def create_started_file(bucket): """Create the started file in gcs for gubernator.""" contents = prow_artifacts.create_started() target = os.path.join(prow_artifacts.get_gcs_dir(bucket), "started.json") util.upload_to_gcs(contents, target)
def run_papermill_job(
notebook_path,
name,
namespace, # pylint: disable=too-many-branches,too-many-statements
repos,
image):
"""Generate a K8s job to run a notebook using papermill
Args:
notebook_path: Path to the notebook. This should be in the form
"{REPO_OWNER}/{REPO}/path/to/notebook.ipynb"
name: Name for the K8s job
namespace: The namespace where the job should run.
repos: Which repos to checkout; if None or empty tries
to infer based on PROW environment variables
image: The docker image to run the notebook in.
"""
util.maybe_activate_service_account()
with open("job.yaml") as hf:
job = yaml.load(hf)
if notebook_path.startswith("/"):
raise ValueError(
"notebook_path={0} should not start with /".format(notebook_path))
# We need to checkout the correct version of the code
# in presubmits and postsubmits. We should check the environment variables
# for the prow environment variables to get the appropriate values.
# We should probably also only do that if the
# See
# https://github.com/kubernetes/test-infra/blob/45246b09ed105698aa8fb928b7736d14480def29/prow/jobs.md#job-environment-variables
if not repos:
repos = argo_build_util.get_repo_from_prow_env()
if not repos:
raise ValueError("Could not get repos from prow environment variable "
"and --repos isn't explicitly set")
repos += ",kubeflow/testing@HEAD"
logging.info("Repos set to %s", repos)
job["spec"]["template"]["spec"]["initContainers"][0]["command"] = [
"/usr/local/bin/checkout_repos.sh",
"--repos=" + repos,
"--src_dir=/src",
"--depth=all",
]
job["spec"]["template"]["spec"]["containers"][0]["image"] = image
full_notebook_path = os.path.join("/src", notebook_path)
job["spec"]["template"]["spec"]["containers"][0]["command"] = [
"python3", "-m", "kubeflow.examples.notebook_tests.execute_notebook",
"--notebook_path", full_notebook_path
]
job["spec"]["template"]["spec"]["containers"][0][
"workingDir"] = os.path.dirname(full_notebook_path)
# The prow bucket to use for results/artifacts
prow_bucket = prow_artifacts.PROW_RESULTS_BUCKET
if os.getenv("REPO_OWNER") and os.getenv("REPO_NAME"):
# Running under prow
prow_dir = prow_artifacts.get_gcs_dir(prow_bucket)
logging.info("Prow artifacts dir: %s", prow_dir)
prow_dir = os.path.join(prow_dir, "artifacts")
if os.getenv("TEST_TARGET_NAME"):
prow_dir = os.path.join(prow_dir,
os.getenv("TEST_TARGET_NAME").lstrip("/"))
prow_bucket, prow_path = util.split_gcs_uri(prow_dir)
else:
prow_path = "notebook-test" + datetime.datetime.now().strftime(
"%H%M%S")
prow_path = prow_path + "-" + uuid.uuid4().hex[0:3]
prow_dir = util.to_gcs_uri(prow_bucket, prow_path)
prow_path = os.path.join(prow_path, name + ".html")
output_gcs = util.to_gcs_uri(NB_BUCKET, prow_path)
job["spec"]["template"]["spec"]["containers"][0]["env"] = [
{
"name": "OUTPUT_GCS",
"value": output_gcs
},
{
"name": "PYTHONPATH",
"value": "/src/kubeflow/testing/py:/src/kubeflow/examples/py"
},
]
logging.info("Notebook will be written to %s", output_gcs)
util.load_kube_config(persist_config=False)
if name:
job["metadata"]["name"] = name
else:
job["metadata"]["name"] = ("notebook-test-" +
datetime.datetime.now().strftime("%H%M%S") +
"-" + uuid.uuid4().hex[0:3])
name = job["metadata"]["name"]
job["metadata"]["namespace"] = namespace
# Create an API client object to talk to the K8s master.
api_client = k8s_client.ApiClient()
batch_api = k8s_client.BatchV1Api(api_client)
logging.info("Creating job:\n%s", yaml.dump(job))
actual_job = batch_api.create_namespaced_job(job["metadata"]["namespace"],
job)
logging.info("Created job %s.%s:\n%s", namespace, name,
yaml.safe_dump(actual_job.to_dict()))
final_job = util.wait_for_job(api_client,
namespace,
name,
timeout=datetime.timedelta(minutes=30))
logging.info("Final job:\n%s", yaml.safe_dump(final_job.to_dict()))
# Download notebook html to artifacts
logging.info("Copying %s to bucket %s", output_gcs, prow_bucket)
storage_client = storage.Client()
bucket = storage_client.get_bucket(NB_BUCKET)
blob = bucket.get_blob(prow_path)
destination_bucket = storage_client.get_bucket(prow_bucket)
bucket.copy_blob(blob, destination_bucket)
if not final_job.status.conditions:
raise RuntimeError("Job {0}.{1}; did not complete".format(
namespace, name))
last_condition = final_job.status.conditions[-1]
if last_condition.type not in ["Complete"]:
logging.error("Job didn't complete successfully")
raise RuntimeError("Job {0}.{1} failed".format(namespace, name))
def run(args, file_handler): # pylint: disable=too-many-statements,too-many-branches
job_type = os.getenv("JOB_TYPE")
repo_owner = os.getenv("REPO_OWNER")
repo_name = os.getenv("REPO_NAME")
pull_base_sha = os.getenv("PULL_BASE_SHA")
