def test_resubmit_failed(cleanup):
cmd = f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT}"
ret = run_command(cmd)
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
agg = ResultsAggregator.load(OUTPUT)
results = agg.get_results_unsafe()
assert results
for result in results:
assert result.return_code == 0
x = results[0]
results[0] = Result(x.name, 1, x.status, x.exec_time_s, x.completion_time)
agg._write_results(results)
results_filename = os.path.join(OUTPUT, RESULTS_FILE)
final_results = load_data(results_filename)
final_results["results"][0]["return_code"] = 1
final_results["results_summary"]["num_failed"] = 1
final_results["results_summary"]["num_successful"] -= 1
dump_data(final_results, results_filename)
summary = ResultsSummary(OUTPUT)
assert summary.get_failed_results()[0].name == "1"
ret = run_command(f"{RESUBMIT_JOBS} {OUTPUT}")
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
summary = ResultsSummary(OUTPUT)
assert len(summary.get_successful_results()) == NUM_COMMANDS
def test_resubmit_missing(cleanup):
cmd = f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT}"
ret = run_command(cmd)
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
agg = ResultsAggregator.load(OUTPUT)
results = agg.get_results_unsafe()
assert results
for result in results:
assert result.return_code == 0
results.pop()
agg._write_results(results)
results_filename = os.path.join(OUTPUT, RESULTS_FILE)
final_results = load_data(results_filename)
missing = final_results["results"].pop()
final_results["results_summary"]["num_missing"] = 1
final_results["results_summary"]["num_successful"] -= 1
final_results["missing_jobs"] = [missing["name"]]
dump_data(final_results, results_filename)
summary = ResultsSummary(OUTPUT)
assert len(summary.get_failed_results()) == 0
assert len(summary.get_successful_results()) == NUM_COMMANDS - 1
ret = run_command(f"{RESUBMIT_JOBS} {OUTPUT}")
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
summary = ResultsSummary(OUTPUT)
assert len(summary.get_successful_results()) == NUM_COMMANDS
def _handle_completion(self, cluster):
result = Status.GOOD
self._results = ResultsAggregator.list_results(self._output)
if len(self._results) != self._config.get_num_jobs():
finished_jobs = {x.name for x in self._results}
all_jobs = {x.name for x in self._config.iter_jobs()}
missing_jobs = sorted(all_jobs.difference(finished_jobs))
logger.error(
"Error in result totals. num_results=%s total_num_jobs=%s",
len(self._results),
self._config.get_num_jobs(),
)
logger.error(
"These jobs did not finish: %s. Check for process crashes or HPC timeouts.",
missing_jobs,
)
result = Status.ERROR
else:
missing_jobs = []
self.write_results_summary(RESULTS_FILE, missing_jobs)
self._log_error_log_messages(self._output)
bytes_consumed = get_directory_size_bytes(self._output,
recursive=False)
event = StructuredLogEvent(
source="submitter",
category=EVENT_CATEGORY_RESOURCE_UTIL,
name=EVENT_NAME_BYTES_CONSUMED,
message="main output directory size",
bytes_consumed=bytes_consumed,
)
log_event(event)
event = StructuredLogEvent(
source="submitter",
category=EVENT_CATEGORY_RESOURCE_UTIL,
name=EVENT_NAME_SUBMIT_COMPLETED,
message="job submission completed",
num_jobs=self.get_num_jobs(),
)
log_event(event)
group = self._config.get_default_submission_group()
if group.submitter_params.generate_reports:
self.generate_reports(self._output,
group.submitter_params.resource_monitor_type)
cluster.mark_complete()
if cluster.config.pipeline_stage_num is not None:
