2
0
mirror of https://github.com/fork-maintainers/iceraven-browser synced 2024-11-03 23:15:31 +00:00
iceraven-browser/taskcluster/fenix_taskgraph/release_promotion.py
2022-03-21 18:38:56 +00:00

203 lines
7.4 KiB
Python

# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
import os
from mozilla_version.fenix import FenixVersion
from taskgraph.actions.registry import register_callback_action
from taskgraph.util.taskcluster import get_artifact
from taskgraph.taskgraph import TaskGraph
from taskgraph.decision import taskgraph_decision
from taskgraph.parameters import Parameters
from taskgraph.util.taskgraph import (
find_decision_task,
find_existing_tasks_from_previous_kinds,
)
RELEASE_PROMOTION_PROJECTS = (
"https://github.com/mozilla-mobile/fenix",
"https://github.com/mozilla-releng/staging-fenix",
)
def is_release_promotion_available(parameters):
return parameters["head_repository"] in RELEASE_PROMOTION_PROJECTS
@register_callback_action(
name="release-promotion",
title="Ship Fenix",
symbol="${input.release_promotion_flavor}",
description="Ship Fenix",
generic=False,
order=500,
context=[],
available=is_release_promotion_available,
schema=lambda graph_config: {
"type": "object",
"properties": {
"build_number": {
"type": "integer",
"default": 1,
"minimum": 1,
"title": "The release build number",
"description": (
"The release build number. Starts at 1 per "
"release version, and increments on rebuild."
),
},
"do_not_optimize": {
"type": "array",
"description": (
"Optional: a list of labels to avoid optimizing out "
"of the graph (to force a rerun of, say, "
"funsize docker-image tasks)."
),
"items": {
"type": "string",
},
},
"revision": {
"type": "string",
"title": "Optional: revision to ship",
"description": ("Optional: the revision to ship."),
},
"release_promotion_flavor": {
"type": "string",
"description": "The flavor of release promotion to perform.",
"default": "build",
"enum": sorted(graph_config["release-promotion"]["flavors"].keys()),
},
"rebuild_kinds": {
"type": "array",
"description": (
"Optional: an array of kinds to ignore from the previous "
"graph(s)."
),
"items": {
"type": "string",
},
},
"previous_graph_ids": {
"type": "array",
"description": (
"Optional: an array of taskIds of decision or action "
"tasks from the previous graph(s) to use to populate "
"our `previous_graph_kinds`."
),
"items": {
"type": "string",
},
},
"version": {
"type": "string",
"description": (
"Optional: override the version for release promotion. "
"Occasionally we'll land a taskgraph fix in a later "
"commit, but want to act on a build from a previous "
"commit. If a version bump has landed in the meantime, "
"relying on the in-tree version will break things."
),
"default": "",
},
"next_version": {
"type": "string",
"description": "Next version.",
"default": "",
},
},
"required": [
"release_promotion_flavor",
"version",
"build_number",
"next_version",
],
},
)
def release_promotion_action(parameters, graph_config, input, task_group_id, task_id):
release_promotion_flavor = input["release_promotion_flavor"]
promotion_config = graph_config["release-promotion"]["flavors"][
release_promotion_flavor
]
target_tasks_method = promotion_config["target-tasks-method"].format(
project=parameters["project"]
)
rebuild_kinds = input.get("rebuild_kinds") or promotion_config.get(
"rebuild-kinds", []
)
do_not_optimize = input.get("do_not_optimize") or promotion_config.get(
"do-not-optimize", []
)
# make parameters read-write
parameters = dict(parameters)
# Build previous_graph_ids from ``previous_graph_ids`` or ``revision``.
previous_graph_ids = input.get("previous_graph_ids")
if not previous_graph_ids:
previous_graph_ids = [find_decision_task(parameters, graph_config)]
# Download parameters from the first decision task
parameters = get_artifact(previous_graph_ids[0], "public/parameters.yml")
# Download and combine full task graphs from each of the previous_graph_ids.
# Sometimes previous relpro action tasks will add tasks, like partials,
# that didn't exist in the first full_task_graph, so combining them is
# important. The rightmost graph should take precedence in the case of
# conflicts.
combined_full_task_graph = {}
for graph_id in previous_graph_ids:
full_task_graph = get_artifact(graph_id, "public/full-task-graph.json")
combined_full_task_graph.update(full_task_graph)
_, combined_full_task_graph = TaskGraph.from_json(combined_full_task_graph)
parameters["existing_tasks"] = find_existing_tasks_from_previous_kinds(
combined_full_task_graph, previous_graph_ids, rebuild_kinds
)
parameters["do_not_optimize"] = do_not_optimize
parameters["target_tasks_method"] = target_tasks_method
parameters["build_number"] = int(input["build_number"])
# When doing staging releases on try, we still want to re-use tasks from
# previous graphs.
parameters["optimize_target_tasks"] = True
parameters["shipping_phase"] = input["release_promotion_flavor"]
version_in_file = read_version_file()
parameters["version"] = (
input["version"] if input.get("version") else read_version_file()
)
version_string = parameters["version"]
if version_string != version_in_file:
raise ValueError(
"Version given in tag ({}) does not match the one in version.txt ({})".format(
version_string, version_in_file
)
)
parameters["head_tag"] = "v{}".format(version_string)
parameters["next_version"] = input["next_version"]
version = FenixVersion.parse(version_string)
if version.is_beta:
release_type = "beta"
elif version.is_release:
release_type = "release"
elif version.is_release_candidate:
release_type = "release"
else:
raise ValueError("Unsupported version type: {}".format(version.version_type))
parameters["release_type"] = release_type
parameters["tasks_for"] = "action"
parameters["pull_request_number"] = None
# make parameters read-only
parameters = Parameters(**parameters)
taskgraph_decision({"root": graph_config.root_dir}, parameters=parameters)
def read_version_file():
with open(os.path.join(os.path.dirname(__file__), "..", "..", "version.txt")) as f:
return f.read().strip()