summaryrefslogtreecommitdiffstats
path: root/taskcluster/gecko_taskgraph/target_tasks.py
diff options
context:
space:
mode:
Diffstat (limited to 'taskcluster/gecko_taskgraph/target_tasks.py')
-rw-r--r--taskcluster/gecko_taskgraph/target_tasks.py1497
1 files changed, 1497 insertions, 0 deletions
diff --git a/taskcluster/gecko_taskgraph/target_tasks.py b/taskcluster/gecko_taskgraph/target_tasks.py
new file mode 100644
index 0000000000..90ebd479ad
--- /dev/null
+++ b/taskcluster/gecko_taskgraph/target_tasks.py
@@ -0,0 +1,1497 @@
+# 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 itertools
+import os
+import re
+from datetime import datetime, timedelta
+
+from redo import retry
+from taskgraph.parameters import Parameters
+from taskgraph.target_tasks import _target_task, get_method
+from taskgraph.util.taskcluster import find_task_id
+
+from gecko_taskgraph import GECKO, try_option_syntax
+from gecko_taskgraph.util.attributes import (
+ match_run_on_hg_branches,
+ match_run_on_projects,
+)
+from gecko_taskgraph.util.hg import find_hg_revision_push_info, get_hg_commit_message
+from gecko_taskgraph.util.platforms import platform_family
+
+# Some tasks show up in the target task set, but are possibly special cases,
+# uncommon tasks, or tasks running against limited hardware set that they
+# should only be selectable with --full.
+UNCOMMON_TRY_TASK_LABELS = [
+ # Platforms and/or Build types
+ r"build-.*-gcp", # Bug 1631990
+ r"mingwclang", # Bug 1631990
+ r"valgrind", # Bug 1631990
+ # Android tasks
+ r"android-geckoview-docs",
+ r"android-hw",
+ # Windows tasks
+ r"windows10-64-ref-hw",
+ r"windows10-aarch64-qr",
+ # Linux tasks
+ r"linux-", # hide all linux32 tasks by default - bug 1599197
+ r"linux1804-32", # hide linux32 tests - bug 1599197
+ # Test tasks
+ r"web-platform-tests.*backlog", # hide wpt jobs that are not implemented yet - bug 1572820
+ r"-ccov",
+ r"-profiling-", # talos/raptor profiling jobs are run too often
+ # Hide shippable versions of tests we have opt versions of because the non-shippable
+ # versions are faster to run. This is mostly perf tests.
+ r"-shippable(?!.*(awsy|browsertime|marionette-headless|mochitest-devtools-chrome-fis|raptor|talos|web-platform-tests-wdspec-headless|mochitest-plain-headless))", # noqa - too long
+]
+
+
+def index_exists(index_path, reason=""):
+ print(f"Looking for existing index {index_path} {reason}...")
+ try:
+ task_id = find_task_id(index_path)
+ print(f"Index {index_path} exists: taskId {task_id}")
+ return True
+ except KeyError:
+ print(f"Index {index_path} doesn't exist.")
+ return False
+
+
+def filter_out_shipping_phase(task, parameters):
+ return (
+ # nightly still here because of geckodriver
+ not task.attributes.get("nightly")
+ and task.attributes.get("shipping_phase") in (None, "build")
+ )
+
+
+def filter_out_devedition(task, parameters):
+ return not task.attributes.get("shipping_product") == "devedition"
+
+
+def filter_out_cron(task, parameters):
+ """
+ Filter out tasks that run via cron.
+ """
+ return not task.attributes.get("cron")
+
+
+def filter_for_project(task, parameters):
+ """Filter tasks by project. Optionally enable nightlies."""
+ run_on_projects = set(task.attributes.get("run_on_projects", []))
+ return match_run_on_projects(parameters["project"], run_on_projects)
+
+
+def filter_for_hg_branch(task, parameters):
+ """Filter tasks by hg branch.
+ If `run_on_hg_branch` is not defined, then task runs on all branches"""
+ run_on_hg_branches = set(task.attributes.get("run_on_hg_branches", ["all"]))
+ return match_run_on_hg_branches(parameters["hg_branch"], run_on_hg_branches)
+
+
+def filter_on_platforms(task, platforms):
+ """Filter tasks on the given platform"""
+ platform = task.attributes.get("build_platform")
+ return platform in platforms
+
+
+def filter_by_uncommon_try_tasks(task, optional_filters=None):
+ """Filters tasks that should not be commonly run on try.
+
+ Args:
+ task (str): String representing the task name.
+ optional_filters (list, optional):
+ Additional filters to apply to task filtering.
+
+ Returns:
+ (Boolean): True if task does not match any known filters.
+ False otherwise.
+ """
+ filters = UNCOMMON_TRY_TASK_LABELS
+ if optional_filters:
+ filters = itertools.chain(filters, optional_filters)
+
+ return not any(re.search(pattern, task) for pattern in filters)
+
+
+def filter_by_regex(task_label, regexes, mode="include"):
+ """Filters tasks according to a list of pre-compiled reguar expressions.
+
+ If mode is "include", a task label must match any regex to pass.
+ If it is "exclude", a task label must _not_ match any regex to pass.
+ """
+ if not regexes:
+ return True
+
+ assert mode in ["include", "exclude"]
+
+ any_match = any(r.search(task_label) for r in regexes)
+ if any_match:
+ return mode == "include"
+ return mode != "include"
+
+
+def filter_release_tasks(task, parameters):
+ platform = task.attributes.get("build_platform")
+ if platform in (
+ "linux",
+ "linux64",
+ "macosx64",
+ "win32",
+ "win64",
+ "win64-aarch64",
+ ):
+ if task.attributes["kind"] == "l10n":
+ # This is on-change l10n
+ return True
+ if (
+ task.attributes["build_type"] == "opt"
+ and task.attributes.get("unittest_suite") != "talos"
+ and task.attributes.get("unittest_suite") != "raptor"
+ ):
+ return False
+
+ if task.attributes.get("shipping_phase") not in (None, "build"):
+ return False
+
+ """ No debug on release, keep on ESR with 4 week cycles, release
+ will not be too different from central, but ESR will live for a long time.
