def _main(config_dir_path: Parameter.REQUIRED,
*,
no_checks=False,
no_debugging=False,
verbose=False):
"""
Run the diagnostic daemon.
:param config_dir_path: Configuration directory to scan for Ambassador YAML files
:param no_checks: If True, don't do Envoy-cluster health checking
:param no_debugging: If True, don't run Flask in debug mode
:param verbose: If True, be more verbose
"""
app.estats = EnvoyStats()
app.health_checks = False
app.debugging = not no_debugging
# This feels like overkill.
app._logger = logging.getLogger(app.logger_name)
app.logger.setLevel(logging.INFO)
if app.debugging or verbose:
app.logger.setLevel(logging.DEBUG)
logging.getLogger().setLevel(logging.DEBUG)
if not no_checks:
app.health_checks = True
app.logger.debug("Starting periodic updates")
app.stats_updater = PeriodicTrigger(app.estats.update, period=5)
app.config_dir_prefix = config_dir_path
app.run(host='0.0.0.0', port=aconf(app).diag_port(), debug=app.debugging)
def load(self):
if self.application.health_checks:
self.application.logger.info("Starting periodic updates")
self.application.stats_updater = PeriodicTrigger(
self.application.estats.update, period=5)
return self.application
def _load_ir(self, aconf: Config, fetcher: ResourceFetcher,
secret_reader: Callable[['IRTLSContext', str, str],
SavedSecret], snapshot: str) -> None:
aconf.load_all(fetcher.sorted())
aconf_path = os.path.join(app.snapshot_path,
"aconf-%s.json" % snapshot)
open(aconf_path, "w").write(aconf.as_json())
ir = IR(aconf, secret_reader=secret_reader)
ir_path = os.path.join(app.snapshot_path, "ir-%s.json" % snapshot)
open(ir_path, "w").write(ir.as_json())
check_scout(app, "update", ir)
econf = EnvoyConfig.generate(ir, "V2")
diag = Diagnostics(ir, econf)
bootstrap_config, ads_config = econf.split_config()
if not self.validate_envoy_config(config=ads_config):
self.logger.info(
"no updates were performed due to invalid envoy configuration, continuing with current configuration..."
)
return
self.logger.info("saving Envoy configuration for snapshot %s" %
snapshot)
with open(app.bootstrap_path, "w") as output:
output.write(json.dumps(bootstrap_config, sort_keys=True,
indent=4))
with open(app.ads_path, "w") as output:
output.write(json.dumps(ads_config, sort_keys=True, indent=4))
app.aconf = aconf
app.ir = ir
app.econf = econf
app.diag = diag
if app.kick:
self.logger.info("running '%s'" % app.kick)
os.system(app.kick)
elif app.ambex_pid != 0:
self.logger.info("notifying PID %d ambex" % app.ambex_pid)
os.kill(app.ambex_pid, signal.SIGHUP)
self.logger.info("configuration updated")
if app.health_checks and not app.stats_updater:
app.logger.info("starting Envoy status updater")
app.stats_updater = PeriodicTrigger(app.watcher.update_estats,
period=5)
def load(self):
# This is a little weird, but whatever.
self.application.watcher = AmbassadorEventWatcher(self.application)
self.application.watcher.start()
if self.application.health_checks:
self.application.logger.info("Starting periodic updates")
self.application.stats_updater = PeriodicTrigger(
self.application.watcher.update_estats, period=5)
return self.application
def run(self):
self.logger.info("starting Scout checker")
self.app.scout_checker = PeriodicTrigger(
lambda: self.check_scout("checkin"), period=86400) # Yup, one day.