# The pipeline directory must be the one above this one.
pipeline_dir = os.path.dirname(self._output)
next_stage = cluster.config.pipeline_stage_num + 1
cmd = (f"jade pipeline submit-next-stage {pipeline_dir} "
f"--stage-num={next_stage} "
f"--return-code={result.value}")
run_command(cmd)
return result
def test_demo_extension(test_data_dir):
"""Should create a config.json file"""
config_file = os.path.join(test_data_dir, "config.json")
if os.path.exists(config_file):
os.remove(config_file)
base = os.path.join(JADE_PATH, "extensions", "demo")
create_demo_config = os.path.join(base, "create_demo_config.sh")
create_merge_config = os.path.join(base, "create_merge_pred_gdp.py")
config_file = os.path.join(test_data_dir, "pipeline.json")
output = os.path.join(test_data_dir, "output")
if os.path.exists(output):
shutil.rmtree(output)
try:
cmd = (
f"jade pipeline create {create_demo_config} {create_merge_config} -c {config_file} -l"
)
returncode = run_command(cmd=cmd)
assert returncode == 0
assert os.path.exists(config_file)
returncode = run_command(
f"jade pipeline submit {config_file} -o {output}")
assert returncode == 0
output_stage1 = os.path.join(output, "output-stage1")
output_stage2 = os.path.join(output, "output-stage2")
assert os.path.exists(output)
assert os.path.exists(output_stage1)
assert os.path.exists(output_stage2)
job_outputs = os.path.join(output_stage1, "job-outputs")
for country in ("australia", "brazil", "united_states"):
results = os.listdir(os.path.join(job_outputs, country))
assert "result.csv" in results
assert "result.png" in results
assert "summary.toml" in results
pred_gdp_file = os.path.join(output_stage2, "pred_gdp.csv")
assert os.path.exists(pred_gdp_file)
df = pd.read_csv(pred_gdp_file)
assert "year" in df.columns
assert "brazil" in df.columns
assert "united_states" in df.columns
assert "australia" in df.columns
finally:
if os.path.exists(output):
shutil.rmtree(output)
if os.path.exists(config_file):
os.remove(config_file)
if os.path.exists(PRED_GDP_COMMANDS_FILE):
os.remove(PRED_GDP_COMMANDS_FILE)
def test_stats__bytes_consumed(example_output):
output = {}
ret = run_command(f"jade stats bytes-consumed -o {example_output}", output)
assert ret == 0
assert len(output["stdout"]) > 0
ret = run_command(f"jade stats bytes-consumed --no-human-readable -o {example_output}", output)
assert ret == 0
bytes_consumed = int(output["stdout"].strip())
assert bytes_consumed > 0
def test_stats__exec_time(example_output):
output = {}
ret = run_command(f"jade stats exec-time -o {example_output}", output)
assert ret == 0
assert len(output["stdout"]) > 0
ret = run_command(f"jade stats exec-time --no-human-readable -o {example_output}", output)
assert ret == 0
exec_time = float(output["stdout"].strip())
assert exec_time > 0
def test_config__show(cleanup):
ret = run_command(f"jade auto-config demo tests/data/demo -c {CONFIG1}")
assert ret == 0
assert os.path.exists(CONFIG1)
output = {}
ret = run_command(f"jade config show {CONFIG1}", output=output)
assert ret == 0
for country in ("australia", "brazil", "united_states"):
assert country in output["stdout"]
def test_config__filter_show_only(cleanup):
ret = run_command(f"jade auto-config demo tests/data/demo -c {CONFIG1}")
assert ret == 0
assert os.path.exists(CONFIG1)
output = {}
ret = run_command(f"jade config filter {CONFIG1} -f country brazil", output=output)
assert ret == 0
assert not os.path.exists(CONFIG2)
assert "brazil" in output["stdout"]
def test_config__assign_blocked_by(cleanup):
config = GenericCommandConfiguration()
base_cmd = "bash my_script.sh"
regular_job_names = []
for i in range(1, 4):
cmd = base_cmd + " " + str(i)
name = f"job_{i}"
job = GenericCommandParameters(
command=cmd,
name=name,
append_job_name=True,
append_output_dir=True,
)
config.add_job(job)
regular_job_names.append(name)
pp_name = "post_process"
post_process_job = GenericCommandParameters(
command="bash run_post_process.sh",
name=pp_name,
append_job_name=True,
append_output_dir=True,
)
config.add_job(post_process_job)
config_file = CONFIG1
config.dump(config_file, indent=2)
ret = run_command(
f"jade config assign-blocked-by {CONFIG1} 3 -o {CONFIG2}")
assert ret == 0
assert os.path.exists(CONFIG2)
config = load_data(CONFIG2)
assert sorted(config["jobs"][3]["blocked_by"]) == sorted(regular_job_names)
os.remove(CONFIG2)
ret = run_command(
f"jade config assign-blocked-by {CONFIG1} 3 1 2 -o {CONFIG2}")
assert ret == 0
assert os.path.exists(CONFIG2)
config = load_data(CONFIG2)
expected = [regular_job_names[1], regular_job_names[2]]
assert sorted(config["jobs"][3]["blocked_by"]) == sorted(expected)