+
+ From June 2019 -> June 2020, we found 1 unique regression on ESR debug
+ and 5 unique regressions on beta/release. Keeping spidermonkey and linux
+ debug finds all but 1 unique regressions (windows found on try) for beta/release.
+
+ ...but debug-only failures started showing up on ESR (esr-91, esr-102) so
+ desktop debug tests were added back for beta.
+ """
+ build_type = task.attributes.get("build_type", "")
+ build_platform = task.attributes.get("build_platform", "")
+ test_platform = task.attributes.get("test_platform", "")
+
+ if parameters["release_type"].startswith("esr") or (
+ parameters["release_type"] == "beta" and "android" not in build_platform
+ ):
+ return True
+
+ # code below here is intended to reduce release debug tasks
+ if task.kind == "hazard" or "toolchain" in build_platform:
+ # keep hazard and toolchain builds around
+ return True
+
+ if build_type == "debug":
+ if "linux" not in build_platform:
+ # filter out windows/mac/android
+ return False
+ if task.kind not in ["spidermonkey"] and "-qr" in test_platform:
+ # filter out linux-qr tests, leave spidermonkey
+ return False
+ if "64" not in build_platform:
+ # filter out linux32 builds
+ return False
+
+ # webrender-android-*-debug doesn't have attributes to find 'debug', using task.label.
+ if task.kind == "webrender" and "debug" in task.label:
+ return False
+ return True
+
+
+def filter_out_missing_signoffs(task, parameters):
+ for signoff in parameters["required_signoffs"]:
+ if signoff not in parameters["signoff_urls"] and signoff in task.attributes.get(
+ "required_signoffs", []
+ ):
+ return False
+ return True
+
+
+def filter_tests_without_manifests(task, parameters):
+ """Remove test tasks that have an empty 'test_manifests' attribute.
+
+ This situation can arise when the test loader (e.g bugbug) decided there
+ weren't any important manifests to run for the given push. We filter tasks
+ out here rather than in the transforms so that the full task graph is still
+ aware that the task exists (which is needed by the backfill action).
+ """
+ if (
+ task.kind == "test"
+ and "test_manifests" in task.attributes
+ and not task.attributes["test_manifests"]
+ ):
+ return False
+ return True
+
+
+def standard_filter(task, parameters):
+ return all(
+ filter_func(task, parameters)
+ for filter_func in (
+ filter_out_cron,
+ filter_for_project,
+ filter_for_hg_branch,
+ filter_tests_without_manifests,
+ )
+ )
+
+
+def accept_raptor_android_build(platform):
+ """Helper function for selecting the correct android raptor builds."""
+ if "android" not in platform:
+ return False
+ if "shippable" not in platform:
+ return False
+ if "p2" in platform and "aarch64" in platform:
+ return False
+ if "p5" in platform and "aarch64" in platform:
+ return False
+ if "g5" in platform:
+ return False
+ if "a51" in platform:
+ return True
+
+
+def filter_unsupported_artifact_builds(task, parameters):
+ try_config = parameters.get("try_task_config", {})
+ if not try_config.get("use-artifact-builds", False):
+ return True
+
+ supports_artifact_builds = task.attributes.get("supports-artifact-builds", True)
+ return supports_artifact_builds
+
+
+def filter_out_shippable(task):
+ return not task.attributes.get("shippable", False)
+
+
+def _try_task_config(full_task_graph, parameters, graph_config):
+ requested_tasks = parameters["try_task_config"]["tasks"]
+ return list(set(requested_tasks) & full_task_graph.graph.nodes)
+
+
+def _try_option_syntax(full_task_graph, parameters, graph_config):
+ """Generate a list of target tasks based on try syntax in
+ parameters['message'] and, for context, the full task graph."""
+ options = try_option_syntax.TryOptionSyntax(
+ parameters, full_task_graph, graph_config
+ )
+ target_tasks_labels = [
+ t.label
+ for t in full_task_graph.tasks.values()
+ if options.task_matches(t)
+ and filter_by_uncommon_try_tasks(t.label)
+ and filter_unsupported_artifact_builds(t, parameters)
+ ]
+
+ attributes = {
+ k: getattr(options, k)
+ for k in [
+ "no_retry",
+ "tag",
+ ]
+ }
+
+ for l in target_tasks_labels:
+ task = full_task_graph[l]
+ if "unittest_suite" in task.attributes:
+ task.attributes["task_duplicates"] = options.trigger_tests
+
+ for l in target_tasks_labels:
+ task = full_task_graph[l]
+ # If the developer wants test jobs to be rebuilt N times we add that value here
+ if options.trigger_tests > 1 and "unittest_suite" in task.attributes:
+ task.attributes["task_duplicates"] = options.trigger_tests
+
+ # If the developer wants test talos jobs to be rebuilt N times we add that value here
+ if (
+ options.talos_trigger_tests > 1
+ and task.attributes.get("unittest_suite") == "talos"
+ ):
+ task.attributes["task_duplicates"] = options.talos_trigger_tests
+
+ # If the developer wants test raptor jobs to be rebuilt N times we add that value here
+ if (
+ options.raptor_trigger_tests
+ and options.raptor_trigger_tests > 1
+ and task.attributes.get("unittest_suite") == "raptor"
+ ):
+ task.attributes["task_duplicates"] = options.raptor_trigger_tests
+
+ task.attributes.update(attributes)
+
+ # Add notifications here as well
+ if options.notifications:
+ for task in full_task_graph:
+ owner = parameters.get("owner")
+ routes = task.task.setdefault("routes", [])
+ if options.notifications == "all":
+ routes.append(f"notify.email.{owner}.on-any")
+ elif options.notifications == "failure":
+ routes.append(f"notify.email.{owner}.on-failed")
+ routes.append(f"notify.email.{owner}.on-exception")
+
+ return target_tasks_labels
+
+
+@_target_task("try_tasks")
+def target_tasks_try(full_task_graph, parameters, graph_config):
+ try_mode = parameters["try_mode"]
+ if try_mode == "try_task_config":
+ return _try_task_config(full_task_graph, parameters, graph_config)
+ if try_mode == "try_option_syntax":
+ return _try_option_syntax(full_task_graph, parameters, graph_config)
+ # With no try mode, we schedule nothing, allowing the user to add tasks
+ # later via treeherder.
