stacktach-winchester/winchester/pipeline_handler.py

478 lines
19 KiB
Python

import abc
import datetime
import logging
import six
import uuid
from notabene import kombu_driver as driver
logger = logging.getLogger(__name__)
@six.add_metaclass(abc.ABCMeta)
class PipelineHandlerBase(object):
"""Base class for Pipeline handlers.
Pipeline handlers perform the actual processing on a set of events
captured by a stream. The handlers are chained together, each handler
in a pipeline is called in order, and receives the output of the previous
handler.
Once all of the handlers in a pipeline have successfully processed the
events (with .handle_events() ), each handler's .commit() method will be
called. If any handler in the chain raises an exception, processing of
events will stop, and each handler's .rollback() method will be called."""
def __init__(self, **kw):
"""Setup the pipeline handler.
A new instance of each handler for a pipeline is used for each
stream (set of events) processed.
:param kw: The parameters listed in the pipeline config file for
this handler (if any).
"""
@abc.abstractmethod
def handle_events(self, events, env):
""" This method handles the actual event processing.
This method receives a list of events and should return a list of
events as well. The return value of this method will be passed to
the next handler's .handle_events() method. Handlers can add new
events simply by adding them to the list they return. New events
(those with unrecognized message_id's), will be saved to the
database if all handlers in this pipeline complete successfully.
Likewise, handlers can omit events from the list they return to
act as a filter for downstream handlers.
Care should be taken to avoid any operation with side-effects in
this method. Pipelines can be re-tried if a handler throws an
error. If you need to perform such operations, such as interacting
with an external system, save the needed information in an instance
variable, and perform the operation in the .commit() method.
:param events: A list of events.
:param env: Just a dictionary, it's passed to each handler, and
can act as a shared scratchpad.
:returns: A list of events.
"""
@abc.abstractmethod
def commit(self):
""" Called when each handler in this pipeline has successfully
completed.
If you have operations with side effects, preform them here.
Exceptions raised here will be logged, but otherwise ignored.
"""
@abc.abstractmethod
def rollback(self):
""" Called if there is an error for any handler while processing a list
of events.
If you need to perform some kind of cleanup, do it here.
Exceptions raised here will be logged, but otherwise ignored.
"""
class LoggingHandler(PipelineHandlerBase):
def handle_events(self, events, env):
emsg = ', '.join("%s: %s" % (event['event_type'], event['message_id'])
for event in events)
logger.info("Received %s events: \n%s" % (len(events), emsg))
return events
def commit(self):
pass
def rollback(self):
pass
class NotabeneException(Exception):
pass
class ConnectionManager(object):
def __init__(self):
# {connection_properties:
# {exchange_properties: (connection, exchange)}}
self.pool = {}
def _extract_params(self, kw):
host = kw.get('host', 'localhost')
user = kw.get('user', 'guest')
password = kw.get('password', 'guest')
port = kw.get('port', 5672)
vhost = kw.get('vhost', '/')
library = kw.get('library', 'librabbitmq')
exchange_name = kw.get('exchange')
exchange_type = kw.get('exchange_type', 'topic')
if exchange_name is None:
raise NotabeneException("No 'exchange' name provided")
connection_dict = {'host': host, 'port': port,
'user': user, 'password': password,
'library': library, 'vhost': vhost}
connection_tuple = tuple(sorted(connection_dict.items()))
exchange_dict = {'exchange_name': exchange_name,
'exchange_type': exchange_type}
exchange_tuple = tuple(sorted(exchange_dict.items()))
return (connection_dict, connection_tuple,
exchange_dict, exchange_tuple)
def get_connection(self, properties, queue_name):
connection_dict, connection_tuple, \
exchange_dict, exchange_tuple = self._extract_params(properties)
connection_info = self.pool.get(connection_tuple)
if connection_info is None:
connection = driver.create_connection(connection_dict['host'],
connection_dict['port'],
connection_dict['user'],
connection_dict['password'],
connection_dict['library'],
connection_dict['vhost'])
connection_info = (connection, {})
self.pool[connection_tuple] = connection_info
connection, exchange_pool = connection_info
exchange = exchange_pool.get(exchange_tuple)
if exchange is None:
exchange = driver.create_exchange(exchange_dict['exchange_name'],
exchange_dict['exchange_type'])
exchange_pool[exchange_tuple] = exchange
# Make sure the queue exists so we don't lose events.
queue = driver.create_queue(queue_name, exchange, queue_name,
channel=connection.channel())
queue.declare()
return (connection, exchange)
# Global ConnectionManager. Shared by all Handlers.
connection_manager = ConnectionManager()
class NotabeneHandler(PipelineHandlerBase):
# Handlers are created per stream, so we have to be smart about
# things like connections to databases and queues.
# We don't want to create too many connections, and we have to
# remember that stream processing has to occur quickly, so
# we want to avoid round-trips where possible.
def __init__(self, **kw):
super(NotabeneHandler, self).__init__(**kw)
global connection_manager
self.queue_name = kw.get('queue_name')
if self.queue_name is None:
raise NotabeneException("No 'queue_name' provided")
self.connection, self.exchange = connection_manager.get_connection(
kw, self.queue_name)
self.env_keys = kw.get('env_keys', [])
def handle_events(self, events, env):
keys = [key for key in self.env_keys]
self.pending_notifications = []
for key in keys:
self.pending_notifications.extend(env.get(key, []))
return events
def _format_notification(self, notification):
"""Core traits are in the root of the notification and extra
traits go in the payload."""