self.logger.info("starting event watcher")
while True:
cmd, arg, rqueue = self.events.get()
# self.logger.info("EVENT: %s" % cmd)
if cmd == 'ESTATS':
# self.logger.info("updating estats")
try:
self._respond(rqueue, 200, 'updating')
self.app.estats.update()
except Exception as e:
self.logger.error("could not update estats: %s" % e)
self.logger.exception(e)
elif cmd == 'CONFIG_FS':
try:
self.load_config_fs(rqueue, arg)
except Exception as e:
self.logger.error("could not reconfigure: %s" % e)
self.logger.exception(e)
elif cmd == 'CONFIG':
version, url = arg
try:
if version == 'kw':
self.load_config_kubewatch(rqueue, url)
elif version == 'watt':
self.load_config_watt(rqueue, url)
else:
raise RuntimeError("config from %s not supported" %
version)
except Exception as e:
self.logger.error("could not reconfigure: %s" % e)
self.logger.exception(e)
elif cmd == 'SCOUT':
try:
self._respond(rqueue, 200, 'checking Scout')
self.check_scout(*arg)
except Exception as e:
self.logger.error("could not reconfigure: %s" % e)
self.logger.exception(e)
else:
self.logger.error("unknown event type: '%s' '%s'" % (cmd, arg))
def _load_ir(self, rqueue: queue.Queue, aconf: Config,
fetcher: ResourceFetcher, secret_handler: SecretHandler,
snapshot: str) -> None:
aconf.load_all(fetcher.sorted())
aconf_path = os.path.join(app.snapshot_path, "aconf-tmp.json")
open(aconf_path, "w").write(aconf.as_json())
ir = IR(aconf, secret_handler=secret_handler)
ir_path = os.path.join(app.snapshot_path, "ir-tmp.json")
open(ir_path, "w").write(ir.as_json())
econf = EnvoyConfig.generate(ir, "V2")
diag = Diagnostics(ir, econf)
bootstrap_config, ads_config = econf.split_config()
if not self.validate_envoy_config(config=ads_config):
self.logger.info(
"no updates were performed due to invalid envoy configuration, continuing with current configuration..."
)
app.check_scout("attempted bad update")
self._respond(
rqueue, 500,
'ignoring: invalid Envoy configuration in snapshot %s' %
snapshot)
return
snapcount = int(os.environ.get('AMBASSADOR_SNAPSHOT_COUNT', "4"))
snaplist: List[Tuple[str, str]] = []
if snapcount > 0:
self.logger.debug("rotating snapshots for snapshot %s" % snapshot)
# If snapcount is 4, this range statement becomes range(-4, -1)
# which gives [ -4, -3, -2 ], which the list comprehension turns
# into [ ( "-3", "-4" ), ( "-2", "-3" ), ( "-1", "-2" ) ]...
# which is the list of suffixes to rename to rotate the snapshots.
snaplist += [(str(x + 1), str(x))
for x in range(-1 * snapcount, -1)]
# After dealing with that, we need to rotate the current file into -1.
snaplist.append(('', '-1'))
# Whether or not we do any rotation, we need to cycle in the '-tmp' file.
snaplist.append(('-tmp', ''))
for from_suffix, to_suffix in snaplist:
for fmt in [
"aconf{}.json", "econf{}.json", "ir{}.json",
"snapshot{}.yaml"
]:
from_path = os.path.join(app.snapshot_path,
fmt.format(from_suffix))
to_path = os.path.join(app.snapshot_path,
fmt.format(to_suffix))
try:
self.logger.debug("rotate: %s -> %s" %
(from_path, to_path))
os.rename(from_path, to_path)
except IOError as e:
self.logger.debug("skip %s -> %s: %s" %
(from_path, to_path, e))
pass
except Exception as e:
self.logger.debug("could not rename %s -> %s: %s" %
(from_path, to_path, e))
self.logger.info("saving Envoy configuration for snapshot %s" %
snapshot)
with open(app.bootstrap_path, "w") as output:
output.write(json.dumps(bootstrap_config, sort_keys=True,
indent=4))
with open(app.ads_path, "w") as output:
output.write(json.dumps(ads_config, sort_keys=True, indent=4))
app.aconf = aconf
app.ir = ir
app.econf = econf
app.diag = diag
if app.kick:
self.logger.info("running '%s'" % app.kick)
os.system(app.kick)
elif app.ambex_pid != 0:
self.logger.info("notifying PID %d ambex" % app.ambex_pid)
os.kill(app.ambex_pid, signal.SIGHUP)
self.logger.info("configuration updated from snapshot %s" % snapshot)
self._respond(rqueue, 200,
'configuration updated from snapshot %s' % snapshot)
if app.health_checks and not app.stats_updater:
app.logger.info("starting Envoy status updater")
app.stats_updater = PeriodicTrigger(app.watcher.update_estats,
period=5)