+ return []
+
+
+@_target_task("try_select_tasks")
+def target_tasks_try_select(full_task_graph, parameters, graph_config):
+ tasks = target_tasks_try_select_uncommon(full_task_graph, parameters, graph_config)
+ return [l for l in tasks if filter_by_uncommon_try_tasks(l)]
+
+
+@_target_task("try_select_tasks_uncommon")
+def target_tasks_try_select_uncommon(full_task_graph, parameters, graph_config):
+ from gecko_taskgraph.decision import PER_PROJECT_PARAMETERS
+
+ projects = ("autoland", "mozilla-central")
+ if parameters["project"] not in projects:
+ projects = (parameters["project"],)
+
+ tasks = set()
+ for project in projects:
+ params = dict(parameters)
+ params["project"] = project
+ parameters = Parameters(**params)
+
+ try:
+ target_tasks_method = PER_PROJECT_PARAMETERS[project]["target_tasks_method"]
+ except KeyError:
+ target_tasks_method = "default"
+
+ tasks.update(
+ get_method(target_tasks_method)(full_task_graph, parameters, graph_config)
+ )
+
+ return sorted(tasks)
+
+
+@_target_task("try_auto")
+def target_tasks_try_auto(full_task_graph, parameters, graph_config):
+ """Target the tasks which have indicated they should be run on autoland
+ (rather than try) via the `run_on_projects` attributes.
+
+ Should do the same thing as the `default` target tasks method.
+ """
+ params = dict(parameters)
+ params["project"] = "autoland"
+ parameters = Parameters(**params)
+
+ regex_filters = parameters["try_task_config"].get("tasks-regex")
+ include_regexes = exclude_regexes = []
+ if regex_filters:
+ include_regexes = [re.compile(r) for r in regex_filters.get("include", [])]
+ exclude_regexes = [re.compile(r) for r in regex_filters.get("exclude", [])]
+
+ return [
+ l
+ for l, t in full_task_graph.tasks.items()
+ if standard_filter(t, parameters)
+ and filter_out_shipping_phase(t, parameters)
+ and filter_out_devedition(t, parameters)
+ and filter_by_uncommon_try_tasks(t.label)
+ and filter_by_regex(t.label, include_regexes, mode="include")
+ and filter_by_regex(t.label, exclude_regexes, mode="exclude")
+ and filter_unsupported_artifact_builds(t, parameters)
+ and filter_out_shippable(t)
+ ]
+
+
+@_target_task("default")
+def target_tasks_default(full_task_graph, parameters, graph_config):
+ """Target the tasks which have indicated they should be run on this project
+ via the `run_on_projects` attributes."""
+ return [
+ l
+ for l, t in full_task_graph.tasks.items()
+ if standard_filter(t, parameters)
+ and filter_out_shipping_phase(t, parameters)
+ and filter_out_devedition(t, parameters)
+ ]
+
+
+@_target_task("autoland_tasks")
+def target_tasks_autoland(full_task_graph, parameters, graph_config):
+ """In addition to doing the filtering by project that the 'default'
+ filter does, also remove any tests running against shippable builds
+ for non-backstop pushes."""
+ filtered_for_project = target_tasks_default(
+ full_task_graph, parameters, graph_config
+ )
+
+ def filter(task):
+ if task.kind != "test":
+ return True
+
+ if parameters["backstop"]:
+ return True
+
+ build_type = task.attributes.get("build_type")
+
+ if not build_type or build_type != "opt" or filter_out_shippable(task):
+ return True
+
+ return False
+
+ return [l for l in filtered_for_project if filter(full_task_graph[l])]
+
+
+@_target_task("mozilla_central_tasks")
+def target_tasks_mozilla_central(full_task_graph, parameters, graph_config):
+ """In addition to doing the filtering by project that the 'default'
+ filter does, also remove any tests running against regular (aka not shippable,
+ asan, etc.) opt builds."""
+ filtered_for_project = target_tasks_default(
+ full_task_graph, parameters, graph_config
+ )
+
+ def filter(task):
+ if task.kind != "test":
+ return True
+
+ build_platform = task.attributes.get("build_platform")
+ build_type = task.attributes.get("build_type")
+ shippable = task.attributes.get("shippable", False)
+
+ if not build_platform or not build_type:
+ return True
+
+ family = platform_family(build_platform)
+ # We need to know whether this test is against a "regular" opt build
+ # (which is to say, not shippable, asan, tsan, or any other opt build
+ # with other properties). There's no positive test for this, so we have to
+ # do it somewhat hackily. Android doesn't have variants other than shippable
+ # so it is pretty straightforward to check for. Other platforms have many
+ # variants, but none of the regular opt builds we're looking for have a "-"
+ # in their platform name, so this works (for now).
