def run_with_captured_reboots(config, platform, monkeypatch):
"""Runs a session to completion using exceptions to capture reboots
Returns the number of reboots and the last scheduler"""
no_envlogs(monkeypatch)
make_reboot_raise(monkeypatch)
krun.util.assign_platform(config, platform)
reboots = 0
manifest = emulate_first_reboot(platform, config)
if manifest.num_execs_left == 0:
sched = ExecutionScheduler(config,
platform.mailer,
platform,
dry_run=True)
return reboots, sched
reboots += 1
# Run the main benchmarking loop
while True:
sched = ExecutionScheduler(config,
platform.mailer,
platform,
dry_run=True)
try:
sched.run()
except _TestReboot:
reboots += 1
else:
# normal exit() from run -- schedule finished
break
return reboots, sched
def test_num_emails_sent_persists0001(self, monkeypatch, mock_platform):
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config, monkeypatch)
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
sched.mailer.recipients = ["noone@localhost"]
assert sched.manifest.num_mails_sent == 0
sched.mailer.send("subject", "body", manifest=sched.manifest)
assert sched.manifest.num_mails_sent == 1
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# suppose a reboot happened now
del sched
del config
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
assert sched.manifest.num_mails_sent == 1
os.unlink(sched.manifest.path)
def test_pre_and_post_cmds0001(self, monkeypatch, mock_platform):
cap_cmds = []
def dummy_run_shell_cmd(cmd, failure_fatal=False, extra_env=None):
cap_cmds.append(cmd)
return "", "", 0
monkeypatch.setattr(krun.util, "run_shell_cmd", dummy_run_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
config.PRE_EXECUTION_CMDS = ["pre1", "pre2"]
config.POST_EXECUTION_CMDS = ["post1", "post2"]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
sched.run()
expect = ["pre1", "pre2", "post1", "post2"] * 8
assert cap_cmds == expect
def run_with_captured_reboots(config, platform, monkeypatch):
"""Runs a session to completion using exceptions to capture reboots
Returns the number of reboots and the last scheduler"""
no_envlogs(monkeypatch)
make_reboot_raise(monkeypatch)
krun.util.assign_platform(config, platform)
reboots = 0
manifest = emulate_first_reboot(platform, config, monkeypatch)
if manifest.num_execs_left == 0:
sched = ExecutionScheduler(config, platform.mailer, platform,
dry_run=True)
return reboots, sched
reboots += 1
# Run the main benchmarking loop
while True:
sched = ExecutionScheduler(config, platform.mailer, platform, dry_run=True)
try:
sched.run()
except _TestReboot:
reboots += 1
else:
# normal exit() from run -- schedule finished
break
return reboots, sched
def test_num_emails_sent_persists0001(self, monkeypatch, mock_platform,
no_results_instantiation_check):
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
sched.mailer.recipients = ["noone@localhost"]
assert sched.manifest.num_mails_sent == 0
sched.mailer.send("subject", "body", manifest=sched.manifest)
assert sched.manifest.num_mails_sent == 1
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# suppose a reboot happened now
del sched
del config
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
assert sched.manifest.num_mails_sent == 1
os.unlink(sched.manifest.path)
def test_audit_differs0001(self, monkeypatch, mock_platform, caplog):
"""Check that if the audit differs, a crash occurs"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config, monkeypatch)
results_path = config.results_filename()
# mutate the audit, so it won't match later
results = Results(config, mock_platform, results_file=results_path)
results.audit._audit["wibble"] = "wut"
results.write_to_file()
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
with pytest.raises(krun.util.FatalKrunError):
sched.run()
expect = "This is only valid if the machine you are using is identical"
assert expect in caplog.text()
os.unlink(sched.manifest.path)
os.unlink(results_path)
def test_audit_differs0001(self, monkeypatch, mock_platform, caplog,
no_results_instantiation_check):
"""Check that if the audit differs, a crash occurs"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config)
results_path = config.results_filename()
# mutate the audit, so it won't match later
results = Results(config, mock_platform, results_file=results_path)
results.audit._audit["wibble"] = "wut"
results.write_to_file()
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
with pytest.raises(krun.util.FatalKrunError):
sched.run()
expect = "This is only valid if the machine you are using is identical"
assert expect in caplog.text
os.unlink(sched.manifest.path)
os.unlink(results_path)
def test_run_schedule_reboot(self, monkeypatch, mock_platform):
