More flexible merging of environments

https://blueprints.launchpad.net/heat/+spec/environment-merging

Since we now support server side merging of environment files, which is good but it only supports the same “last one wins” merge strategy that we previously had in heatclient.

In some situations more flexibility is required, e.g when composing a deployment via multiple environment files where parameter key collisions will occur.

Problem description

Consider this scenario

heat stack-create foo -f foo.yaml -e one.yaml -e two.yaml

one.yaml contains

parameters:
ControllerServices:
  • Keystone

two.yaml contains

parameters:
ControllerServices:
  • Glance

With the current environment merging I will always get:

parameters:
ControllerServices:
  • Glance

But what I actually want is:

parameters:
ControllerServices:
  • Keystone

  • Glance

So, I need some way to specify that the heat server-side environment merging will append to, rather than overwrite the ControllerServices parameter. (Same problem exists for e.g json map parameters, where we probably want to offer the option of shallow and deep merge vs just overwriting).

Note this is the exact same problem faced (and fixed) by some other tools that accept potentially overlapping sections of yaml data, such as cloud-init[1]

[1] http://cloudinit.readthedocs.io/en/latest/topics/merging.html

Proposed change

Since parameters/parameter_defaults are simple key/value pairs it’s hard to come up with an interface that adds merge strategy data within the parameters map, so we’d probably need a separate environment section, e.g:

  1. Specify global merge strategy for all parameters:

merge_strategy:

list: extend dict: merge # we could support “merge” and “deep_merge” here? string: append

  1. Specify per-parameter merge strategy

merge_strategy:
parameters:

ControllerServices: extend

This is pretty similar to how the cloud-init “merge_how” directive works.

If multiple environments specified conflicting merge_strategy values, we’d raise a validation error.

Alternatives

The alternative is basically client-side munging of templates, which would work, but makes the templates much less portable (e.g we’ll end up implementing a TripleO specific solution to this in tripleo-common, which will result in our environments not working direct to heat via heatclient)

Implementation

Since jdob added support for merging a list of environment_files in engine/service.py, this should just be a case of adding support for these optional strategies to the _merge_environments function in that file.

Assignee(s)

Primary assignee:

ramishra

Milestones

Target Milestone for completion:

newton-2

Work Items

  • Update heat-engine code to support new optional merge_strategy key

  • Comprehensive tests & docs for this new interface

  • Add support to heatclient (not to support the merging, just to accept the key)

Dependencies

None