+ is_regular_opt = (
+ family == "android" and not shippable
+ ) or "-" not in build_platform
+
+ if build_type != "opt" or not is_regular_opt:
+ return True
+
+ return False
+
+ return [l for l in filtered_for_project if filter(full_task_graph[l])]
+
+
+@_target_task("graphics_tasks")
+def target_tasks_graphics(full_task_graph, parameters, graph_config):
+ """In addition to doing the filtering by project that the 'default'
+ filter does, also remove artifact builds because we have csets on
+ the graphics branch that aren't on the candidate branches of artifact
+ builds"""
+ filtered_for_project = target_tasks_default(
+ full_task_graph, parameters, graph_config
+ )
+
+ def filter(task):
+ if task.attributes["kind"] == "artifact-build":
+ return False
+ return True
+
+ return [l for l in filtered_for_project if filter(full_task_graph[l])]
+
+
+@_target_task("mozilla_beta_tasks")
+def target_tasks_mozilla_beta(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a promotable beta or release build
+ of desktop, plus android CI. The candidates build process involves a pipeline
+ of builds and signing, but does not include beetmover or balrog jobs."""
+
+ return [
+ l
+ for l, t in full_task_graph.tasks.items()
+ if filter_release_tasks(t, parameters) and standard_filter(t, parameters)
+ ]
+
+
+@_target_task("mozilla_release_tasks")
+def target_tasks_mozilla_release(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a promotable beta or release build
+ of desktop, plus android CI. The candidates build process involves a pipeline
+ of builds and signing, but does not include beetmover or balrog jobs."""
+
+ return [
+ l
+ for l, t in full_task_graph.tasks.items()
+ if filter_release_tasks(t, parameters) and standard_filter(t, parameters)
+ ]
+
+
+@_target_task("mozilla_esr102_tasks")
+def target_tasks_mozilla_esr102(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a promotable beta or release build
+ of desktop, without android CI. The candidates build process involves a pipeline
+ of builds and signing, but does not include beetmover or balrog jobs."""
+
+ def filter(task):
+ if not filter_release_tasks(task, parameters):
+ return False
+
+ if not standard_filter(task, parameters):
+ return False
+
+ platform = task.attributes.get("build_platform")
+
+ # Android is not built on esr102.
+ if platform and "android" in platform:
+ return False
+
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("mozilla_esr115_tasks")
+def target_tasks_mozilla_esr115(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a promotable beta or release build
+ of desktop, without android CI. The candidates build process involves a pipeline
+ of builds and signing, but does not include beetmover or balrog jobs."""
+
+ def filter(task):
+ if not filter_release_tasks(task, parameters):
+ return False
+
+ if not standard_filter(task, parameters):
+ return False
+
+ platform = task.attributes.get("build_platform")
+
+ # Android is not built on esr115.
+ if platform and "android" in platform:
+ return False
+
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("promote_desktop")
+def target_tasks_promote_desktop(full_task_graph, parameters, graph_config):
+ """Select the superset of tasks required to promote a beta or release build
+ of a desktop product. This should include all non-android
+ mozilla_{beta,release} tasks, plus l10n, beetmover, balrog, etc."""
+
+ def filter(task):
+ # Bug 1758507 - geckoview ships in the promote phase
+ if not parameters["release_type"].startswith("esr") and is_geckoview(
+ task, parameters
+ ):
+ return True
+
+ if task.attributes.get("shipping_product") != parameters["release_product"]:
+ return False
+
+ # 'secondary' balrog/update verify/final verify tasks only run for RCs
+ if parameters.get("release_type") != "release-rc":
+ if "secondary" in task.kind:
+ return False
+
+ if not filter_out_missing_signoffs(task, parameters):
+ return False
+
+ if task.attributes.get("shipping_phase") == "promote":
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+def is_geckoview(task, parameters):
+ return (
+ task.attributes.get("shipping_product") == "fennec"
+ and task.kind in ("beetmover-geckoview", "upload-symbols")
+ and parameters["release_product"] == "firefox"
+ )
+
+
+@_target_task("push_desktop")
+def target_tasks_push_desktop(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to push a build of desktop to cdns.
+ Previous build deps will be optimized out via action task."""
+ filtered_for_candidates = target_tasks_promote_desktop(
+ full_task_graph,
+ parameters,
+ graph_config,
+ )
+
+ def filter(task):
+ if not filter_out_missing_signoffs(task, parameters):
+ return False
+ # Include promotion tasks; these will be optimized out
+ if task.label in filtered_for_candidates:
+ return True
+
+ if (
+ task.attributes.get("shipping_product") == parameters["release_product"]
+ and task.attributes.get("shipping_phase") == "push"
+ ):
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("ship_desktop")
+def target_tasks_ship_desktop(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to ship desktop.
+ Previous build deps will be optimized out via action task."""
+ is_rc = parameters.get("release_type") == "release-rc"
+ if is_rc:
+ # ship_firefox_rc runs after `promote` rather than `push`; include
+ # all promote tasks.
+ filtered_for_candidates = target_tasks_promote_desktop(
+ full_task_graph,
+ parameters,
+ graph_config,
+ )
+ else:
+ # ship_firefox runs after `push`; include all push tasks.