def dummy_shell_cmd(text):
pass
def dummy_execv(text, lst):
pass
monkeypatch.setattr(os, "execv", dummy_execv)
monkeypatch.setattr(subprocess, "call", dummy_shell_cmd)
monkeypatch.setattr(krun.util, "run_shell_cmd", dummy_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
resume=False,
reboot=True,
dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
with pytest.raises(AssertionError):
sched.run()
assert len(sched) == 7
os.unlink(os.path.join(TEST_DIR, "example_results.json.bz2"))
def test_pre_and_post_cmds0002(self, monkeypatch, mock_platform):
"""Check that the pre/post commands use a shell and don't just exec(3)"""
config = Config(os.path.join(TEST_DIR, "example.krun"))
tmp_file = os.path.join(TEST_DIR, "prepost.txt")
# commands use shell syntax
config.PRE_EXECUTION_CMDS = ["echo 'pre' > %s" % tmp_file]
config.POST_EXECUTION_CMDS = ["echo 'post' >> %s" % tmp_file]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
sched.run()
with open(tmp_file) as fh:
got = fh.read()
os.unlink(tmp_file)
assert got == "pre\npost\n"
def test_pre_and_post_cmds0002(self, monkeypatch, mock_platform):
"""Check that the pre/post commands use a shell and don't just exec(3)"""
config = Config(os.path.join(TEST_DIR, "example.krun"))
tmp_file = os.path.join(TEST_DIR, "prepost.txt")
# commands use shell syntax
config.PRE_EXECUTION_CMDS = ["echo 'pre' > %s" % tmp_file]
config.POST_EXECUTION_CMDS = ["echo 'post' >> %s" % tmp_file]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform, resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
sched.run()
with open(tmp_file) as fh:
got = fh.read()
os.unlink(tmp_file)
assert got == "pre\npost\n"
def test_post_exec_cmds0002(self, monkeypatch, mock_platform):
config = Config(os.path.join(TEST_DIR, "example.krun"))
path = os.path.join(TEST_DIR, "shell-out")
cmd = "echo ${KRUN_RESULTS_FILE}:${KRUN_LOG_FILE} > %s" % path
config.POST_EXECUTION_CMDS = [cmd]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform, resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
sched.run()
with open(path) as fh:
got = fh.read().strip()
os.unlink(path)
elems = got.split(":")
assert elems[0].endswith(".json.bz2")
assert elems[1].endswith(".log")
def test_pre_and_post_cmds0001(self, monkeypatch, mock_platform):
cap_cmds = []
def dummy_run_shell_cmd(cmd, failure_fatal=False, extra_env=None):
cap_cmds.append(cmd)
return "", "", 0
monkeypatch.setattr(krun.util, "run_shell_cmd", dummy_run_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
config.PRE_EXECUTION_CMDS = ["pre1", "pre2"]
config.POST_EXECUTION_CMDS = ["post1", "post2"]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform, resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
sched.run()
expect = ["pre1", "pre2", "post1", "post2"] * 8
assert cap_cmds == expect
def test_boot_loop0001(self, monkeypatch, mock_platform, caplog):
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config, monkeypatch)
# Simulate a boot loop
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
sched.manifest.num_reboots = 9998 # way too many
sched.manifest.update_num_reboots() # increments and writes out file
with pytest.raises(krun.util.FatalKrunError):
sched.run()
expect = ("HALTING now to prevent an infinite reboot loop: "
"INVARIANT num_reboots <= num_jobs violated. Krun was about "
"to execute reboot number: 10000. 1 jobs have been "
"completed, 7 are left to go.")
assert expect in caplog.text()
os.unlink(config.results_filename())
os.unlink(sched.manifest.path)
def test_post_exec_cmds0002(self, monkeypatch, mock_platform):
config = Config(os.path.join(TEST_DIR, "example.krun"))
path = os.path.join(TEST_DIR, "shell-out")
cmd = "echo ${KRUN_RESULTS_FILE}:${KRUN_LOG_FILE} > %s" % path
config.POST_EXECUTION_CMDS = [cmd]
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
resume=False,
dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
sched.run()
with open(path) as fh:
got = fh.read().strip()
os.unlink(path)
elems = got.split(":")
assert elems[0].endswith(".json.bz2")
assert elems[1].endswith(".log")
def test_boot_loop0001(self, monkeypatch, mock_platform, caplog,
no_results_instantiation_check):
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config)
# Simulate a boot loop
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
sched.manifest.num_reboots = 9998 # way too many
sched.manifest.update_num_reboots() # increments and writes out file
with pytest.raises(krun.util.FatalKrunError):
sched.run()
expect = ("HALTING now to prevent an infinite reboot loop: "
"INVARIANT num_reboots <= num_jobs violated. Krun was about "
"to execute reboot number: 10000. 1 jobs have been "
"completed, 7 are left to go.")
