
- Currently only prompting for the load balancer after all other hosts have
been entered
- Removed spurious echo
|
|
Rename ha_proxy variables and methods to 'master_lb' to better
future-proof things.
|
|
Reverse the order we ask two questions: What variant the user wants
to install and which hosts to install on. This lets us avoid asking
for multiple masters for 3.0 installs.
|
|
|
|
|
|
|
|
|
|
This adds the ability to quickly set up a multi-master environment.
|
|
atomic-openshift-installer: connect_to error handling
|
|
|