+ filtered_for_candidates = target_tasks_push_desktop(
+ full_task_graph,
+ parameters,
+ graph_config,
+ )
+
+ def filter(task):
+ if not filter_out_missing_signoffs(task, parameters):
+ return False
+ # Include promotion tasks; these will be optimized out
+ if task.label in filtered_for_candidates:
+ return True
+
+ if (
+ task.attributes.get("shipping_product") != parameters["release_product"]
+ or task.attributes.get("shipping_phase") != "ship"
+ ):
+ return False
+
+ if "secondary" in task.kind:
+ return is_rc
+ return not is_rc
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("pine_tasks")
+def target_tasks_pine(full_task_graph, parameters, graph_config):
+ """Bug 1339179 - no mobile automation needed on pine"""
+
+ def filter(task):
+ platform = task.attributes.get("build_platform")
+ # disable mobile jobs
+ if str(platform).startswith("android"):
+ return False
+ # disable asan
+ if platform == "linux64-asan":
+ return False
+ # disable non-pine and tasks with a shipping phase
+ if standard_filter(task, parameters) or filter_out_shipping_phase(
+ task, parameters
+ ):
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("kaios_tasks")
+def target_tasks_kaios(full_task_graph, parameters, graph_config):
+ """The set of tasks to run for kaios integration"""
+
+ def filter(task):
+ # We disable everything in central, and adjust downstream.
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("ship_geckoview")
+def target_tasks_ship_geckoview(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to ship geckoview nightly. The
+ nightly build process involves a pipeline of builds and an upload to
+ maven.mozilla.org."""
+ index_path = (
+ f"{graph_config['trust-domain']}.v2.{parameters['project']}.revision."
+ f"{parameters['head_rev']}.taskgraph.decision-ship-geckoview"
+ )
+ if os.environ.get("MOZ_AUTOMATION") and retry(
+ index_exists,
+ args=(index_path,),
+ kwargs={
+ "reason": "to avoid triggering multiple nightlies off the same revision",
+ },
+ ):
+ return []
+
+ def filter(task):
+ # XXX Starting 69, we don't ship Fennec Nightly anymore. We just want geckoview to be
+ # uploaded
+ return task.attributes.get("shipping_product") == "fennec" and task.kind in (
+ "beetmover-geckoview",
+ "upload-symbols",
+ )
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("custom-car_perf_testing")
+def target_tasks_custom_car_perf_testing(full_task_graph, parameters, graph_config):
+ """Select tasks required for running daily performance tests for custom chromium-as-release."""
+
+ def filter(task):
+ platform = task.attributes.get("test_platform")
+ attributes = task.attributes
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+
+ try_name = attributes.get("raptor_try_name")
+
+ # Completely ignore all non-shippable platforms
+ if "shippable" not in platform:
+ return False
+
+ # ignore all windows 7 perf jobs scheduled automatically
+ if "windows7" in platform or "windows10-32" in platform:
+ return False
+
+ # Desktop selection only for CaR
+ if "android" not in platform:
+ if "browsertime" in try_name and "custom-car" in try_name:
+ return True
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("general_perf_testing")
+def target_tasks_general_perf_testing(full_task_graph, parameters, graph_config):
+ """
+ Select tasks required for running performance tests 3 times a week.
+ """
+
+ def filter(task):
+ platform = task.attributes.get("test_platform")
+ attributes = task.attributes
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+
+ try_name = attributes.get("raptor_try_name")
+
+ # Completely ignore all non-shippable platforms
+ if "shippable" not in platform:
+ return False
+
+ # ignore all windows 7 perf jobs scheduled automatically
+ if "windows7" in platform or "windows10-32" in platform:
+ return False
+
+ # Desktop selection
+ if "android" not in platform:
+ # Select some browsertime tasks as desktop smoke-tests
+ if "browsertime" in try_name:
+ if "chrome" in try_name:
+ return True
+ if "chromium" in try_name:
+ return True
+ # chromium-as-release has it's own cron
+ if "custom-car" in try_name:
+ return False
+ if "-live" in try_name:
+ return True
+ if "-fis" in try_name:
+ return False
+ if "linux" in platform:
+ if "speedometer" in try_name:
+ return True
+ if "safari" and "benchmark" in try_name:
+ # Speedometer 3 is broken on Safari, see bug 1802922
+ if "speedometer3" in try_name:
+ return False
+ return True
+ else:
+ # Don't run tp6 raptor tests
+ if "tp6" in try_name:
+ return False
+ # Android selection
+ elif accept_raptor_android_build(platform):
+ if "chrome-m" in try_name and (
+ ("ebay" in try_name and "live" not in try_name)
+ or (
+ "live" in try_name
+ and ("facebook" in try_name or "dailymail" in try_name)
+ )
+ ):
+ return False
+ # Ignore all fennec tests here, we run those weekly
+ if "fennec" in try_name:
+ return False
+ # Only run webrender tests
+ if "chrome-m" not in try_name and "-qr" not in platform:
+ return False
+ # Select live site tests
+ if "-live" in try_name:
+ return True
+ # Select fenix resource usage tests
+ if "fenix" in try_name:
+ # Bug 1816421 disable fission perf tests
+ if "-fis" in try_name:
+ return False
+ if "-power" in try_name:
+ return True
+ # Select geckoview resource usage tests
+ if "geckoview" in try_name:
+ # Bug 1816421 disable fission perf tests
+ if "-fis" in try_name:
+ return False
+ # Run cpu+memory, and power tests
+ cpu_n_memory_task = "-cpu" in try_name and "-memory" in try_name
+ power_task = "-power" in try_name
+ # Ignore cpu+memory+power tests
+ if power_task and cpu_n_memory_task:
+ return False
+ if cpu_n_memory_task:
+ return False
+ if power_task:
+ return "browsertime" in try_name
+ # Select browsertime-specific tests
+ if "browsertime" in try_name:
+ if "speedometer" in try_name:
+ return True
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+def make_desktop_nightly_filter(platforms):
+ """Returns a filter that gets all nightly tasks on the given platform."""