assert expect in caplog.text
os.unlink(config.results_filename())
os.unlink(sched.manifest.path)
def test_run_schedule(self, monkeypatch, mock_platform):
json_file = os.path.join(TEST_DIR, "example_results.json.bz2")
def dummy_shell_cmd(text):
pass
monkeypatch.setattr(subprocess, 'call', dummy_shell_cmd)
monkeypatch.setattr(krun.util, 'run_shell_cmd', dummy_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
resume=False,
reboot=False,
dry_run=True,
started_by_init=False)
sched.build_schedule()
assert len(sched) == 8
sched.run()
assert len(sched) == 0
results = Results(Config(os.path.join(TEST_DIR, "example.krun")),
mock_platform,
results_file=json_file)
for k, execs in results.data.iteritems():
assert type(execs) is list
for one_exec in execs:
assert type(one_exec) is list
assert all([type(x) is float for x in one_exec])
for k, execs in results.eta_estimates.iteritems():
assert type(execs) is list
assert all([type(x) is float for x in execs])
assert type(results.starting_temperatures) is dict
assert type(results.reboots) is int
assert type(results.audit) is type(Audit(dict()))
assert type(results.config) is type(Config())
assert type(results.error_flag) is bool
os.unlink(json_file)
def test_error_flag_persists0002(self, monkeypatch, mock_platform,
no_results_instantiation_check):
"""Check a changed dmesg will correctly set the error flag"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
# pretend dmesg changes
def dummy_check_for_dmesg_changes(self):
return True
monkeypatch.setattr(mock_platform, 'check_dmesg_for_changes',
dummy_check_for_dmesg_changes)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config)
results_path = config.results_filename()
# To start, the error flag is not set
results = Results(config, mock_platform, results_file=results_path)
assert not results.error_flag
# run an execution where the dmesg changes
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# reload results and check the error flag is now set
results = Results(config, mock_platform, results_file=results_path)
assert results.error_flag
os.unlink(sched.manifest.path)
os.unlink(results_path)
def test_error_flag_persists0001(self, monkeypatch, mock_platform,
no_results_instantiation_check):
"""Check a failing exec will correctly set the error flag"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
# pretend exec fails
def dummy_job_run(self, mailer, dry):
measurements = self.make_empty_measurement()
return measurements, None, 'E' # measurements, instr_data, flag
monkeypatch.setattr(ExecutionJob, 'run', dummy_job_run)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config)
results_path = config.results_filename()
# To start, the error flag is not set
results = Results(config, mock_platform, results_file=results_path)
assert not results.error_flag
# run a (failing) execution, which will dump the results file
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform,
dry_run=True)
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# reload results and check the error flag is now set
results = Results(config, mock_platform, results_file=results_path)
assert results.error_flag
os.unlink(sched.manifest.path)
os.unlink(results_path)
def test_run_schedule_reboot(self, monkeypatch, mock_platform):
def dummy_shell_cmd(text):
pass
def dummy_execv(text, lst):
pass
monkeypatch.setattr(os, "execv", dummy_execv)
monkeypatch.setattr(subprocess, "call", dummy_shell_cmd)
monkeypatch.setattr(krun.util, "run_shell_cmd", dummy_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform, resume=False,
reboot=True, dry_run=True,
started_by_init=True)
sched.build_schedule()
assert len(sched) == 8
with pytest.raises(AssertionError):
sched.run()
assert len(sched) == 7
os.unlink(os.path.join(TEST_DIR, "example_results.json.bz2"))
def test_run_schedule(self, monkeypatch, mock_platform):
json_file = os.path.join(TEST_DIR, "example_results.json.bz2")
def dummy_shell_cmd(text):
pass
monkeypatch.setattr(subprocess, 'call', dummy_shell_cmd)
monkeypatch.setattr(krun.util, 'run_shell_cmd', dummy_shell_cmd)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
sched = ExecutionScheduler(config,
mock_platform.mailer,
mock_platform, resume=False,
reboot=False, dry_run=True,
started_by_init=False)
sched.build_schedule()
assert len(sched) == 8
sched.run()
assert len(sched) == 0
results = Results(Config(os.path.join(TEST_DIR, "example.krun")),
mock_platform, results_file=json_file)