# app.scout_updater = PeriodicTrigger(lambda: app.watcher.check_scout("30s"), period=30)
# Don't use app.check_scout; it will deadlock. And don't bother doing the Scout
# update until after we've taken care of Envoy.
self.check_scout("update")
def _load_ir(self, rqueue: queue.Queue, aconf: Config,
fetcher: ResourceFetcher, secret_handler: SecretHandler,
snapshot: str) -> None:
aconf.load_all(fetcher.sorted())
aconf_path = os.path.join(app.snapshot_path, "aconf-tmp.json")
open(aconf_path, "w").write(aconf.as_json())
ir = IR(aconf, secret_handler=secret_handler)
ir_path = os.path.join(app.snapshot_path, "ir-tmp.json")
open(ir_path, "w").write(ir.as_json())
econf = EnvoyConfig.generate(ir, "V2")
diag = Diagnostics(ir, econf)
bootstrap_config, ads_config = econf.split_config()
if not self.validate_envoy_config(config=ads_config,
retries=self.app.validation_retries):
self.logger.info(
"no updates were performed due to invalid envoy configuration, continuing with current configuration..."
)
# Don't use app.check_scout; it will deadlock.
self.check_scout("attempted bad update")
self._respond(
rqueue, 500,
'ignoring: invalid Envoy configuration in snapshot %s' %
snapshot)
return
snapcount = int(os.environ.get('AMBASSADOR_SNAPSHOT_COUNT', "4"))
snaplist: List[Tuple[str, str]] = []
if snapcount > 0:
self.logger.debug("rotating snapshots for snapshot %s" % snapshot)
# If snapcount is 4, this range statement becomes range(-4, -1)
# which gives [ -4, -3, -2 ], which the list comprehension turns
# into [ ( "-3", "-4" ), ( "-2", "-3" ), ( "-1", "-2" ) ]...
# which is the list of suffixes to rename to rotate the snapshots.
snaplist += [(str(x + 1), str(x))
for x in range(-1 * snapcount, -1)]
# After dealing with that, we need to rotate the current file into -1.
snaplist.append(('', '-1'))
# Whether or not we do any rotation, we need to cycle in the '-tmp' file.
snaplist.append(('-tmp', ''))
for from_suffix, to_suffix in snaplist:
for fmt in [
"aconf{}.json", "econf{}.json", "ir{}.json",
"snapshot{}.yaml"
]:
from_path = os.path.join(app.snapshot_path,
fmt.format(from_suffix))
to_path = os.path.join(app.snapshot_path,
fmt.format(to_suffix))
try:
self.logger.debug("rotate: %s -> %s" %
(from_path, to_path))
os.rename(from_path, to_path)
except IOError as e:
self.logger.debug("skip %s -> %s: %s" %
(from_path, to_path, e))
pass
except Exception as e:
self.logger.debug("could not rename %s -> %s: %s" %
(from_path, to_path, e))
app.latest_snapshot = snapshot
self.logger.info("saving Envoy configuration for snapshot %s" %
snapshot)
with open(app.bootstrap_path, "w") as output:
output.write(json.dumps(bootstrap_config, sort_keys=True,
indent=4))
with open(app.ads_path, "w") as output:
output.write(json.dumps(ads_config, sort_keys=True, indent=4))
app.aconf = aconf
app.ir = ir
app.econf = econf
app.diag = diag
if app.kick:
self.logger.info("running '%s'" % app.kick)
os.system(app.kick)
elif app.ambex_pid != 0:
self.logger.info("notifying PID %d ambex" % app.ambex_pid)
os.kill(app.ambex_pid, signal.SIGHUP)
if app.ir.k8s_status_updates:
for name in app.ir.k8s_status_updates.keys():
kind, update = app.ir.k8s_status_updates[name]
self.logger.info(
f"doing K8s status update for {kind} {name}...")