# For presubmit/postsubmit jobs, find the list of files changed by the PR.
diff_command = []
if job_type == "presubmit":
diff_command = ["git", "diff", "--name-only", "master"]
elif job_type == "postsubmit":
diff_command = ["git", "diff", "--name-only", pull_base_sha + "^", pull_base_sha]
changed_files = []
if job_type == "presubmit" or job_type == "postsubmit":
changed_files = util.run(diff_command,
cwd=os.path.join(args.repos_dir, repo_owner, repo_name)).splitlines()
for f in changed_files:
logging.info("File %s is modified.", f)
if args.release:
generate_env_from_head(args)
workflows = []
if args.config_file:
workflows.extend(parse_config_file(args.config_file, args.repos_dir))
create_started_file(args.bucket)
util.maybe_activate_service_account()
util.configure_kubectl(args.project, args.zone, args.cluster)
util.load_kube_config()
workflow_names = []
ui_urls = {}
for w in workflows:
# Create the name for the workflow
# We truncate sha numbers to prevent the workflow name from being too large.
# Workflow name should not be more than 63 characters because its used
# as a label on the pods.
workflow_name = os.getenv("JOB_NAME") + "-" + w.name
ks_cmd = get_ksonnet_cmd(w)
# Print ksonnet version
util.run([ks_cmd, "version"])
# Skip this workflow if it is scoped to a different job type.
if w.job_types and not job_type in w.job_types:
logging.info("Skipping workflow %s because job type %s is not one of "
"%s.", w.name, job_type, w.job_types)
continue
# If we are scoping this workflow to specific directories, check if any files
# modified match the specified regex patterns.
dir_modified = False
if w.include_dirs:
for f in changed_files:
for d in w.include_dirs:
if fnmatch.fnmatch(f, d):
dir_modified = True
logging.info("Triggering workflow %s because %s in dir %s is modified.",
w.name, f, d)
break
if dir_modified:
break
# Only consider modified files when the job is pre or post submit, and if
# the include_dirs stanza is defined.
if job_type != "periodic" and w.include_dirs and not dir_modified:
logging.info("Skipping workflow %s because no code modified in %s.",
w.name, w.include_dirs)
continue
if job_type == "presubmit":
workflow_name += "-{0}".format(os.getenv("PULL_NUMBER"))
workflow_name += "-{0}".format(os.getenv("PULL_PULL_SHA")[0:7])
elif job_type == "postsubmit":
workflow_name += "-{0}".format(os.getenv("PULL_BASE_SHA")[0:7])
workflow_name += "-{0}".format(os.getenv("BUILD_NUMBER"))
salt = uuid.uuid4().hex[0:4]
# Add some salt. This is mostly a convenience for the case where you
# are submitting jobs manually for testing/debugging. Since the prow should
# vend unique build numbers for each job.
workflow_name += "-{0}".format(salt)
workflow_names.append(workflow_name)
# Create a new environment for this run
env = workflow_name
util.run([ks_cmd, "env", "add", env], cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component,
"name", workflow_name],
cwd=w.app_dir)
# Set the prow environment variables.
prow_env = []
names = ["JOB_NAME", "JOB_TYPE", "BUILD_ID", "BUILD_NUMBER",
"PULL_BASE_SHA", "PULL_NUMBER", "PULL_PULL_SHA", "REPO_OWNER",
"REPO_NAME"]
names.sort()
for v in names:
if not os.getenv(v):
continue
prow_env.append("{0}={1}".format(v, os.getenv(v)))
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "prow_env",
",".join(prow_env)], cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "namespace",
get_namespace(args)], cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "bucket",
args.bucket], cwd=w.app_dir)
if args.release:
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "versionTag",
os.getenv("VERSION_TAG")], cwd=w.app_dir)
# Set any extra params. We do this in alphabetical order to make it easier to verify in
# the unittest.
param_names = w.params.keys()
param_names.sort()
for k in param_names:
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, k,
"{0}".format(w.params[k])], cwd=w.app_dir)
# For debugging print out the manifest
util.run([ks_cmd, "show", env, "-c", w.component], cwd=w.app_dir)
util.run([ks_cmd, "apply", env, "-c", w.component], cwd=w.app_dir)
ui_url = ("http://testing-argo.kubeflow.org/workflows/kubeflow-test-infra/{0}"
"?tab=workflow".format(workflow_name))
ui_urls[workflow_name] = ui_url
logging.info("URL for workflow: %s", ui_url)
success = True
workflow_phase = {}
try:
results = argo_client.wait_for_workflows(get_namespace(args),
workflow_names,
timeout=datetime.timedelta(minutes=180),
status_callback=argo_client.log_status)
for r in results:
phase = r.get("status", {}).get("phase")
name = r.get("metadata", {}).get("name")
workflow_phase[name] = phase
if phase != "Succeeded":
success = False
logging.info("Workflow %s/%s finished phase: %s", get_namespace(args), name, phase)
except util.TimeoutError:
success = False
logging.exception("Time out waiting for Workflows %s to finish", ",".join(workflow_names))
except Exception as e:
# We explicitly log any exceptions so that they will be captured in the
# build-log.txt that is uploaded to Gubernator.
logging.exception("Exception occurred: %s", e)
raise
finally:
success = prow_artifacts.finalize_prow_job(args.bucket, success, workflow_phase, ui_urls)
# Upload logs to GCS. No logs after this point will appear in the
# file in gcs
file_handler.flush()
util.upload_file_to_gcs(
file_handler.baseFilename,
os.path.join(prow_artifacts.get_gcs_dir(args.bucket), "build-log.txt"))
return success
def load_tekton_run(
params,
test_target_name,
tekton_run, # pylint: disable=too-many-branches
bucket,
repo_owner,
repo_under_test,
pull_revision):
"""Load Tekton configs and override information from Prow.
Args:
params: Extra parameters to be passed into Tekton pipelines.
test_target_name: test target name as classname in JUNIT.
tekton_run: File path to the PipelineRun config.
bucket: GCS bucket to write artifacts to.
"""
with open(tekton_run) as f:
config = yaml.load(f)
if config.get("kind", "") != "PipelineRun":
raise ValueError("Invalid config (not PipelineRun): " + config)
if not "generateName" in config["metadata"]:
raise ValueError("TektonPipeline is missing generateName")
logging.info("Reading Tekton PipelineRun config: %s",
config["metadata"]["generateName"])
workflow_name = config["metadata"]["generateName"]
artifacts_gcs = prow_artifacts.get_gcs_dir(bucket)
junit_path = "artifacts/junit_{run_name}".format(run_name=workflow_name)
args = {
"test-target-name": test_target_name,
"artifacts-gcs": artifacts_gcs,
"junit-path": junit_path,
}
for p in params:
args[p["name"]] = p["value"]
for param in config.get("spec", {}).get("params", []):
n = param.get("name", "")
v = param.get("value", "")
if n and v and not n in args:
args[n] = v
config["spec"]["params"] = []
for n in args:
logging.info("Writing Tekton param: %s -> %s", n, args[n])
config["spec"]["params"].append({
"name": n,
"value": args[n],
})
# Points to the revision under test.
repo_url = "https://github.com/{owner}/{name}.git".format(
owner=repo_owner, name=repo_under_test)
replacing_param = [
{
"name": "url",
"value": repo_url
},
{
"name": "revision",
"value": pull_revision
},
]
job_type = os.getenv("JOB_TYPE", "").lower()
if job_type in ["presubmit", "postsubmit"]:
logging.info("Job is type %s; looking for url %s", job_type, repo_url)
foundRepo = False
for resource in config["spec"].get("resources", []):
if resource.get("resourceSpec", {}).get("type", "") != "git":
pass
for param in resource.get("resourceSpec", {}).get("params", []):
if param.get("name", "") != "url":
continue
if param.get("value", "") == repo_url:
foundRepo = True
resource["resourceSpec"]["params"] = replacing_param
break
if not foundRepo:
raise ValueError(
("The TektonPipelineRun is missing a pipeline git "
"resource that matches the repo being tested by "
"prow. The pipeline parameters must include "
"a git resource whose URL is {0}".format(repo_url)))
else:
logging.info("Job is type %s; not looking for repo", job_type)
return config
def push_kfctl_to_gcs(kfctl_path):
bucket = prow_artifacts.PROW_RESULTS_BUCKET
logging.info("Bucket name: %s", prow_artifacts.get_gcs_dir(bucket))
gcs_path = os.path.join(
prow_artifacts.get_gcs_dir(bucket) + "/artifacts/build_bin/kfctl")
util.upload_file_to_gcs(kfctl_path, gcs_path)
def run(args, file_handler): # pylint: disable=too-many-statements,too-many-branches
# Check https://github.com/kubernetes/test-infra/blob/master/prow/jobs.md
# for a description of the injected environment variables.
job_type = os.getenv("JOB_TYPE")
repo_owner = os.getenv("REPO_OWNER")
repo_name = os.getenv("REPO_NAME")
base_branch_name = os.getenv("PULL_BASE_REF")
pull_base_sha = os.getenv("PULL_BASE_SHA")
