The tc-admin
library supports administration of the runtime configuration of a Taskcluster deployment.
This means creation and maintenance of resources such as roles, hooks, and worker pools, based on version-controlled specifications..
This library is used as a dependency of a Python application containing code and configuration specific to the taskcluster deployment(s) being administered.
The project should contain a tc-admin.py
that serves to define the application configuration.
Assuming that is in place, the tool is easy to use:
After installing the app, run tc-admin generate
to generate the expected set of resources (use --json
to get JSON output).
This will require TASKCLUSTER_ROOT_URL
to be set in the environment, to know which deployment to talk to.
Similarly, tc-admin current
will generate the current set of resources (optionally with --json
).
To compare them, run tc-admin diff
.
If the configuration includes secrets, you may want to pass the --without-secrets
option.
This option skips managing the content of secrets, and thus needs neither access to secret values nor Taskcluster credentials to fetch secrets.
Run tc-admin apply
to apply the changes.
Note that only apply
will require Taskcluster credentials, and it's a good practice to only set TC credentials when running this command.
See tc-admin <command> --help
for more useful options.
Checks are a way to double-check that purpose-specific invariants are satisfied in a Taskcluster deployment. For example, it may be important to check that only specific repository roles have scopes to create tasks in critical worker pools. Checks are defined as normal Python tests, and have access to the current and generated configurations.
If the app has checks set up, then tc-admin check
will run those checks.
The operation of this tool is pretty simple: it generates a set of expected Taskcluster resources, downloads existing resources from the Taskcluster API, and compares them. A collection of resources also specifies the set of "managed" resources -- this allows deletion of resources that are no longer expected, without risk of deleting everything in the Taskcluster API.
After generation, resources can be "modified". This is typically used to make minor changes to resources depending on environment. For exmaple, in a staging environment, hook schedules might be removed.
A tc-admin app is configured by tc-admin.py
.
This is a Python file which is responsible for creating and customizing an AppConfig
object.
from tcadmin.appconfig import AppConfig
appconfig = AppConfig()
# .. customize
The tc-admin
command looks for tc-admin.py
in the current directory, or in the directory given by $TC_ADMIN_PY
or command-line argument --tc-admin-py
.
Like most Python modules, the global __file__
is set when tc-admin.py
is executed, and can be used to determine relative paths.
Before tc-admin.py
is executed, the current working directory is changed to the directory containing it.
This enables relative imports as well as loading files with relative paths (such as with LocalLoader, below).
This library can also be used programmatically.
Simply create an AppConfig object directly and call tcadmin.main.main
with it:
from tcadmin.appconfig import AppConfig
from tcadmin.main import main
def boot():
appconfig = AppConfig()
# .. customize
main(appconfig)
if __name__ == "__main__":
boot()
Note that the current directory is not automatically set in this case.
The AppConfig object contains a number of properties that can be customized, described below.
During execution, the current AppConfig object is available from AppConfig.current()
.
This can be useful when generators or modifiers are defined in separate Python modules.
Generators generate expected Taskcluster resources and defined the managed resource names.
Each generator is an async function that is given a Resources
object and is expected to call resources.manage
and resources.update
.
Generators are registered with appconfig.generators.register
, most easily with a decorator:
@appconfig.generators.register
async def update_resources(resources):
# modify in place ...
When generating secrets, respect the --with-secrets
option, and generate secrets without values when it is false.
You can also use this option to determine whether the generation process requires access to secret values.
This allows generation runs with --without-secrets
to occur without any credentials or access to secret values.
@appconfig.generators.register
async def update_resources(resources):
# modify in place ...
if appconfig.options.get('--with-secrets'):
secretstore = load_secret_values()
resources.add(Secret(
name="top-secret/cookie-recipe",
secret=secretstore.decrypt('recipes/double-chocolate-chip')))
else:
resources.add(Secret(name="top-secret/cookie-recipe"))
Modifiers are responsible for modifying an existing set of resources. Since resources are immutable, the signature differs slightly from generators:
@appconfig.modifiers.register
async def modify_resources(resources):
# return new set of resources
return resources.map(..)