# Include the pp job in blocking-job-indexes.
ret = run_command(
f"jade config assign-blocked-by {CONFIG1} 3 1 2 3 -o {CONFIG2}")
assert ret != 0
# Invalid job index
ret = run_command(
f"jade config assign-blocked-by {CONFIG1} 47 1 2 -o {CONFIG2}")
assert ret != 0
def test_config__filter_copy(cleanup):
ret = run_command(f"jade auto-config demo tests/data/demo -c {CONFIG1}")
assert ret == 0
assert os.path.exists(CONFIG1)
ret = run_command(f"jade config filter {CONFIG1} -o {CONFIG2}")
assert ret == 0
assert os.path.exists(CONFIG2)
config1 = load_data(CONFIG1)
config2 = load_data(CONFIG2)
assert config1 == config2
def test_config__filter_range(cleanup):
ret = run_command(f"jade auto-config demo tests/data/demo -c {CONFIG1}")
assert ret == 0
assert os.path.exists(CONFIG1)
ret = run_command(f"jade config filter {CONFIG1} -o {CONFIG2} 0 1")
assert ret == 0
assert os.path.exists(CONFIG2)
config1 = load_data(CONFIG1)
config2 = load_data(CONFIG2)
assert config2["jobs"] == [config1["jobs"][0], config1["jobs"][1]]
def test_collect_stats():
output_dir = os.path.join(tempfile.gettempdir(), "test-stats-output")
try:
ret = run_command(f"jade stats collect -i1 -o {output_dir} -d 1 -f")
assert ret == 0
cmd = f"jade stats show -o {output_dir} cpu disk mem net"
output = {}
ret = run_command(cmd, output=output)
assert ret == 0
for term in ("IOPS", "read_bytes", "bytes_recv", "idle"):
assert term in output["stdout"]
finally:
if os.path.exists(output_dir):
shutil.rmtree(output_dir)
def run_jobs(config_file, distributed_submitter, output, num_processes,
verbose):
"""Starts jobs on HPC."""
match = re.search(r"batch_(\d+)\.json", config_file)
assert match
batch_id = match.group(1)
os.makedirs(output, exist_ok=True)
mgr = JobRunner(config_file, output=output, batch_id=batch_id)
# Logging has to get enabled after the JobRunner is created because we need the node ID
# is what makes the file unique.
filename = os.path.join(output,
f"run_jobs_batch_{batch_id}_{mgr.node_id}.log")
level = logging.DEBUG if verbose else logging.INFO
setup_event_logging(mgr.event_filename)
logger = setup_logging(__name__,
filename,
file_level=level,
console_level=level,
mode="w")
logger.info(get_cli_string())
group = mgr.config.get_default_submission_group()
if group.submitter_params.node_setup_script:
cmd = f"{group.submitter_params.node_setup_script} {config_file} {output}"
ret = run_command(cmd)
if ret != 0:
logger.error("Failed to run node setup script %s: %s", cmd, ret)
sys.exit(ret)
status = mgr.run_jobs(distributed_submitter=distributed_submitter,
verbose=verbose,
num_processes=num_processes)
ret = status.value
if group.submitter_params.node_shutdown_script:
cmd = f"{group.submitter_params.node_shutdown_script} {config_file} {output}"
ret2 = run_command(cmd)
if ret2 != 0:
logger.error("Failed to run node shutdown script %s: %s", cmd,
ret2)
if status == Status.GOOD and distributed_submitter:
start = time.time()
_try_submit_jobs(output, verbose=verbose)
logger.info("try-submit-jobs took %s seconds", time.time() - start)
sys.exit(ret)
def _submit(self, verbose):
for stage in self._stages:
os.environ["JADE_PIPELINE_STAGE_ID"] = str(self._cur_stage_id)
stage_info = {
"stage_id":
self._cur_stage_id,
"output_directory":
self.get_stage_output_path(self._output, self._cur_stage_id)
}
self._status_info["stages"].append(stage_info)
self._dump_status()
self._run_auto_config(stage)
cmd = self._make_submit_cmd(stage, verbose)
start = time.time()
ret = run_command(cmd)
end = time.time()
exec_time = end - start
result = Result(str(self._cur_stage_id), ret, "finished",
exec_time, end)
self._status_info["stages"][-1]["result"] = serialize_result(
result)
self._dump_status()
if ret != 0:
raise ExecutionError(f"stage {self._cur_stage_id} failed")
self._cur_stage_id += 1
logger.info("Finished execution pipeline")
def generate_reports(directory):
"""Create reports summarizing the output results of a set of jobs.