+
+ def filter(task, parameters):
+ return all(
+ [
+ filter_on_platforms(task, platforms),
+ filter_for_project(task, parameters),
+ task.attributes.get("shippable", False),
+ # Tests and nightly only builds don't have `shipping_product` set
+ task.attributes.get("shipping_product")
+ in {None, "firefox", "thunderbird"},
+ task.kind not in {"l10n"}, # no on-change l10n
+ ]
+ )
+
+ return filter
+
+
+@_target_task("sp-perftests")
+def target_tasks_speedometer_tests(full_task_graph, parameters, graph_config):
+ def filter(task):
+ platform = task.attributes.get("test_platform")
+ attributes = task.attributes
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+ if "windows10-32" not in platform:
+ try_name = attributes.get("raptor_try_name")
+ if (
+ "browsertime" in try_name
+ and "speedometer" in try_name
+ and "chrome" in try_name
+ ):
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("nightly_linux")
+def target_tasks_nightly_linux(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of linux. The
+ nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter({"linux64-shippable", "linux-shippable"})
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("nightly_macosx")
+def target_tasks_nightly_macosx(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of macosx. The
+ nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter({"macosx64-shippable"})
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("nightly_win32")
+def target_tasks_nightly_win32(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of win32 and win64.
+ The nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter({"win32-shippable"})
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("nightly_win64")
+def target_tasks_nightly_win64(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of win32 and win64.
+ The nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter({"win64-shippable"})
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("nightly_win64_aarch64")
+def target_tasks_nightly_win64_aarch64(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of win32 and win64.
+ The nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter({"win64-aarch64-shippable"})
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("nightly_asan")
+def target_tasks_nightly_asan(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of asan. The
+ nightly build process involves a pipeline of builds, signing,
+ and, eventually, uploading the tasks to balrog."""
+ filter = make_desktop_nightly_filter(
+ {"linux64-asan-reporter-shippable", "win64-asan-reporter-shippable"}
+ )
+ return [l for l, t in full_task_graph.tasks.items() if filter(t, parameters)]
+
+
+@_target_task("daily_releases")
+def target_tasks_daily_releases(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to identify if we should release.
+ If we determine that we should the task will communicate to ship-it to
+ schedule the release itself."""
+
+ def filter(task):
+ return task.kind in ["maybe-release"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("nightly_desktop")
+def target_tasks_nightly_desktop(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required for a nightly build of linux, mac,
+ windows."""
+ index_path = (
+ f"{graph_config['trust-domain']}.v2.{parameters['project']}.revision."
+ f"{parameters['head_rev']}.taskgraph.decision-nightly-desktop"
+ )
+ if os.environ.get("MOZ_AUTOMATION") and retry(
+ index_exists,
+ args=(index_path,),
+ kwargs={
+ "reason": "to avoid triggering multiple nightlies off the same revision",
+ },
+ ):
+ return []
+
+ # Tasks that aren't platform specific
+ release_filter = make_desktop_nightly_filter({None})
+ release_tasks = [
+ l for l, t in full_task_graph.tasks.items() if release_filter(t, parameters)
+ ]
+ # Avoid duplicate tasks.
+ return list(
+ set(target_tasks_nightly_win32(full_task_graph, parameters, graph_config))
+ | set(target_tasks_nightly_win64(full_task_graph, parameters, graph_config))
+ | set(
+ target_tasks_nightly_win64_aarch64(
+ full_task_graph, parameters, graph_config
+ )
+ )
+ | set(target_tasks_nightly_macosx(full_task_graph, parameters, graph_config))
+ | set(target_tasks_nightly_linux(full_task_graph, parameters, graph_config))
+ | set(target_tasks_nightly_asan(full_task_graph, parameters, graph_config))
+ | set(release_tasks)
+ )
+
+
+# Run Searchfox analysis once daily.
+@_target_task("searchfox_index")
+def target_tasks_searchfox(full_task_graph, parameters, graph_config):
+ """Select tasks required for indexing Firefox for Searchfox web site each day"""
+ return [
+ "searchfox-linux64-searchfox/debug",
+ "searchfox-macosx64-searchfox/debug",
+ "searchfox-win64-searchfox/debug",
+ "searchfox-android-armv7-searchfox/debug",
+ "source-test-file-metadata-bugzilla-components",
+ "source-test-file-metadata-test-info-all",
+ "source-test-wpt-metadata-summary",
+ ]
+
+
+# Run build linux64-plain-clang-trunk/opt on mozilla-central/beta with perf tests
+@_target_task("linux64_clang_trunk_perf")
+def target_tasks_build_linux64_clang_trunk_perf(
+ full_task_graph, parameters, graph_config
+):
+ """Select tasks required to run perf test on linux64 build with clang trunk"""
+
+ # Only keep tasks generated from platform `linux1804-64-clang-trunk-qr/opt`
+ def filter(task_label):
+ if "linux1804-64-clang-trunk-qr/opt" in task_label:
+ return True
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t.label)]
+
+
+# Run Updatebot's cron job 4 times daily.
+@_target_task("updatebot_cron")
+def target_tasks_updatebot_cron(full_task_graph, parameters, graph_config):
+ """Select tasks required to run Updatebot's cron job"""
+ return ["updatebot-cron"]
+
+
+@_target_task("customv8_update")
+def target_tasks_customv8_update(full_task_graph, parameters, graph_config):
+ """Select tasks required for building latest d8/v8 version."""
+ return ["toolchain-linux64-custom-v8"]
+
+
+@_target_task("chromium_update")
+def target_tasks_chromium_update(full_task_graph, parameters, graph_config):
+ """Select tasks required for building latest chromium versions."""
+ return [
+ "fetch-linux64-chromium",
+ "fetch-win32-chromium",
+ "fetch-win64-chromium",
+ "fetch-mac-chromium",
+ "toolchain-linux64-custom-car",
+ "toolchain-win64-custom-car",
+ ]
+
+
+@_target_task("file_update")
+def target_tasks_file_update(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to perform nightly in-tree file updates"""
+
+ def filter(task):
+ # For now any task in the repo-update kind is ok
+ return task.kind in ["repo-update"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("l10n_bump")
+def target_tasks_l10n_bump(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to perform l10n bumping."""
