/manifest-bouncer

Primary LanguagePythonApache License 2.0Apache-2.0

manifest-bouncer

A tool to performs checks on kubernetes/openshift manifests.

Usage

usage: manifest-bouncer [-h] [-v] [--warn-only] [--enable-all] [--disable-all]
                        [--enable-limits] [--enable-requests]
                        MANIFEST

Run checks on k8s/openshift manifests.

positional arguments:
  MANIFEST              manifest to check (YAML or JSON). Use '-' to read from
                        STDIN

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         print full report
  --warn-only           do not return an error if the checks fail
  --enable-all          run all the checks. To disable a specific check, use
                        the `--disable-<check>` form.
  --disable-all         Don't run any checks. To enable a specific check, use
                        the `--enable-<check>` form.
  --enable-limits, --disable-limits
                        check that limits are defined (Default: ENABLED)
  --enable-requests, --disable-requests
                        check that requests are defined (Default: DISABLED)

If all the tests are successful it will exit with 0. Otherwise it will return with 1 if one or more checks fails.

If the manifest is of kind: List, this tool will perform the checks on the list items.

Adding checks

Checks can be added by simply contributing a new class to the checks directory. The added class must:

  • The class name must start with Check, typically: Check<CheckName>
  • It must inherit from CheckBase (exported from lib.base): class CheckMyCheck(CheckBase).
  • If it should be passed only on a specific template kinds then it's possible to define a class level attribute whitelist that lists all matching kinds.
  • Within the class, each method that begins with check_ will be run as a check. Checks must raise an AssertionError if they don't pass, other exceptions will be treated as runtime errors. These methods must have the following signature: (self, m), where m is the manifest.
  • The check_ methods can be further filtered for specific kinds using the @whitelist(*kinds) decorator.
  • Note that you can use the @whitelist decorator without the class variable whitelist (and viceversa).
  • Every class that implements a check should have an enable_parameter class varoabñe that will generate two command line options: --enable-<checkname> and --disable-checkname>. It also requires a description class variable.

Example:

from lib.base import CheckBase


class CheckMyCheck(CheckBase):
    whitelist = ["DeploymentConfig", "StatefulSet", "ReplicaSet"]
    enable_parameter = "mycheck"
    help = "check if mycheck passes"

    def check_foo(self, m):
        assert False, "this always fails" # will raise a check failure

    @whitelist("StatefulSet", "ReplicaSet") # this will be run only on the listed kinds
    def check_bar(self, m):
        assert True

Since the enable_parameter is my-check it will generate two new options: --enable-mycheck and --disable-mycheck (useful along with --enable-all).

Installation

Create and enter the virtualenv environment:

python3 -m venv venv
source venv/bin/activate

# make sure you are running the latest setuptools
pip install --upgrade pip setuptools

Install the package:

python setup.py install

# or alternatively use this for a devel environment
python setup.py develop

Requirements

Please see setup.py.

Future work

  • Readiness and Liveness Probe
  • Not using the latest tag for container images
  • Sequencing of objects (A Role should be listed before RoleBinding)
  • All containers should be named
  • All ports should be named
  • Container name must be reported in the report
  • CheckLimits and CheckRequests should support an extra argument that ensures the values are within those boundaries

Licence

Apache License Version 2.0.

Authors

These tools have been written by the Red Hat App-SRE Team.