for k, execs in results.data.iteritems():
assert type(execs) is list
for one_exec in execs:
assert type(one_exec) is list
assert all([type(x) is float for x in one_exec])
for k, execs in results.eta_estimates.iteritems():
assert type(execs) is list
assert all([type(x) is float for x in execs])
assert type(results.starting_temperatures) is dict
assert type(results.reboots) is int
assert type(results.audit) is type(Audit(dict()))
assert type(results.config) is type(Config())
assert type(results.error_flag) is bool
os.unlink(json_file)
def test_error_flag_persists0001(self, monkeypatch, mock_platform):
"""Check a failing exec will correctly set the error flag"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
# pretend exec fails
def dummy_job_run(self, mailer, dry):
measurements = self.make_empty_measurement()
return measurements, None, 'E' # measurements, instr_data, flag
monkeypatch.setattr(ExecutionJob, 'run', dummy_job_run)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config, monkeypatch)
results_path = config.results_filename()
# To start, the error flag is not set
results = Results(config, mock_platform, results_file=results_path)
assert not results.error_flag
# run a (failing) execution, which will dump the results file
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# reload results and check the error flag is now set
results = Results(config, mock_platform, results_file=results_path)
assert results.error_flag
os.unlink(sched.manifest.path)
os.unlink(results_path)
def test_error_flag_persists0002(self, monkeypatch, mock_platform):
"""Check a changed dmesg will correctly set the error flag"""
make_reboot_raise(monkeypatch)
no_envlogs(monkeypatch)
# pretend dmesg changes
def dummy_check_for_dmesg_changes(self):
return True
monkeypatch.setattr(mock_platform, 'check_dmesg_for_changes',
dummy_check_for_dmesg_changes)
config = Config(os.path.join(TEST_DIR, "example.krun"))
krun.util.assign_platform(config, mock_platform)
emulate_first_reboot(mock_platform, config, monkeypatch)
results_path = config.results_filename()
# To start, the error flag is not set
results = Results(config, mock_platform, results_file=results_path)
assert not results.error_flag
# run an execution where the dmesg changes
sched = ExecutionScheduler(config, mock_platform.mailer, mock_platform,
dry_run=True)
try:
sched.run()
except _TestReboot:
pass
else:
assert False
# reload results and check the error flag is now set
results = Results(config, mock_platform, results_file=results_path)
assert results.error_flag
os.unlink(sched.manifest.path)
os.unlink(results_path)
def inner_main(mailer, on_first_invocation, config, args):
out_file = config.results_filename()
out_file_exists = os.path.exists(out_file)
instr_dir = util.get_instr_json_dir(config)
instr_dir_exists = os.path.exists(instr_dir)
envlog_dir = util.get_envlog_dir(config)
envlog_dir_exists = os.path.exists(envlog_dir)
if out_file_exists and not os.path.isfile(out_file):
util.fatal(
"Output file '%s' exists but is not a regular file" % out_file)
if out_file_exists and on_first_invocation:
util.fatal("Output results file '%s' already exists. "
"Move the file away before running Krun." % out_file)
if instr_dir_exists and on_first_invocation:
util.fatal("Instrumentation dir '%s' exists." % instr_dir)
if envlog_dir_exists and on_first_invocation:
util.fatal("Env log dir '%s' exists." % envlog_dir)
if not out_file_exists and not on_first_invocation:
util.fatal("No results file to resume. Expected '%s'" % out_file)
# Initialise platform instance and assign to VM defs.
# This needs to be done early, so VM sanity checks can run.
platform = detect_platform(mailer, config)
platform.quick_mode = args.quick
platform.no_user_change = args.no_user_change
platform.no_tickless_check = args.no_tickless_check
platform.no_pstate_check = args.no_pstate_check
platform.hardware_reboots = args.hardware_reboots
# Create the instrumentation directory if required
if on_first_invocation:
# We only want make a dir if >=1 VM is in instrumentation mode.
for vm in config.VMS.itervalues():
if vm['vm_def'].instrument:
util.make_instr_dir(config)
break
debug("Checking platform preliminaries")
platform.check_preliminaries()
# Make a bit of noise if this is a virtualised environment
if platform.is_virtual():
warn("This appears to be a virtualised host. The results will be flawed. "
"Use bare-metal for reliable results!")