+
+ def filter(task):
+ # For now any task in the repo-update kind is ok
+ return task.kind in ["l10n-bump"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("merge_automation")
+def target_tasks_merge_automation(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to perform repository merges."""
+
+ def filter(task):
+ # For now any task in the repo-update kind is ok
+ return task.kind in ["merge-automation"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("scriptworker_canary")
+def target_tasks_scriptworker_canary(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to run scriptworker canaries."""
+
+ def filter(task):
+ # For now any task in the repo-update kind is ok
+ return task.kind in ["scriptworker-canary"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("cron_bouncer_check")
+def target_tasks_bouncer_check(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to perform bouncer version verification."""
+
+ def filter(task):
+ if not filter_for_project(task, parameters):
+ return False
+ # For now any task in the repo-update kind is ok
+ return task.kind in ["cron-bouncer-check"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("staging_release_builds")
+def target_tasks_staging_release(full_task_graph, parameters, graph_config):
+ """
+ Select all builds that are part of releases.
+ """
+
+ def filter(task):
+ if not task.attributes.get("shipping_product"):
+ return False
+ if parameters["release_type"].startswith(
+ "esr"
+ ) and "android" in task.attributes.get("build_platform", ""):
+ return False
+ if parameters["release_type"] != "beta" and "devedition" in task.attributes.get(
+ "build_platform", ""
+ ):
+ return False
+ if task.attributes.get("shipping_phase") == "build":
+ return True
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("release_simulation")
+def target_tasks_release_simulation(full_task_graph, parameters, graph_config):
+ """
+ Select builds that would run on push on a release branch.
+ """
+ project_by_release = {
+ "nightly": "mozilla-central",
+ "beta": "mozilla-beta",
+ "release": "mozilla-release",
+ "esr102": "mozilla-esr102",
+ "esr115": "mozilla-esr115",
+ }
+ target_project = project_by_release.get(parameters["release_type"])
+ if target_project is None:
+ raise Exception("Unknown or unspecified release type in simulation run.")
+
+ def filter_for_target_project(task):
+ """Filter tasks by project. Optionally enable nightlies."""
+ run_on_projects = set(task.attributes.get("run_on_projects", []))
+ return match_run_on_projects(target_project, run_on_projects)
+
+ def filter_out_android_on_esr(task):
+ if parameters["release_type"].startswith(
+ "esr"
+ ) and "android" in task.attributes.get("build_platform", ""):
+ return False
+ return True
+
+ return [
+ l
+ for l, t in full_task_graph.tasks.items()
+ if filter_release_tasks(t, parameters)
+ and filter_out_cron(t, parameters)
+ and filter_for_target_project(t)
+ and filter_out_android_on_esr(t)
+ ]
+
+
+@_target_task("codereview")
+def target_tasks_codereview(full_task_graph, parameters, graph_config):
+ """Select all code review tasks needed to produce a report"""
+
+ def filter(task):
+ # Ending tasks
+ if task.kind in ["code-review"]:
+ return True
+
+ # Analyzer tasks
+ if task.attributes.get("code-review") is True:
+ return True
+
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("nothing")
+def target_tasks_nothing(full_task_graph, parameters, graph_config):
+ """Select nothing, for DONTBUILD pushes"""
+ return []
+
+
+@_target_task("daily_beta_perf")
+def target_tasks_daily_beta_perf(full_task_graph, parameters, graph_config):
+ """
+ Select performance tests on the beta branch to be run daily
+ """
+ index_path = (
+ f"{graph_config['trust-domain']}.v2.{parameters['project']}.revision."
+ f"{parameters['head_rev']}.taskgraph.decision-daily-beta-perf"
+ )
+ if os.environ.get("MOZ_AUTOMATION") and retry(
+ index_exists,
+ args=(index_path,),
+ kwargs={
+ "reason": "to avoid triggering multiple daily beta perftests off of the same revision",
+ },
+ ):
+ return []
+
+ def filter(task):
+ platform = task.attributes.get("test_platform")
+ attributes = task.attributes
+ try_name = attributes.get("raptor_try_name")
+
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+
+ if platform and accept_raptor_android_build(platform):
+ # Select browsertime & geckoview specific tests
+ if "browsertime" and "geckoview" in try_name:
+ if "g5" in platform:
+ return False
+ if "power" in try_name:
+ return False
+ if "cpu" in try_name:
+ return False
+ if "profiling" in try_name:
+ return False
+ if "-live" in try_name:
+ return False
+ if "speedometer" in try_name:
+ return True
+ if "webgl" in try_name:
+ return True
+ if "tp6m" in try_name:
+ return True
+
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("weekly_release_perf")
+def target_tasks_weekly_release_perf(full_task_graph, parameters, graph_config):
+ """
+ Select performance tests on the release branch to be run weekly
+ """
+
+ def filter(task):
+ platform = task.attributes.get("test_platform")
+ attributes = task.attributes
+ try_name = attributes.get("raptor_try_name")
+
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+
+ if platform and accept_raptor_android_build(platform):
+ # Select browsertime & geckoview specific tests
+ if "browsertime" and "geckoview" in try_name:
+ if "g5" in platform:
+ return False
+ if "power" in try_name:
+ return False
+ if "cpu" in try_name:
+ return False
+ if "profiling" in try_name:
+ return False
+ if "-live" in try_name:
+ return False
+ if "speedometer" in try_name:
+ return True
+ if "webgl" in try_name:
+ return True
+ if "tp6m" in try_name:
+ return True
+ if "youtube-playback" in try_name:
+ return True
+
+ return False
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("raptor_tp6m")
+def target_tasks_raptor_tp6m(full_task_graph, parameters, graph_config):
+ """
+ Select tasks required for running raptor cold page-load tests on fenix and refbrow
+ """
+
+ def filter(task):
+ platform = task.attributes.get("build_platform")
+ attributes = task.attributes
+
+ if platform and "android" not in platform:
+ return False
+ if attributes.get("unittest_suite") != "raptor":
+ return False
+ try_name = attributes.get("raptor_try_name")
+ if "-cold" in try_name and "shippable" in platform:
+ # Get browsertime amazon smoke tests
+ if (
+ "browsertime" in try_name
+ and "amazon" in try_name
+ and "search" not in try_name
+ and "fenix" in try_name
+ ):
+ return True
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("backfill_all_browsertime")
+def target_tasks_backfill_all_browsertime(full_task_graph, parameters, graph_config):
+ """
+ Search for revisions that contains patches that were reviewed by perftest reviewers
+ and landed the day before the cron is running. Trigger backfill-all-browsertime action
+ task on each of them.