Modifiers are called sequentially in the order in which they were registered.
Callbacks are external function from your own application that can be executed at specific times during the tc-admin apply
execution:
- A
before_apply
callback will run before a resource is created, updated or deleted, - A
after_apply
callback will run after a resource has beencreated, updated or deleted.
Supported actions are :
- create
- update
- delete
By default all actions are used.
All resources are supported by callbacks, and enabled by default. If you want to limit your callback to some resources, you need to specify them using their class (not a string).
You can declare your callbacks as:
from tcadmin.resources import Secret
async def my_action(action, resource):
print("Got a callback on", action, resource)
# Will call your function when a secret has been updated or deleted
appconfig.callbacks.add("after_apply", my_action, actions=["update", "delete"], resources=[Secret, ])
Apps can add additional command-line options, the values of which are then available during resource generation.
To register an option, call appconfig.options.add
, with the full option name and any of the following keyword options:
required
- if True, the option is requiredhelp
- help string to be shown intc-admin generate --help
default
- default value for the option
To retrieve the option value during generation, call appconfig.options.get(name)
.
All together, then:
appconfig.options.add("--branch", help="configuration branch to read from")
@appconfig.generators.register
async def update_resources(resources):
branch = appconfig.options.get("--branch")
# ...
As a special case, the --with-secrets
secret is available through this same mechanism.
The appconfig.check_path
property gives the path of the checks to run for tc-admin check
, relative to the current directory.
This directory is a "normal" pytest directory.
To help distinguish checks from tests, include a pytest.ini
in this directory:
[pytest]
python_classes = Check*
python_files = check_*.py
python_functions = check_*
The appconfig.description_prefix
property allows the users to customize the prefix of the description.
This can be customized in the tc-admin.py
as follows:
from tcadmin.appconfig import AppConfig
appconfig = AppConfig()
appconfig.description_prefix = "YOUR_CUSTOM_PREFIX"
The DEFAULT value of the description_prefix is *DO NOT EDIT* - This resource is configured automatically.\n\n
For the common case of a configuration that applies to only one Taskcluster deployment, specify that deployment's root URL in tc-admin.py
:
from tcadmin.appconfig import AppConfig
appconfig = AppConfig()
appconfig.root_url = "https://taskcluster.example.com"
To support more complex cases, this value can also be an async callable.
It will be invoked once, after the click
options have been processed, so it can access appconfig.options
if necessary.
The current root URL is available from an async helper function:
from tcadmin.util.root_url import root_url
async def foo():
print(await root_url())
This will retrieve the value from the AppConfig or, if that is not set, from TASKCLUSTER_ROOT_URL
.
If both are set, and the values do not match, it will produce an error message.
Most uses of this library load configuration data from some easily-modified YAML files.
The tcadmin.util.config
package provides some support for loading and parsing these files.
All of this is entirely optional; use what is appropriate to the purpose.
First, define a loader that can load data from files.
from tcadmin.util.config import LocalLoader
loader = LocalLoader()
The LocalLoader class knows how to load configuration from files relative to tc-admin.py
.
It has a load
method that will load data, optionally parsing it as YAML:
data = loader.load("data.bin")
aliases = await loader.load("aliases.yml", parse="yaml")
You can also define your own loader class.
Just implement the load_raw
method to return bytes, given a filename.
YAML data is inconvenient to deal with in Python, introducig a lot of [".."]
noise.
Commonly, config files are either a top-level array, or a top-level object with named "stanzas" of configuration.
The ConfigList and ConfigDict classes support these formats.
We suggest using these with the Python attrs
library.