core_keys = ['event_type', 'message_id', 'timestamp', 'service']
core = dict((key, notification.get(key)) for key in core_keys)
payload = dict((key, notification[key])
for key in notification.keys()
if key not in core_keys)
core['payload'] = payload
# Notifications require "publisher_id", not "service" ...
publisher = core.get('service')
if not publisher:
publisher = "stv3"
core['publisher_id'] = publisher
del core['service']
core['timestamp'] = str(core['timestamp'])
return core
def commit(self):
for notification in self.pending_notifications:
notification = self._format_notification(notification)
logger.debug("Publishing '%s' to '%s' with routing_key '%s'" %
(notification['event_type'], self.exchange,
self.queue_name))
try:
driver.send_notification(notification, self.queue_name,
self.connection, self.exchange)
except Exception as e:
logger.exception(e)
def rollback(self):
pass
class UsageException(Exception):
def __init__(self, code, message):
super(UsageException, self).__init__(message)
self.code = code
class UsageHandler(PipelineHandlerBase):
def __init__(self, **kw):
super(UsageHandler, self).__init__(**kw)
self.warnings = []
def _get_audit_period(self, event):
apb = event.get('audit_period_beginning')
ape = event.get('audit_period_ending')
return apb, ape
def _is_exists(self, event):
return event['event_type'] == 'compute.instance.exists'
def _is_non_EOD_exists(self, event):
# For non-EOD .exists, we just check that the APB and APE are
# not 24hrs apart. We could check that it's not midnight, but
# could be possible (though unlikely).
# And, if we don't find any extras, don't error out ...
# we'll do that later.
apb, ape = self._get_audit_period(event)
return (self._is_exists(event) and apb and ape
and ape.date() != (apb.date() + datetime.timedelta(days=1)))
def _is_EOD_exists(self, event):
# We could have several .exists records, but only the
# end-of-day .exists will have audit_period_* time of
# 00:00:00 and be 24hrs apart.
apb, ape = self._get_audit_period(event)
return (self._is_exists(event) and apb and ape
and apb.time() == datetime.time(0, 0, 0)
and ape.time() == datetime.time(0, 0, 0)
and ape.date() == (apb.date() + datetime.timedelta(days=1)))
def _extract_launched_at(self, exists):
if not exists.get('launched_at'):
raise UsageException("U1", ".exists has no launched_at value.")
return exists['launched_at']
def _extract_interesting_events(self, events, interesting):
return [event for event in events
if event['event_type'] in interesting]
def _find_deleted_events(self, events):
interesting = ['compute.instance.delete.end']
return self._extract_interesting_events(events, interesting)
def _verify_fields(self, this, that, fields):
for field in fields:
if field not in this and field not in that:
continue
if this[field] != that[field]:
raise UsageException("U2",
"Conflicting '%s' values ('%s' != '%s')"
% (field, this[field], that[field]))
def _confirm_delete(self, exists, delete_events, fields):
deleted_at = exists.get('deleted_at')
state = exists.get('state')
apb, ape = self._get_audit_period(exists)
if not deleted_at and delete_events:
raise UsageException("U6", ".deleted events found but .exists "
"has no deleted_at value.")
if deleted_at and state != "deleted":
raise UsageException("U3", ".exists state not 'deleted' but "
".exists deleted_at is set.")
if deleted_at and not delete_events:
# We've already confirmed it's in the "deleted" state.
launched_at = exists.get('launched_at')
if deleted_at < launched_at:
raise UsageException("U4",
".exists deleted_at < .exists launched_at.")
# Is the deleted_at within this audit period?
if (apb and ape and deleted_at >= apb and deleted_at <= ape):
raise UsageException("U5", ".exists deleted_at in audit "
"period, but no matching .delete event found.")
if len(delete_events) > 1:
raise UsageException("U7", "Multiple .delete.end events")
if delete_events:
self._verify_fields(exists, delete_events[0], fields)
def _confirm_launched_at(self, block, exists):
if exists.get('state') != 'active':
return
apb, ape = self._get_audit_period(exists)
# Does launched_at have a value within this audit period?
# If so, we should have a related event. Otherwise, this
# instance was created previously.
launched_at = self._extract_launched_at(exists)
if (apb and ape and
launched_at >= apb and launched_at <= ape and
len(block) == 0):
raise UsageException("U8", ".exists launched_at in audit "
"period, but no related events found.")