+ """
+ from gecko_taskgraph.actions.util import get_decision_task_id, get_pushes
+
+ def date_is_yesterday(date):
+ yesterday = datetime.today() - timedelta(days=1)
+ date = datetime.fromtimestamp(date)
+ return date.date() == yesterday.date()
+
+ def reviewed_by_perftest(push):
+ try:
+ commit_message = get_hg_commit_message(
+ os.path.join(GECKO, graph_config["product-dir"]), rev=push
+ )
+ except Exception as e:
+ print(e)
+ return False
+
+ for line in commit_message.split("\n\n"):
+ if line.lower().startswith("bug ") and "r=" in line:
+ if "perftest-reviewers" in line.split("r=")[-1]:
+ print(line)
+ return True
+ return False
+
+ pushes = get_pushes(
+ project=parameters["head_repository"],
+ end_id=int(parameters["pushlog_id"]),
+ depth=200,
+ full_response=True,
+ )
+ for push_id in sorted([int(p) for p in pushes.keys()], reverse=True):
+ push_rev = pushes[str(push_id)]["changesets"][-1]
+ push_info = find_hg_revision_push_info(
+ "https://hg.mozilla.org/integration/" + parameters["project"], push_rev
+ )
+ pushdate = int(push_info["pushdate"])
+ if date_is_yesterday(pushdate) and reviewed_by_perftest(push_rev):
+ from gecko_taskgraph.actions.util import trigger_action
+
+ print(
+ f"Revision {push_rev} was created yesterday and was reviewed by "
+ f"#perftest-reviewers."
+ )
+ try:
+ push_decision_task_id = get_decision_task_id(
+ parameters["project"], push_id
+ )
+ except Exception:
+ print(f"Could not find decision task for push {push_id}")
+ continue
+ try:
+ trigger_action(
+ action_name="backfill-all-browsertime",
+ # This lets the action know on which push we want to add a new task
+ decision_task_id=push_decision_task_id,
+ )
+ except Exception as e:
+ print(f"Failed to trigger action for {push_rev}: {e}")
+
+ return []
+
+
+@_target_task("condprof")
+def target_tasks_condprof(full_task_graph, parameters, graph_config):
+ """
+ Select tasks required for building conditioned profiles.
+ """
+ for name, task in full_task_graph.tasks.items():
+ if task.kind == "condprof":
+ if "a51" not in name: # bug 1765348
+ yield name
+
+
+@_target_task("system_symbols")
+def target_tasks_system_symbols(full_task_graph, parameters, graph_config):
+ """
+ Select tasks for scraping and uploading system symbols.
+ """
+ for name, task in full_task_graph.tasks.items():
+ if task.kind in [
+ "system-symbols",
+ "system-symbols-upload",
+ "system-symbols-reprocess",
+ ]:
+ yield name
+
+
+@_target_task("perftest")
+def target_tasks_perftest(full_task_graph, parameters, graph_config):
+ """
+ Select perftest tasks we want to run daily
+ """
+ for name, task in full_task_graph.tasks.items():
+ if task.kind != "perftest":
+ continue
+ if task.attributes.get("cron", False):
+ yield name
+
+
+@_target_task("perftest-on-autoland")
+def target_tasks_perftest_autoland(full_task_graph, parameters, graph_config):
+ """
+ Select perftest tasks we want to run daily
+ """
+ for name, task in full_task_graph.tasks.items():
+ if task.kind != "perftest":
+ continue
+ if task.attributes.get("cron", False) and any(
+ test_name in name for test_name in ["view"]
+ ):
+ yield name
+
+
+@_target_task("l10n-cross-channel")
+def target_tasks_l10n_cross_channel(full_task_graph, parameters, graph_config):
+ """Select the set of tasks required to run l10n cross-channel."""
+
+ def filter(task):
+ return task.kind in ["l10n-cross-channel"]
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]
+
+
+@_target_task("are-we-esmified-yet")
+def target_tasks_are_we_esmified_yet(full_task_graph, parameters, graph_config):
+ """
+ select the task to track the progress of the esmification project
+ """
+ return [
+ l for l, t in full_task_graph.tasks.items() if t.kind == "are-we-esmified-yet"
+ ]
+
+
+@_target_task("eslint-build")
+def target_tasks_eslint_build(full_task_graph, parameters, graph_config):
+ """Select the task to run additional ESLint rules which require a build."""
+
+ for name, task in full_task_graph.tasks.items():
+ if task.kind != "source-test":
+ continue
+ if "eslint-build" in name:
+ yield name
+
+
+@_target_task("holly_tasks")
+def target_tasks_holly(full_task_graph, parameters, graph_config):
+ """Bug 1814661: only run updatebot tasks on holly"""
+
+ def filter(task):
+ return task.kind == "updatebot"
+
+ return [l for l, t in full_task_graph.tasks.items() if filter(t)]