text = json.dumps(update)
with open(f'/tmp/kstat-{kind}-{name}', 'w') as out:
out.write(text)
cmd = [
'/ambassador/kubestatus', kind, '-f',
f'metadata.name={name}', '-u', '/dev/fd/0'
]
self.logger.info(f"Running command: {cmd}")
try:
rc = subprocess.run(cmd,
input=text.encode('utf-8'),
timeout=5)
self.logger.info(f'...update finished, rc {rc.returncode}')
except subprocess.TimeoutExpired as e:
self.logger.error(f'...update timed out, {e}')
self.logger.info("configuration updated from snapshot %s" % snapshot)
self._respond(rqueue, 200,
'configuration updated from snapshot %s' % snapshot)
if app.health_checks and not app.stats_updater:
app.logger.info("starting Envoy status updater")
app.stats_updater = PeriodicTrigger(app.watcher.update_estats,
period=5)
# Check our environment...
self.check_environment()
self.chime()
def _load_ir(self, rqueue: queue.Queue, aconf: Config,
fetcher: ResourceFetcher,
secret_reader: Callable[['IRTLSContext', str, str],
SavedSecret], snapshot: str) -> None:
aconf.load_all(fetcher.sorted())
aconf_path = os.path.join(app.snapshot_path, "aconf-tmp.json")
open(aconf_path, "w").write(aconf.as_json())
ir = IR(aconf, secret_reader=secret_reader)
ir_path = os.path.join(app.snapshot_path, "ir-tmp.json")
open(ir_path, "w").write(ir.as_json())
econf = EnvoyConfig.generate(ir, "V2")
diag = Diagnostics(ir, econf)
bootstrap_config, ads_config = econf.split_config()
if not self.validate_envoy_config(config=ads_config):
self.logger.info(
"no updates were performed due to invalid envoy configuration, continuing with current configuration..."
)
app.check_scout("attempted bad update")
self._respond(
rqueue, 500,
'ignoring: invalid Envoy configuration in snapshot %s' %
snapshot)
return
self.logger.info("rotating snapshots for snapshot %s" % snapshot)
for from_suffix, to_suffix in [('-3', '-4'), ('-2', '-3'),
('-1', '-2'), ('', '-1'), ('-tmp', '')]:
for fmt in [
"aconf{}.json", "econf{}.json", "ir{}.json",
"snapshot{}.yaml"
]:
try:
from_path = os.path.join(app.snapshot_path,
fmt.format(from_suffix))
to_path = os.path.join(app.snapshot_path,
fmt.format(to_suffix))
self.logger.debug("rotate: %s -> %s" %
(from_path, to_path))
os.rename(from_path, to_path)
except IOError as e:
self.logger.debug("skip %s -> %s: %s" %
(from_path, to_path, e))
except Exception as e:
self.logger.debug("could not rename %s -> %s: %s" %
(from_path, to_path, e))
self.logger.info("saving Envoy configuration for snapshot %s" %
snapshot)
with open(app.bootstrap_path, "w") as output:
output.write(json.dumps(bootstrap_config, sort_keys=True,
indent=4))
with open(app.ads_path, "w") as output:
output.write(json.dumps(ads_config, sort_keys=True, indent=4))
app.aconf = aconf
app.ir = ir
app.econf = econf
app.diag = diag
if app.kick:
self.logger.info("running '%s'" % app.kick)
os.system(app.kick)
elif app.ambex_pid != 0:
self.logger.info("notifying PID %d ambex" % app.ambex_pid)
os.kill(app.ambex_pid, signal.SIGHUP)
self.logger.info("configuration updated from snapshot %s" % snapshot)
self._respond(rqueue, 200,
'configuration updated from snapshot %s' % snapshot)
if app.health_checks and not app.stats_updater:
app.logger.info("starting Envoy status updater")
app.stats_updater = PeriodicTrigger(app.watcher.update_estats,
period=5)
# Don't use app.check_scout; it will deadlock. And don't bother doing the Scout
# update until after we've taken care of Envoy.
self.check_scout("update")