Parameters
----------
directory : str
output directory
"""
commands = (
(f"jade show-results -o {directory}", "results.txt"),
(f"jade show-events -o {directory} --categories Error",
"errors.txt"),
(f"jade stats show -o {directory}", "stats.txt"),
)
reports = []
for cmd in commands:
output = {}
ret = run_command(cmd[0], output=output)
if ret != 0:
return ret
filename = os.path.join(directory, cmd[1])
with open(filename, "w") as f_out:
f_out.write(cmd[0] + "\n\n")
f_out.write(output["stdout"])
reports.append(filename)
logger.info("Generated reports %s.", " ".join(reports))
return 0
def check_status(self, name=None, job_id=None):
field_names = ("jobid", "name", "state")
cmd = f"squeue -u {self.USER} --Format \"{','.join(field_names)}\" -h"
if name is not None:
cmd += f" -n {name}"
elif job_id is not None:
cmd += f" -j {job_id}"
else:
# Mutual exclusivity should be handled in HpcManager.
assert False
output = {}
ret = run_command(cmd, output)
if ret != 0:
logger.error("Failed to run squeue command=[%s] ret=%s err=%s",
cmd, ret, output["stderr"])
raise ExecutionError(f"squeue command failed: {ret}")
stdout = output["stdout"]
logger.debug("squeue output: [%s]", stdout)
fields = stdout.split()
if not fields:
# No jobs are currently running.
return HpcJobInfo("", "", HpcJobStatus.NONE)
assert len(fields) == len(field_names)
job_info = HpcJobInfo(fields[0],
fields[1],
self._STATUSES.get(fields[2],
HpcJobStatus.UNKNOWN))
return job_info
def test_cancel_on_failure_detect_by_submitter(cleanup):
# HpcSubmitter handles the cancellation because the blocked job will be in the 2nd batch.
cmd = f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT} -n2 -b2"
ret = run_command(cmd)
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
summary = ResultsSummary(OUTPUT)
assert len(summary.get_successful_results()) == 1
assert len(summary.get_failed_results()) == 1
assert len(summary.get_canceled_results()) == 6
results = summary.get_results_by_type()
assert len(results["successful"]) == 1
assert len(results["failed"]) == 1
assert len(results["canceled"]) == 6
def test_run_generic_commands(generic_command_fixture):
commands = [
"ls .",
"ls invalid-file-path",
]
with open(TEST_FILENAME, "w") as f_out:
for command in commands:
f_out.write(command + "\n")
inputs = GenericCommandInputs(TEST_FILENAME)
config = GenericCommandConfiguration(job_inputs=inputs)
for job_param in inputs.iter_jobs():
config.add_job(job_param)
assert config.get_num_jobs() == 2
config.dump(CONFIG_FILE)
cmds = (
f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT} -p 0.1",
# Test with higher queue depth. This exercises the code paths but
# doesn't actually verify the functionality.
# The infrastructure to do that is currently lacking. TODO
f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT} -p 0.1 -q 32",
)
for cmd in cmds:
ret = run_command(cmd)
assert ret == 0
def test_run_command__stdout():
"""Should run a command as a subprocess"""
command = "echo 'Hello Disco'"
output = {}
ret = run_command(command, output)
assert ret == 0
assert "stdout" in output
assert "Hello Disco" in output["stdout"]
def test_cancel_on_failure_detect_by_runner(cleanup):
# JobRunner handles the cancellation in JobQueue because the blocked job is in the batch
# along with the blocking job.
cmd = f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT} -n2 -b8"
ret = run_command(cmd)
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
summary = ResultsSummary(OUTPUT)
assert len(summary.get_successful_results()) == 1
assert len(summary.get_failed_results()) == 1
assert len(summary.get_canceled_results()) == 6
results = summary.get_results_by_type()
assert len(results["successful"]) == 1
assert len(results["failed"]) == 1
assert len(results["canceled"]) == 6
def test_stats__show(example_output):
output = {}
ret = run_command(f"jade stats show -o {example_output}", output)
assert ret == 0
assert len(output["stdout"]) > 0
for text in ("Network stat", "Memory stat", "Disk stat", "CPU stat"):
assert text in output["stdout"]
def _try_submit_jobs(output, verbose):
try_submit_cmd = f"jade try-submit-jobs {output}"
if verbose:
try_submit_cmd += " --verbose"
ret = run_command(try_submit_cmd)
if ret != 0:
logger = logging.getLogger(__name__)
logger.error("Failed to run '%s' ret=%s", try_submit_cmd, ret)
def test_run_command():
"""Should run a command as a subprocess"""
command = "ls -l /dirnotexit"
output = {}
ret = run_command(command, output)
assert ret != 0
assert "stderr" in output
assert "No such file or directory" in output["stderr"]
def _qstat(self):
"""Run the PBS qstat command and return the stdout split to rows.
