def run_notebook_test(notebook_path, expected_messages, parameters=None):
output_path = execute_notebook(notebook_path, parameters=parameters)
import nbformat #pylint: disable=import-error
import nbconvert #pylint: disable=import-error
actual_output = open(output_path, 'r').read()
nb = nbformat.reads(actual_output, as_version=4)
html_exporter = nbconvert.HTMLExporter()
(html_output, _) = html_exporter.from_notebook_node(nb)
gcs_path = os.getenv("OUTPUT_GCS")
kf_util.upload_to_gcs(html_output, gcs_path)
for expected_message in expected_messages:
if not expected_message in actual_output:
logger.error(actual_output)
assert False, "Unable to find from output: " + expected_message
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(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 create_finished_file(bucket, success, workflow_phase, ui_urls):
"""Create the started file in gcs for gubernator."""
contents = create_finished(success, workflow_phase, ui_urls)
target = os.path.join(get_gcs_dir(bucket), "finished.json")
util.upload_to_gcs(contents, target)