# For presubmit/postsubmit jobs, find the list of files changed by the PR.
diff_command = []
if job_type == "presubmit":
# We need to get a common ancestor for the PR and the base branch
cloned_repo_dir = os.path.join(args.repos_dir, repo_owner, repo_name)
_ = util.run([
"git", "fetch", "origin",
base_branch_name + ":refs/remotes/origin/" + base_branch_name
],
cwd=cloned_repo_dir)
diff_command = ["git", "diff", "--name-only"]
diff_branch = "remotes/origin/{}".format(base_branch_name)
try:
common_ancestor = util.run(
["git", "merge-base", "HEAD", diff_branch],
cwd=cloned_repo_dir)
diff_command.append(common_ancestor)
except subprocess.CalledProcessError as e:
logging.warning(
"git merge-base failed; see "
"https://github.com/kubeflow/kubeflow/issues/3523. Diff "
"will be computed against the current master and "
"therefore files not changed in the PR might be "
"considered when determining which tests to trigger")
diff_command.append(diff_branch)
elif job_type == "postsubmit":
# See: https://git-scm.com/docs/git-diff
# This syntax compares the commit before pull_base_sha with the commit
# at pull_base_sha
diff_command = [
"git", "diff", "--name-only", pull_base_sha + "^", pull_base_sha
]
changed_files = []
if job_type in ("presubmit", "postsubmit"):
changed_files = util.run(diff_command,
cwd=os.path.join(args.repos_dir, repo_owner,
repo_name)).splitlines()
for f in changed_files:
logging.info("File %s is modified.", f)
if args.release:
generate_env_from_head(args)
workflows = []
config = {}
if args.config_file:
config, new_workflows = parse_config_file(args.config_file,
args.repos_dir)
workflows.extend(new_workflows)
# Add any paths to the python path
extra_py_paths = []
for p in config.get("python_paths", []):
# Assume that python_paths are in the format $REPO_OWNER/$REPO_NAME/path,
# we need to ensure that the repo is checked out if it is different from
# the current one, and if the repo is not kubeflow/testing (which is already
# checked out).
segments = p.split("/")
if ((segments[0] != repo_owner or segments[1] != repo_name)
and not p.startswith("kubeflow/testing")):
logging.info("Need to clone %s/%s", segments[0], segments[1])
util.clone_repo(
os.path.join(args.repos_dir, segments[0], segments[1]),
segments[0], segments[1])
path = os.path.join(args.repos_dir, p)
extra_py_paths.append(path)
kf_test_path = os.path.join(args.repos_dir, "kubeflow/testing/py")
if kf_test_path not in extra_py_paths:
logging.info("Adding %s to extra python paths", kf_test_path)
extra_py_paths.append(kf_test_path)
logging.info("Extra python paths: %s", ":".join(extra_py_paths))
# Create an initial version of the file with no urls
create_started_file(args.bucket, {})
util.maybe_activate_service_account()
util.configure_kubectl(args.project, args.zone, args.cluster)
util.load_kube_config()
workflow_names = []
ui_urls = {}
for w in workflows: # pylint: disable=too-many-nested-blocks
# Create the name for the workflow
# We truncate sha numbers to prevent the workflow name from being too large.
# Workflow name should not be more than 63 characters because its used
# as a label on the pods.
workflow_name = os.getenv("JOB_NAME") + "-" + w.name
# Skip this workflow if it is scoped to a different job type.
if w.job_types and not job_type in w.job_types:
logging.info(
"Skipping workflow %s because job type %s is not one of "
"%s.", w.name, job_type, w.job_types)
continue
# If we are scoping this workflow to specific directories, check if any files
# modified match the specified regex patterns.
dir_modified = False
if w.include_dirs:
for f in changed_files:
for d in w.include_dirs:
if fnmatch.fnmatch(f, d):
dir_modified = True
logging.info(
"Triggering workflow %s because %s in dir %s is modified.",
w.name, f, d)
break
if dir_modified:
break
# Only consider modified files when the job is pre or post submit, and if
# the include_dirs stanza is defined.
if job_type != "periodic" and w.include_dirs and not dir_modified:
logging.info(
"Skipping workflow %s because no code modified in %s.", w.name,
w.include_dirs)
continue
if job_type == "presubmit":
# When not running under prow we might not set all environment variables
if os.getenv("PULL_NUMBER"):
workflow_name += "-{0}".format(os.getenv("PULL_NUMBER"))
if os.getenv("PULL_PULL_SHA"):
workflow_name += "-{0}".format(os.getenv("PULL_PULL_SHA")[0:7])
elif job_type == "postsubmit":
if os.getenv("PULL_BASE_SHA"):
workflow_name += "-{0}".format(os.getenv("PULL_BASE_SHA")[0:7])
# Append the last 4 digits of the build number
if os.getenv("BUILD_NUMBER"):
workflow_name += "-{0}".format(os.getenv("BUILD_NUMBER")[-4:])
salt = uuid.uuid4().hex[0:4]
# Add some salt. This is mostly a convenience for the case where you
# are submitting jobs manually for testing/debugging. Since the prow should
# vend unique build numbers for each job.
workflow_name += "-{0}".format(salt)
workflow_names.append(workflow_name)
# check if ks workflow and run
if w.app_dir:
ks_cmd = ks_util.get_ksonnet_cmd(w.app_dir)
# Print ksonnet version
util.run([ks_cmd, "version"])
# Create a new environment for this run
env = workflow_name
util.run([
ks_cmd, "env", "add", env, "--namespace=" + get_namespace(args)
],
cwd=w.app_dir)
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component, "name",
workflow_name
],
cwd=w.app_dir)
# Set the prow environment variables.
prow_env = []
names = [
"JOB_NAME", "JOB_TYPE", "BUILD_ID", "BUILD_NUMBER",
"PULL_BASE_SHA", "PULL_NUMBER", "PULL_PULL_SHA", "REPO_OWNER",
"REPO_NAME"
]
names.sort()
for v in names:
if not os.getenv(v):
continue
prow_env.append("{0}={1}".format(v, os.getenv(v)))
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component,
"prow_env", ",".join(prow_env)
],
cwd=w.app_dir)
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component,
"namespace",
get_namespace(args)
],
cwd=w.app_dir)
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component, "bucket",
args.bucket
],
cwd=w.app_dir)
if args.release:
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component,
"versionTag",
os.getenv("VERSION_TAG")
],
cwd=w.app_dir)