# TODO(sandy): Confirm the events we got set launched_at
# properly.
def _get_core_fields(self):
"""Broken out so derived classes can define their
own trait list."""
return ['launched_at', 'instance_flavor_id', 'tenant_id',
'os_architecture', 'os_version', 'os_distro']
def _do_checks(self, block, exists):
interesting = ['compute.instance.rebuild.start',
'compute.instance.resize.prep.start',
'compute.instance.rescue.start']
self._confirm_launched_at(block, exists)
fields = self._get_core_fields()
last_interesting = None
for event in block:
if event['event_type'] in interesting:
last_interesting = event
if last_interesting:
self._verify_fields(last_interesting, exists, fields)
elif self._is_non_EOD_exists(exists):
self.warnings.append("Non-EOD .exists found "
"(msg_id: %s) "
"with no operation event." %
exists['message_id'])
deleted = self._find_deleted_events(block)
delete_fields = ['launched_at', 'deleted_at']
self._confirm_delete(exists, deleted, delete_fields)
def _base_notification(self, exists):
apb, ape = self._get_audit_period(exists)
return {
'payload': {
'audit_period_beginning': str(apb),
'audit_period_ending': str(ape),
'launched_at': str(exists.get('launched_at', '')),
'deleted_at': str(exists.get('deleted_at', '')),
'instance_id': exists.get('instance_id', ''),
'tenant_id': exists.get('tenant_id', ''),
'display_name': exists.get('display_name', ''),
'instance_type': exists.get('instance_flavor', ''),
'instance_flavor_id': exists.get('instance_flavor_id', ''),
'state': exists.get('state', ''),
'state_description': exists.get('state_description', ''),
'bandwidth': {'public': {
'bw_in': exists.get('bandwidth_in', ''),
'bw_out': exists.get('bandwidth_out', '')}},
'image_meta': {
'org.openstack__1__architecture':
exists.get('os_architecture', ''),
'org.openstack__1__os_version': exists.get('os_version', ''),
'org.openstack__1__os_distro': exists.get('os_distro', ''),
'org.rackspace__1__options': exists.get('rax_options', '')
}},
'original_message_id': exists.get('message_id', '')}
def _process_block(self, block, exists):
error = None
try:
self._do_checks(block, exists)
event_type = "compute.instance.exists.verified"
except UsageException as e:
error = e
event_type = "compute.instance.exists.failed"
logger.warn("Stream %s UsageException: (%s) %s" %
(self.stream_id, e.code, e))
apb, ape = self._get_audit_period(exists)
logger.warn("Stream %s deleted_at: %s, launched_at: %s, "
"state: %s, APB: %s, APE: %s, #events: %s" %
(self.stream_id, exists.get("deleted_at"),
exists.get("launched_at"), exists.get("state"),
apb, ape, len(block)))
if len(block) > 1:
logger.warn("Events for Stream: %s" % self.stream_id)
for event in block:
logger.warn("^Event: %s - %s" %
(event['timestamp'], event['event_type']))
events = []
# We could have warnings, but a valid event list.
if self.warnings:
instance_id = exists.get('instance_id', 'n/a')
warning_event = {'event_type': 'compute.instance.exists.warnings',
'message_id': str(uuid.uuid4()),
'timestamp': exists.get('timestamp',
datetime.datetime.utcnow()),
'stream_id': int(self.stream_id),
'instance_id': exists.get('instance_id'),
'warnings': self.warnings}
events.append(warning_event)
new_event = self._base_notification(exists)
new_event.update({'event_type': event_type,
'message_id': str(uuid.uuid4()),
'timestamp': str(exists.get('timestamp',
datetime.datetime.utcnow())),
'stream_id': int(self.stream_id),
'error': str(error),
'error_code': error and error.code})
events.append(new_event)
return events
def handle_events(self, events, env):
self.env = env
self.stream_id = env['stream_id']
self.warnings = []
new_events = []
block = []
for event in events:
if self._is_exists(event):
new_events.extend(self._process_block(block, event))
block = []
else:
block.append(event)
# Final block should be empty.
if block:
new_event = {'event_type': "compute.instance.exists.failed",
'message_id': str(uuid.uuid4()),
'timestamp': block[0].get('timestamp',
datetime.datetime.utcnow()),
'stream_id': int(self.stream_id),
'instance_id': block[0].get('instance_id'),
'error': "Notifications, but no .exists "
"notification found.",
'error_code': "U0"
}
new_events.append(new_event)
env['usage_notifications'] = new_events
return events
def commit(self):
pass
def rollback(self):
pass