Define a class that inherits from either of these classes, specifies the filename to load from, and has an Item
class for the items in the collection:
from tcadmin.util.config import ConfigList
class Workers(ConfigList):
filename = "workers.yml"
@attr.s
class Item:
workerId = attr.ib(type=str)
bigness = attr.ib(type=int, default=1)
Then simply call await Workers.load(loader)
to load a workers.yml
that looks something like
- workerId: small
bigness: 5
- workerId: huge
bigness: 5000
The ConfigDict class is similar, but parses files like
small:
bigness: 5
huge:
bigness: 5000
ConfigList creates new Item
instances from array elements item
with Item(**item)
.
ConfigDict creates new Item
instances from k: item
with Item(k, **item)
.
This approach is compatible with attrs
, where in the latter case k
should be the first attribute defined.
If array elements or object values are not themselves YAML objects, add a class method named transform_item
to transform the data in the YAML file into a Python dictionary.
For example:
class Workers(ConfigList):
@classmethod
def transform_item(cls, item):
# given a simple string, assume that is the workerId and apply defaults
if isinstance(item, str):
return {"workerId": item}
return item
...
The tcadmin.resources
package contains clasess for defining Taskcluster resources and collections.
from tcadmin.resources import Resources
The Resources
class defines a collection of resources and tracks what resources are managed.
Resources found in the Taskcluster deployment that match the "managed" patterns but are not generated will be deleted on apply
.
The class has the following methods:
resources.add(resource)
- add a resource to the collection. The resource must be managed.resources.update(iterable)
- add an iterable full of resources to the collection. All resources must be managed.resources.manage(pattern)
- consider reources matching regular expression stringpattern
to be managedresources.is_managed(id)
- return true if the given resource is managedresources.filter(pattern)
- return a new Resources object containing only resources matching the given regular expression stringresources.map(functor)
- return a new Resources object, with fuctor applied to each resource. This is typically used in modifiers.
Resources must be unique -- tc-admin cannot manage multiple hooks with the same name, for example. However, some resource kinds support merging, where adding a resource with the same identity as one that already exists "merges" it into the existing resource. See the description of roles, below.
The remaining classes represent individual resources.
Each has an id
formed from its kind and the unique identifier for the resource in the Taskcluster.
For example, Hook=release-hooks/beta-release
.
Resources are immutable once created, but can be "evolved" (returning a new resource) with rsrc.evolve(**updates)
.
Resources with descriptions automatically prepend a "DO NOT EDIT" prefix to dissuade users from editing them in the Taskcluster UI.
from tcadmin.resources import Hook, Binding
hook = Hook(
hookGroupId=..,
hookId=..,
name=..,
description=..,
owner=..,
emailOnError=..,
schedule=(.., ..),
bindings=(.., ..),
task={..},
triggerSchema={..})
Most of these fields correspond directly to the Taskcluster definition.
Both schedule
and bindings
must be tuples, not lists (as lists are mutable).
The items in schedule
are cron-like strings.
The items in bindings
are instances of Binding(exchange=.., routingKeyPattern=..)
.
from tcadmin.resources import Secret
secret = Secret(
name=..,
secret=..)
# or, when not managing secret values
secret = Secret(name=..)
Secrets are managed using the Secret resource type. While Taskcluster supports expiration times on secrets, this library sets those times the far future, effectively creating non-expiring secrets
This library is careful to not display secret values in its output.
Instead, it displays <unknown>
when not managing secret values, and displays a salted hash of the secret value when managing secret values.
The salted hash allows tc-admin diff
to show that a secret value has changed, without revealing the value of that secret.
The salt includes a per-run salt, and the name of the secret, with the result that even if two secrets have the same value, they will be shown with different hashes in tc-admin generate
.
from tcadmin.resources import Role
role = Role(
roleId=..,
description=..,
scopes=(.., ..))
As with hooks, scopes
must be a tuple (not a list) of strings.
Roles can be merged if their descriptions match. The resulting role contains the union of the scopes of the input roles. This functionality makes management of roles easier in cases where different parts of the generation process may add scopes to the same role.