# Set any extra params. We do this in alphabetical order to make it easier to verify in
# the unittest.
param_names = w.params.keys()
param_names.sort()
for k in param_names:
util.run([
ks_cmd, "param", "set", "--env=" + env, w.component, k,
"{0}".format(w.params[k])
],
cwd=w.app_dir)
# For debugging print out the manifest
util.run([ks_cmd, "show", env, "-c", w.component], cwd=w.app_dir)
util.run([ks_cmd, "apply", env, "-c", w.component], cwd=w.app_dir)
ui_url = (
"http://testing-argo.kubeflow.org/workflows/kubeflow-test-infra/{0}"
"?tab=workflow".format(workflow_name))
ui_urls[workflow_name] = ui_url
logging.info("URL for workflow: %s", ui_url)
else:
w.kwargs["name"] = workflow_name
w.kwargs["namespace"] = get_namespace(args)
if TEST_TARGET_ARG_NAME not in w.kwargs:
w.kwargs[TEST_TARGET_ARG_NAME] = w.name
logging.info(
"Workflow %s doesn't set arg %s; defaulting to %s", w.name,
TEST_TARGET_ARG_NAME, w.kwargs[TEST_TARGET_ARG_NAME])
# TODO(https://github.com/kubeflow/testing/issues/467): We shell out
# to e2e_tool in order to dumpy the Argo workflow to a file which then
# reimport. We do this because importing the py_func module appears
# to break when we have to dynamically adjust sys.path to insert
# new paths. Setting PYTHONPATH before launching python however appears
# to work which is why we shell out to e2e_tool.
command = [
"python", "-m", "kubeflow.testing.e2e_tool", "show", w.py_func
]
for k, v in w.kwargs.items():
# The fire module turns underscores in parameter names into hyphens
# so we convert underscores in parameter names to hyphens
command.append("--{0}={1}".format(k.replace("_", "-"), v))
with tempfile.NamedTemporaryFile(delete=False) as hf:
workflow_file = hf.name
command.append("--output=" + hf.name)
env = os.environ.copy()
env["PYTHONPATH"] = ":".join(extra_py_paths)
util.run(command, env=env)
with open(workflow_file) as hf:
wf_result = yaml.load(hf)
group, version = wf_result['apiVersion'].split('/')
k8s_co = k8s_client.CustomObjectsApi()
workflow_name = wf_result["metadata"]["name"]
py_func_result = k8s_co.create_namespaced_custom_object(
group=group,
version=version,
namespace=wf_result["metadata"]["namespace"],
plural='workflows',
body=wf_result)
logging.info("Created workflow:\n%s",
yaml.safe_dump(py_func_result))
ui_url = (
"http://testing-argo.kubeflow.org/workflows/kubeflow-test-infra/{0}"
"?tab=workflow".format(workflow_name))
ui_urls[workflow_name] = ui_url
logging.info("URL for workflow: %s", ui_url)
# We delay creating started.json until we know the Argo workflow URLs
create_started_file(args.bucket, ui_urls)
workflow_success = False
workflow_phase = {}
workflow_status_yamls = {}
results = []
try:
results = argo_client.wait_for_workflows(
get_namespace(args),
workflow_names,
timeout=datetime.timedelta(minutes=180),
status_callback=argo_client.log_status)
workflow_success = True
except util.ExceptionWithWorkflowResults as e:
# We explicitly log any exceptions so that they will be captured in the
# build-log.txt that is uploaded to Gubernator.
logging.exception("Exception occurred: %s", e)
results = e.workflow_results
raise
finally:
prow_artifacts_dir = prow_artifacts.get_gcs_dir(args.bucket)
# Upload logs to GCS. No logs after this point will appear in the
# file in gcs
file_handler.flush()
util.upload_file_to_gcs(
file_handler.baseFilename,
os.path.join(prow_artifacts_dir, "build-log.txt"))
# Upload workflow status to GCS.
for r in results:
phase = r.get("status", {}).get("phase")
name = r.get("metadata", {}).get("name")
workflow_phase[name] = phase
workflow_status_yamls[name] = yaml.safe_dump(
r, default_flow_style=False)
if phase != "Succeeded":
workflow_success = False
logging.info("Workflow %s/%s finished phase: %s",
get_namespace(args), name, phase)
for wf_name, wf_status in workflow_status_yamls.items():
util.upload_to_gcs(
wf_status,
os.path.join(prow_artifacts_dir,
'{}.yaml'.format(wf_name)))
all_tests_success = prow_artifacts.finalize_prow_job(
args.bucket, workflow_success, workflow_phase, ui_urls)
return all_tests_success
def run(args, file_handler): # pylint: disable=too-many-statements,too-many-branches
# Check https://github.com/kubernetes/test-infra/blob/master/prow/jobs.md
# for a description of the injected environment variables.
job_type = os.getenv("JOB_TYPE")
repo_owner = os.getenv("REPO_OWNER")
repo_name = os.getenv("REPO_NAME")
base_branch_name = os.getenv("PULL_BASE_REF")
pull_base_sha = os.getenv("PULL_BASE_SHA")