Returns
-------
qstat_rows : list | None
List of strings where each string is a row in the qstat printout.
Returns None if qstat is empty.
"""
cmd = "qstat -u {user}".format(user=self.USER)
output = {}
run_command(cmd, output)
if not output["stdout"]:
# No jobs are currently running.
return None
qstat_rows = output["stdout"].split("\n")
return qstat_rows
def submit(self, filename):
output = {}
ret = run_command("qsub {}".format(filename), output)
if ret == 0:
result = Status.GOOD
job_id = output["stdout"]
else:
result = Status.ERROR
job_id = None
return result, job_id, output["stderr"]
def test_stats__plot(example_output):
path = os.path.join(example_output, "stats")
try:
ret = run_command(f"jade stats plot -o {example_output}")
assert ret == 0
for stat in ("Cpu", "Disk", "Memory", "Network"):
filename = os.path.join(
path, stat + "StatsViewer__resource_monitor_batch_0.html")
assert os.path.exists(filename)
finally:
if os.path.exists(path):
shutil.rmtree(path)
def main():
status = load_data(os.environ["JADE_PIPELINE_STATUS_FILE"])
cur_stage = status["stages"][-1]
cur_stage_output = cur_stage["output_directory"]
previous_stage = status["stages"][-2]
previous_stage_output = previous_stage["output_directory"]
script = "jade/extensions/demo/merge_pred_gdp.py"
with open(PRED_GDP_COMMANDS_FILE, "w") as f_out:
cmd = f"python {script} run {previous_stage_output} {cur_stage_output}"
f_out.write(cmd + "\n")
cmd = "jade config create pred_gdp_commands.txt -c config-stage2.json"
sys.exit(run_command(cmd))
def _run_command(self, cmd):
orig = os.getcwd()
os.chdir(self._path)
try:
output = {}
ret = run_command(cmd, output=output)
if ret != 0:
raise ExecutionError(
f"[{cmd}] failed: {ret}: {output['stderr']}")
return output["stdout"].strip()
finally:
os.chdir(orig)
def main():
config = PipelineConfig(
**load_data(os.environ["JADE_PIPELINE_STATUS_FILE"]))
cur_stage = config.stages[-1]
cur_stage_output = cur_stage.path
previous_stage = config.stages[-2]
previous_stage_output = previous_stage.path
script = "jade/extensions/demo/merge_pred_gdp.py"
with open(PRED_GDP_COMMANDS_FILE, "w") as f_out:
cmd = f"python {script} run {previous_stage_output} {cur_stage_output}"
f_out.write(cmd + "\n")
cmd = "jade config create pred_gdp_commands.txt -c config-stage2.json"
sys.exit(run_command(cmd))
def test_try_add_blocked_jobs(cleanup):
num_commands = 5
commands = ['echo "hello world"'] * num_commands
with open(TEST_FILENAME, "w") as f_out:
for command in commands:
f_out.write(command + "\n")
inputs = GenericCommandInputs(TEST_FILENAME)
config = GenericCommandConfiguration(job_inputs=inputs)
jobs = list(inputs.iter_jobs())
for i, job_param in enumerate(jobs):
if i == num_commands - 1:
job_param.blocked_by = set([1, 2, 3, 4])
config.add_job(job_param)
config.dump(CONFIG_FILE)
for option in ("--try-add-blocked-jobs", "--no-try-add-blocked-jobs"):
cmd = f"{SUBMIT_JOBS} {CONFIG_FILE} --output={OUTPUT} -h {FAKE_HPC_CONFIG} -p 0.1 {option}"
ret = run_command(cmd)
assert ret == 0
ret = run_command(f"{WAIT} --output={OUTPUT} -p 0.01")
assert ret == 0
events_file = os.path.join(OUTPUT, "submit_jobs_events.log")
events_summary = EventsSummary(OUTPUT, preload=True)
submit_events = events_summary.list_events(EVENT_NAME_HPC_SUBMIT)
if option == "--try-add-blocked-jobs":
assert len(submit_events) == 1
event = submit_events[0]
assert event.data["batch_size"] == num_commands
shutil.rmtree(OUTPUT)
else:
assert len(submit_events) == 2
event1 = submit_events[0]
event2 = submit_events[1]
assert event1.data["batch_size"] == num_commands - 1
assert event2.data["batch_size"] == 1