For example:
resources.add(Role(roleId="my-role", description="My Role", scopes=["scope1"]))
resources.add(Role(roleId="my-role", description="My Role", scopes=["scope2"]))
This will result in a single Role with scopes ["scope1", "scope2"]
.
from tcadmin.resources import Client
client = Client(
clientId=..,
description=..,
scopes=(.., ..))
Clients work much like roles.
As with roles, scopes
must be a tuple (not a list) of strings.
This library does not manage access tokens: it discards them from the response to auth.createClient
.
The expectation is that project admins who need credentials for the managed clients will call auth.resetAccessToken
and use the returned token.
Clients configured by this library have an expiration date far in the future. Like roles, the clients managed here last "forever".
from tcadmin.resources import WorkerPool
workerPool = WorkerPool(
workerPoolId=..,
providerId=..,
description=..,
owner=..,
config={..},
emailOnError=..)
All attributes of this class match the Taskcluster definition.
The library provides a number of utilities for common application requirements.
NOTE: only functions described in this README are considered stable. Other functions defined by the library may change without notice.
As an aid to writing checks, tc-admin supplies local implementations of various scope-related algorithms.
from tcadmin.util.scopes import satisfies, normalizeScopes, Resolver
The satisfies
function determines scope satisfaction, without any expansion.
Satisfaction means that the first argument contains all scopes in the second argument.
assert satisfies(['balloons:*', 'cake:birthday'], ['baloons:mylar:happy-birthday'])
The normalizeScopes
function normalizes a scopeset, removing redundant scopes and sorting.
assert normalizedScopes(['balloons:*', 'balloons:mylar:*']) == ['baloons:*']
Finally, Resolver
can perform scope expansion.
It is initialized with a dictionary mapping roleIds to scope lists.
Alternately, it can be initialized from a Resources
instance using Resolver.from_resources(resources)
.
Its expandScopes
method behaves identically to the remote call auth.expandScopes
.
resolver = Resolver.from_resources(resources)
assert resolver.expandScopes(['assume:clown:grimaldi']) == ['assume:clown:grimaldi', 'ruffle:full']
The library uses aiohttp
to communicate with Taskcluster, and establishes a single session for efficiency.
Applications can use the same session for any other HTTP operations.
from tcadmin.util.session import aiohttp_session
async def foo():
# ...
async with aiohttp_session().get(url) as response:
response.raise_for_status()
result = await response.read()
Tests and checks can set this value using with_aiohttp_session
:
from tcadmin.util.sessions import with_aiohttp_session
import pytest
@pytest.mark.asyncio
@with_aiohttp_session
async def test_something():
# ...
A MatchList is a list of regular expressions that can determine whether a given
string matches one of those patterns. Patterns are rooted at the left, but
should use $
where required to match the end of the string.
from tcadmin.util.matchlist import MatchList
ml = MatchList()
ml.add("fo+$")
ml.add("ba+r$")
assert ml.matches("foo")
This functionality is used to track managed resources, but may be useful otherwise.
To install for development, in a virtualenv:
pip install -e [path]
And to run flake8 and tests:
python setup.py flake8
python setup.py test
The library uses Black to format code.
pip install black
black tcadmin
To release:
- update version in
setup.py
andgit commit -m "vX.Y.Z"
git tag vX.Y.Z
git push
git push --tags
./release.sh --real
and enter your pypi credentials when prompted (omit the--real
to try it against the testing pypi, if you're not sure)- go to https://github.com/taskcluster/tc-admin/releases/tag/vX.Y.Z and create a new release from the tag with a brief description of the changes
- build and publish a new version of the linux/amd64 tc-admin docker image:
pass git pull
pass hub.docker.com/taskclusterbot # fetch credentials for making docker release
docker logout
docker login # use credentials output above
docker build -t taskcluster/tc-admin:X.Y.Z .
docker push taskcluster/tc-admin:X.Y.Z