# For presubmit/postsubmit jobs, find the list of files changed by the PR.
diff_command = []
if job_type == "presubmit":
# We need to get a common ancestor for the PR and the base branch
cloned_repo_dir = os.path.join(args.repos_dir, repo_owner, repo_name)
_ = util.run(["git", "fetch", "origin", base_branch_name + ":refs/remotes/origin/" +
base_branch_name], cwd=cloned_repo_dir)
diff_command = ["git", "diff", "--name-only"]
diff_branch = "remotes/origin/{}".format(base_branch_name)
try:
common_ancestor = util.run(["git", "merge-base", "HEAD", diff_branch],
cwd=cloned_repo_dir)
diff_command.append(common_ancestor)
except subprocess.CalledProcessError as e:
logging.warning("git merge-base failed; see "
"https://github.com/kubeflow/kubeflow/issues/3523. Diff "
"will be computed against the current master and "
"therefore files not changed in the PR might be "
"considered when determining which tests to trigger")
diff_command.append(diff_branch)
elif job_type == "postsubmit":
# See: https://git-scm.com/docs/git-diff
# This syntax compares the commit before pull_base_sha with the commit
# at pull_base_sha
diff_command = ["git", "diff", "--name-only", pull_base_sha + "^", pull_base_sha]
changed_files = []
if job_type in ("presubmit", "postsubmit"):
changed_files = util.run(diff_command,
cwd=os.path.join(args.repos_dir, repo_owner, repo_name)).splitlines()
for f in changed_files:
logging.info("File %s is modified.", f)
if args.release:
generate_env_from_head(args)
workflows = []
if args.config_file:
workflows.extend(parse_config_file(args.config_file, args.repos_dir))
# Create an initial version of the file with no urls
create_started_file(args.bucket, {})
util.maybe_activate_service_account()
util.configure_kubectl(args.project, args.zone, args.cluster)
util.load_kube_config()
workflow_names = []
ui_urls = {}
for w in workflows:
# Create the name for the workflow
# We truncate sha numbers to prevent the workflow name from being too large.
# Workflow name should not be more than 63 characters because its used
# as a label on the pods.
workflow_name = os.getenv("JOB_NAME") + "-" + w.name
ks_cmd = ks_util.get_ksonnet_cmd(w.app_dir)
# Print ksonnet version
util.run([ks_cmd, "version"])
# Skip this workflow if it is scoped to a different job type.
if w.job_types and not job_type in w.job_types:
logging.info("Skipping workflow %s because job type %s is not one of "
"%s.", w.name, job_type, w.job_types)
continue
# If we are scoping this workflow to specific directories, check if any files
# modified match the specified regex patterns.
dir_modified = False
if w.include_dirs:
for f in changed_files:
for d in w.include_dirs:
if fnmatch.fnmatch(f, d):
dir_modified = True
logging.info("Triggering workflow %s because %s in dir %s is modified.",
w.name, f, d)
break
if dir_modified:
break
# Only consider modified files when the job is pre or post submit, and if
# the include_dirs stanza is defined.
if job_type != "periodic" and w.include_dirs and not dir_modified:
logging.info("Skipping workflow %s because no code modified in %s.",
w.name, w.include_dirs)
continue
if job_type == "presubmit":
workflow_name += "-{0}".format(os.getenv("PULL_NUMBER"))
workflow_name += "-{0}".format(os.getenv("PULL_PULL_SHA")[0:7])
elif job_type == "postsubmit":
workflow_name += "-{0}".format(os.getenv("PULL_BASE_SHA")[0:7])
# Append the last 4 digits of the build number
workflow_name += "-{0}".format(os.getenv("BUILD_NUMBER")[-4:])
salt = uuid.uuid4().hex[0:4]
# Add some salt. This is mostly a convenience for the case where you
# are submitting jobs manually for testing/debugging. Since the prow should
# vend unique build numbers for each job.
workflow_name += "-{0}".format(salt)
workflow_names.append(workflow_name)
# Create a new environment for this run
env = workflow_name
util.run([ks_cmd, "env", "add", env, "--namespace=" + get_namespace(args)],
cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component,
"name", workflow_name],
cwd=w.app_dir)
# Set the prow environment variables.
prow_env = []
names = ["JOB_NAME", "JOB_TYPE", "BUILD_ID", "BUILD_NUMBER",
"PULL_BASE_SHA", "PULL_NUMBER", "PULL_PULL_SHA", "REPO_OWNER",
"REPO_NAME"]
names.sort()
for v in names:
if not os.getenv(v):
continue
prow_env.append("{0}={1}".format(v, os.getenv(v)))
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "prow_env",
",".join(prow_env)], cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "namespace",
get_namespace(args)], cwd=w.app_dir)
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "bucket",
args.bucket], cwd=w.app_dir)
if args.release:
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, "versionTag",
os.getenv("VERSION_TAG")], cwd=w.app_dir)