platform.collect_audit()
# At this point the config file is OK, and on-disk state is consistent,
# so let's daemonise (if requested).
if args.daemonise:
util.daemonise()
if not on_first_invocation:
# output file must exist, due to check above
assert(out_file_exists)
debug("Using pre-recorded initial temperature readings")
manifest = ManifestManager(config, platform)
platform_temps = {}
for sensor, tup in manifest.starting_temperatures.iteritems():
platform_temps[sensor] = tup[1]
platform.starting_temperatures = platform_temps
else:
manifest = ManifestManager(config, platform, new_file=True)
if manifest.num_execs_left == 0:
# No executions, or all skipped
fatal("Empty schedule!")
try:
info(("Wait %s secs to allow system to cool prior to "
"collecting initial temperature readings") %
config.TEMP_READ_PAUSE)
# This part is wrapped in hooks, so that if daemons or networking are
# taken down for process executions, then the initial temperature
# reading gets the same treatment.
util.run_shell_cmd_list(config.PRE_EXECUTION_CMDS,)
platform.sleep(config.TEMP_READ_PAUSE)
debug("Taking fresh initial temperature readings")
platform.starting_temperatures = platform.take_temperature_readings()
manifest.set_starting_temperatures(platform.starting_temperatures)
# Write out an empty results file. After the initial reboot Krun
# will expect this to exist.
Results.ok_to_instantiate = True
results = Results(config, platform)
results.write_to_file()
except:
raise
finally:
util.run_shell_cmd_list(config.POST_EXECUTION_CMDS,)
log_path = config.log_filename(resume=False)
util.log_and_mail(mailer, debug,
"Benchmarking started",
"Benchmarking started.\nLogging to %s" %
log_path, bypass_limiter=True)
util.reboot(manifest, platform)
# Assign platform to VM defs -- needs to happen early for sanity checks
util.assign_platform(config, platform)
sanity_checks(config, platform)
# Build job queue -- each job is an execution
sched = ExecutionScheduler(config,
mailer,
platform,
dry_run=args.dry_run)
sched.run()
def inner_main(mailer, config, args):
out_file = config.results_filename()
out_file_exists = os.path.exists(out_file)
if out_file_exists and not os.path.isfile(out_file):
util.fatal(
"Output file '%s' exists but is not a regular file" % out_file)
if out_file_exists and not args.resume:
util.fatal("Output file '%s' already exists. "
"Either resume the session (--resume) or "
"move the file away" % out_file)
if not out_file_exists and args.resume:
util.fatal("No results file to resume. Expected '%s'" % out_file)
if args.started_by_init and not args.reboot:
util.fatal("--started-by-init makes no sense without --reboot")
if args.started_by_init and not args.resume:
util.fatal("--started-by-init makes no sense without --resume")
if args.develop:
warn("Developer mode enabled. Results will not be reliable.")
# Initialise platform instance and assign to VM defs.
# This needs to be done early, so VM sanity checks can run.
platform = detect_platform(mailer)
if not args.develop:
debug("Checking platform preliminaries")
platform.check_preliminaries()
else:
# Needed to skip the use of certain tools and techniques.
# E.g. switching user.
warn("Not checking platform prerequisites due to developer mode")
platform.developer_mode = True
platform.collect_audit()
# If the user has asked for resume-mode, the current platform must
# be an identical machine to the current one.
error_msg = ("You have asked Krun to resume an interrupted benchmark. " +
"This is only valid if the machine you are using is " +
"identical to the one on which the last results were " +
"gathered, which is not the case.")
current = None
if args.resume:
# output file must exist, due to check above
assert(out_file_exists)
current = Results(config, platform, results_file=out_file)
from krun.audit import Audit
if not Audit(platform.audit) == current.audit:
util.fatal(error_msg)
debug("Using pre-recorded initial temperature readings")
platform.starting_temperatures = current.starting_temperatures
else:
# Touch the config file to update its mtime. This is required
# by resume-mode which uses the mtime to determine the name of
# the log file, should this benchmark be resumed.
_, _, rc = util.run_shell_cmd("touch " + args.filename)
if rc != 0:
util.fatal("Could not touch config file: " + args.filename)
info(("Wait %s secs to allow system to cool prior to "
"collecting initial temperature readings") %
config.TEMP_READ_PAUSE)
if args.develop or args.dry_run:
info("SIMULATED: time.sleep(%s)" % config.TEMP_READ_PAUSE)
else:
time.sleep(config.TEMP_READ_PAUSE)
debug("Taking fresh initial temperature readings")
platform.starting_temperatures = platform.take_temperature_readings()
# Assign platform to VM defs -- needs to happen early for sanity checks
util.assign_platform(config, platform)
sanity_checks(config, platform)
# Build job queue -- each job is an execution
sched = ExecutionScheduler(config,
mailer,
platform,
resume=args.resume,
reboot=args.reboot,
dry_run=args.dry_run,
started_by_init=args.started_by_init)
sched.build_schedule()
sched.run()
def inner_main(mailer, on_first_invocation, config, args):
out_file = config.results_filename()
out_file_exists = os.path.exists(out_file)
instr_dir = util.get_instr_json_dir(config)
instr_dir_exists = os.path.exists(instr_dir)
envlog_dir = util.get_envlog_dir(config)
envlog_dir_exists = os.path.exists(envlog_dir)
if out_file_exists and not os.path.isfile(out_file):
util.fatal("Output file '%s' exists but is not a regular file" %
out_file)
if out_file_exists and on_first_invocation:
util.fatal("Output results file '%s' already exists. "
"Move the file away before running Krun." % out_file)
if instr_dir_exists and on_first_invocation:
util.fatal("Instrumentation dir '%s' exists." % instr_dir)
if envlog_dir_exists and on_first_invocation:
util.fatal("Env log dir '%s' exists." % envlog_dir)
if not out_file_exists and not on_first_invocation:
util.fatal("No results file to resume. Expected '%s'" % out_file)
# Initialise platform instance and assign to VM defs.
# This needs to be done early, so VM sanity checks can run.
platform = detect_platform(mailer, config)
platform.quick_mode = args.quick
platform.no_user_change = args.no_user_change
platform.no_tickless_check = args.no_tickless_check
platform.no_pstate_check = args.no_pstate_check
platform.hardware_reboots = args.hardware_reboots
# Create the instrumentation directory if required
if on_first_invocation:
# We only want make a dir if >=1 VM is in instrumentation mode.
for vm in config.VMS.itervalues():
if vm['vm_def'].instrument:
util.make_instr_dir(config)
break
debug("Checking platform preliminaries")
platform.check_preliminaries()
# Make a bit of noise if this is a virtualised environment
if platform.is_virtual():
warn(
"This appears to be a virtualised host. The results will be flawed. "
"Use bare-metal for reliable results!")
platform.collect_audit()
# At this point the config file is OK, and on-disk state is consistent,
# so let's daemonise (if requested).
if args.daemonise:
util.daemonise()
if not on_first_invocation:
# output file must exist, due to check above
assert (out_file_exists)
debug("Using pre-recorded initial temperature readings")
manifest = ManifestManager(config, platform)
platform_temps = {}
for sensor, tup in manifest.starting_temperatures.iteritems():
platform_temps[sensor] = tup[1]
platform.starting_temperatures = platform_temps
else:
manifest = ManifestManager(config, platform, new_file=True)
if manifest.num_execs_left == 0:
# No executions, or all skipped
fatal("Empty schedule!")
try:
info(("Wait %s secs to allow system to cool prior to "
"collecting initial temperature readings") %
config.TEMP_READ_PAUSE)
# This part is wrapped in hooks, so that if daemons or networking are
# taken down for process executions, then the initial temperature
# reading gets the same treatment.
util.run_shell_cmd_list(config.PRE_EXECUTION_CMDS, )
platform.sleep(config.TEMP_READ_PAUSE)
debug("Taking fresh initial temperature readings")
platform.starting_temperatures = platform.take_temperature_readings(
)
manifest.set_starting_temperatures(platform.starting_temperatures)
# Write out an empty results file. After the initial reboot Krun
# will expect this to exist.
Results.ok_to_instantiate = True
results = Results(config, platform)
results.write_to_file()
except:
raise
finally:
util.run_shell_cmd_list(config.POST_EXECUTION_CMDS, )
log_path = config.log_filename(resume=False)
util.log_and_mail(mailer,
debug,
"Benchmarking started",
"Benchmarking started.\nLogging to %s" % log_path,
bypass_limiter=True)
util.reboot(manifest, platform)
# Assign platform to VM defs -- needs to happen early for sanity checks
util.assign_platform(config, platform)
sanity_checks(config, platform)
# Build job queue -- each job is an execution
sched = ExecutionScheduler(config, mailer, platform, dry_run=args.dry_run)
sched.run()