# Set any extra params. We do this in alphabetical order to make it easier to verify in
# the unittest.
param_names = w.params.keys()
param_names.sort()
for k in param_names:
util.run([ks_cmd, "param", "set", "--env=" + env, w.component, k,
"{0}".format(w.params[k])], cwd=w.app_dir)
# For debugging print out the manifest
util.run([ks_cmd, "show", env, "-c", w.component], cwd=w.app_dir)
util.run([ks_cmd, "apply", env, "-c", w.component], cwd=w.app_dir)
ui_url = ("http://testing-argo.kubeflow.org/workflows/kubeflow-test-infra/{0}"
"?tab=workflow".format(workflow_name))
ui_urls[workflow_name] = ui_url
logging.info("URL for workflow: %s", ui_url)
# We delay creating started.json until we know the Argo workflow URLs
create_started_file(args.bucket, ui_urls)
workflow_success = False
workflow_phase = {}
workflow_status_yamls = {}
results = []
try:
results = argo_client.wait_for_workflows(
get_namespace(args), workflow_names,
timeout=datetime.timedelta(minutes=180),
status_callback=argo_client.log_status
)
workflow_success = True
except util.ExceptionWithWorkflowResults as e:
# We explicitly log any exceptions so that they will be captured in the
# build-log.txt that is uploaded to Gubernator.
logging.exception("Exception occurred: %s", e)
results = e.workflow_results
raise
finally:
prow_artifacts_dir = prow_artifacts.get_gcs_dir(args.bucket)
# Upload logs to GCS. No logs after this point will appear in the
# file in gcs
file_handler.flush()
util.upload_file_to_gcs(
file_handler.baseFilename,
os.path.join(prow_artifacts_dir, "build-log.txt"))
# Upload workflow status to GCS.
for r in results:
phase = r.get("status", {}).get("phase")
name = r.get("metadata", {}).get("name")
workflow_phase[name] = phase
workflow_status_yamls[name] = yaml.safe_dump(r, default_flow_style=False)
if phase != "Succeeded":
workflow_success = False
logging.info("Workflow %s/%s finished phase: %s", get_namespace(args), name, phase)
for wf_name, wf_status in workflow_status_yamls.items():
util.upload_to_gcs(
wf_status,
os.path.join(prow_artifacts_dir, '{}.yaml'.format(wf_name)))
all_tests_success = prow_artifacts.finalize_prow_job(
args.bucket, workflow_success, workflow_phase, ui_urls)
return all_tests_success
def run(args, file_handler): # pylint: disable=too-many-statements,too-many-branches
# Print ksonnet version
util.run(["ks", "version"])
if args.release:
generate_env_from_head(args)
workflows = []
if args.config_file:
workflows.extend(parse_config_file(args.config_file, args.repos_dir))
if args.app_dir and args.component:
# TODO(jlewi): We can get rid of this branch once all repos are using a prow_config.xml file.
workflows.append(
WorkflowComponent("legacy", args.app_dir, args.component, {}))
create_started_file(args.bucket)
util.maybe_activate_service_account()
util.configure_kubectl(args.project, args.zone, args.cluster)
util.load_kube_config()
api_client = k8s_client.ApiClient()
workflow_names = []
ui_urls = {}
for w in workflows:
# Create the name for the workflow
# We truncate sha numbers to prevent the workflow name from being too large.
# Workflow name should not be more than 63 characters because its used
# as a label on the pods.
workflow_name = os.getenv("JOB_NAME") + "-" + w.name
job_type = os.getenv("JOB_TYPE")
if job_type == "presubmit":
workflow_name += "-{0}".format(os.getenv("PULL_NUMBER"))
workflow_name += "-{0}".format(os.getenv("PULL_PULL_SHA")[0:7])
elif job_type == "postsubmit":
workflow_name += "-{0}".format(os.getenv("PULL_BASE_SHA")[0:7])
workflow_name += "-{0}".format(os.getenv("BUILD_NUMBER"))
salt = uuid.uuid4().hex[0:4]
# Add some salt. This is mostly a convenience for the case where you
# are submitting jobs manually for testing/debugging. Since the prow should
# vend unique build numbers for each job.
workflow_name += "-{0}".format(salt)
workflow_names.append(workflow_name)
# Create a new environment for this run
env = workflow_name
util.run(["ks", "env", "add", env], cwd=w.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, w.component, "name",
workflow_name
],
cwd=w.app_dir)
# Set the prow environment variables.
prow_env = []
names = [
"JOB_NAME", "JOB_TYPE", "BUILD_ID", "BUILD_NUMBER",
"PULL_BASE_SHA", "PULL_NUMBER", "PULL_PULL_SHA", "REPO_OWNER",
"REPO_NAME"
]
names.sort()
for v in names:
if not os.getenv(v):
continue
prow_env.append("{0}={1}".format(v, os.getenv(v)))
util.run([
"ks", "param", "set", "--env=" + env, w.component, "prow_env",
",".join(prow_env)
],
cwd=w.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, w.component, "namespace",
get_namespace(args)
],
cwd=w.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, w.component, "bucket",
args.bucket
],
cwd=w.app_dir)
if args.release:
util.run([
"ks", "param", "set", "--env=" + env, w.component,
"versionTag",
os.getenv("VERSION_TAG")
],
cwd=w.app_dir)
# Set any extra params. We do this in alphabetical order to make it easier to verify in
# the unittest.
param_names = w.params.keys()
param_names.sort()
for k in param_names:
util.run([
"ks", "param", "set", "--env=" + env, w.component, k,
"{0}".format(w.params[k])
],
cwd=w.app_dir)
# For debugging print out the manifest
util.run(["ks", "show", env, "-c", w.component], cwd=w.app_dir)
util.run(["ks", "apply", env, "-c", w.component], cwd=w.app_dir)
ui_url = (
"http://testing-argo.kubeflow.org/workflows/kubeflow-test-infra/{0}"
"?tab=workflow".format(workflow_name))
ui_urls[workflow_name] = ui_url
logging.info("URL for workflow: %s", ui_url)
success = True
workflow_phase = {}
try:
results = argo_client.wait_for_workflows(
api_client,
get_namespace(args),
workflow_names,
status_callback=argo_client.log_status)
for r in results:
phase = r.get("status", {}).get("phase")
name = r.get("metadata", {}).get("name")
workflow_phase[name] = phase
if phase != "Succeeded":
success = False
logging.info("Workflow %s/%s finished phase: %s",
get_namespace(args), name, phase)
except util.TimeoutError:
success = False
logging.error("Time out waiting for Workflows %s to finish",
",".join(workflow_names))
except Exception as e:
# We explicitly log any exceptions so that they will be captured in the
# build-log.txt that is uploaded to Gubernator.
logging.error("Exception occurred: %s", e)
raise
finally:
success = prow_artifacts.finalize_prow_job(args.bucket, success,
workflow_phase, ui_urls)
# Upload logs to GCS. No logs after this point will appear in the
# file in gcs
file_handler.flush()
util.upload_file_to_gcs(
file_handler.baseFilename,
os.path.join(prow_artifacts.get_gcs_dir(args.bucket),
"build-log.txt"))
return success
def run(args, file_handler):
create_started_file(args.bucket)
util.maybe_activate_service_account()
util.configure_kubectl(args.project, args.zone, args.cluster)
util.load_kube_config()
# Create the name for the workflow
# We truncate sha numbers to prevent the workflow name from being too large.
# Workflow name should not be more than 63 characters because its used
# as a label on the pods.
workflow_name = os.getenv("JOB_NAME")
job_type = os.getenv("JOB_TYPE")
if job_type == "presubmit":
workflow_name += "-{0}".format(os.getenv("PULL_NUMBER"))
workflow_name += "-{0}".format(os.getenv("PULL_PULL_SHA")[0:7])
elif job_type == "postsubmit":
workflow_name += "-{0}".format(os.getenv("PULL_BASE_SHA")[0:7])
workflow_name += "-{0}".format(os.getenv("BUILD_NUMBER"))
salt = uuid.uuid4().hex[0:4]
# Add some salt. This is mostly a convenience for the case where you
# are submitting jobs manually for testing/debugging. Since the prow should
# vend unique build numbers for each job.
workflow_name += "-{0}".format(salt)
# Create a new environment for this run
env = workflow_name
util.run(["ks", "env", "add", env], cwd=args.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, args.component, "name",
workflow_name
],
cwd=args.app_dir)
util.load_kube_config()
api_client = k8s_client.ApiClient()
# Set the prow environment variables.
prow_env = []
names = [
"JOB_NAME", "JOB_TYPE", "BUILD_ID", "BUILD_NUMBER", "PULL_BASE_SHA",
"PULL_NUMBER", "PULL_PULL_SHA", "REPO_OWNER", "REPO_NAME"
]
names.sort()
for v in names:
if not os.getenv(v):
continue
prow_env.append("{0}={1}".format(v, os.getenv(v)))
util.run([
"ks", "param", "set", "--env=" + env, args.component, "prow_env",
",".join(prow_env)
],
cwd=args.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, args.component, "namespace",
NAMESPACE
],
cwd=args.app_dir)
util.run([
"ks", "param", "set", "--env=" + env, args.component, "bucket",
args.bucket
],
cwd=args.app_dir)
# For debugging print out the manifest
util.run(["ks", "show", env, "-c", args.component], cwd=args.app_dir)
util.run(["ks", "apply", env, "-c", args.component], cwd=args.app_dir)
ui_url = (
"http://testing-argo.kubeflow.io/timeline/kubeflow-test-infra/{0}"
";tab=workflow".format(workflow_name))
logging.info("URL for workflow: %s", ui_url)
success = False
try:
results = argo_client.wait_for_workflow(
api_client,
NAMESPACE,
workflow_name,
status_callback=argo_client.log_status)
if results["status"]["phase"] == "Succeeded":
success = True
logging.info("Workflow %s/%s finished phase: %s", NAMESPACE,
workflow_name, results["status"]["phase"])
except util.TimeoutError:
success = False
logging.error("Time out waiting for Workflow %s/%s to finish",
NAMESPACE, workflow_name)
finally:
create_finished_file(args.bucket, success)
# Upload logs to GCS. No logs after this point will appear in the
# file in gcs
file_handler.flush()
upload_file_to_gcs(
file_handler.baseFilename,
os.path.join(prow_artifacts.get_gcs_dir(args.bucket),
"build-log.txt"))
return success
def create_finished_file(bucket, success):
"""Create the started file in gcs for gubernator."""
contents = prow_artifacts.create_finished(success)
target = os.path.join(prow_artifacts.get_gcs_dir(bucket), "finished.json")
upload_to_gcs(contents, target)
