[ENH] Using msgpack instead of json

Using msgpack instead of json results in faster (de)serialize and
less memory usage. Redis is also capable of msgpack within its lua
api i.e. https://github.com/kengonakajima/lua-msgpack-native.

====== Benchmark =======
JSON median size: 387
MSGPACK median size: 329
------------------------
Diff: 16.20%

JSON
* Serialize: 39286
* Deserialize: 30713

MSGPACK
* Serialize: 23483
* Deserialize: 12602
---------------------
DIFF
* Serialize: 50.35%
* Deserialize: 83.62%

Data extracted from spamhaus-collector
Measurements based on deduplicator-expert
460 events in total process by deducplicator-expert

Signed-off-by: Sebastian Waldbauer <[email protected]>

Author: waldbauer-certat

debian/control

@@ -20,6 +20,7 @@ Build-Depends: debhelper (>= 4.1.16),
                python3-sphinx-rtd-theme,
                python3-termstyle,
                python3-tz,
+               python3-msgpack,
                quilt,
                rsync,
                safe-rm
@@ -40,6 +41,7 @@ Depends: bash-completion,
          python3-requests (>= 2.2.1),
          python3-termstyle (>= 0.1.10),
          python3-tz,
+         python3-msgpack,
          redis-server,
          systemd,
          ${misc:Depends},

intelmq/bots/parsers/json/parser.py

@@ -25,7 +25,9 @@ def process(self):
         for line in lines:
             new_event = MessageFactory.unserialize(line,
                                                    harmonization=self.harmonization,
-                                                   default_type='Event')
+                                                   default_type='Event',
+                                                   use_packer="json")
+
             event = self.new_event(report)
             event.update(new_event)
             if 'raw' not in event:

intelmq/lib/bot.py

@@ -13,6 +13,7 @@
 import inspect
 import io
 import json
+import msgpack
 import logging
 import os
 import re
@@ -318,8 +319,8 @@ def start(self, starting: bool = True, error_on_pipeline: bool = True,
                     self.logger.error('Pipeline failed.')
                 self.__disconnect_pipelines()
 
-            except exceptions.DecodingError as exc:
-                self.logger.exception('Could not decode message from pipeline. No retries useful.')
+            except exceptions.UnserializationError as exc:
+                self.logger.exception('Could not unserialize message from pipeline. No retries useful.')
 
                 # ensure that we do not re-process the faulty message
                 self.__error_retries_counter = self.error_max_retries + 1

intelmq/lib/exceptions.py

@@ -168,3 +168,12 @@ def __init__(self, encodings=None, exception: UnicodeDecodeError = None,
             suffixes.append('with reason %r' % exception.reason)
         suffix = (' ' + ' '.join(suffixes)) if suffixes else ''
         super().__init__("Could not decode string%s." % suffix)
+
+
+class UnserializationError(IntelMQException, ValueError):
+    """
+    Unrecoverable error during message unserialization
+    """
+    def __init__(self, exception: Exception = None, object: bytes = None):
+        self.object = object
+        super().__init__("Could not unserialize message%s." % exception)

intelmq/lib/message.py

@@ -10,6 +10,7 @@
 import warnings
 from collections import defaultdict
 from typing import Any, Dict, Iterable, Optional, Sequence, Union
+import msgpack
 
 import intelmq.lib.exceptions as exceptions
 import intelmq.lib.harmonization
@@ -54,8 +55,8 @@ def from_dict(message: dict, harmonization=None,
         return class_reference(message, auto=True, harmonization=harmonization)
 
     @staticmethod
-    def unserialize(raw_message: str, harmonization: dict = None,
-                    default_type: Optional[str] = None) -> dict:
+    def unserialize(raw_message: bytes, harmonization: dict = None,
+                    default_type: Optional[str] = None, use_packer: str = "msgpack") -> dict:
         """
         Takes JSON-encoded Message object, returns instance of correct class.
 
@@ -68,12 +69,12 @@ def unserialize(raw_message: str, harmonization: dict = None,
             MessageFactory.from_dict
             MessageFactory.serialize
         """
-        message = Message.unserialize(raw_message)
+        message = Message.unserialize(raw_message, use_packer=use_packer)
         return MessageFactory.from_dict(message, harmonization=harmonization,
                                         default_type=default_type)
 
     @staticmethod
-    def serialize(message):
+    def serialize(message) -> bytes:
         """
         Takes instance of message-derived class and makes JSON-encoded Message.
 
@@ -121,7 +122,7 @@ def __init__(self, message: Union[dict, tuple] = (), auto: bool = False,
         elif isinstance(message, tuple):
             self.iterable = dict(message)
         else:
-            raise ValueError("Type %r of message can't be handled, must be dict or tuple.", type(message))
+            raise ValueError("Type %r of message can't be handled, must be dict or tuple." % type(message))
         for key, value in self.iterable.items():
             if not self.add(key, value, sanitize=False, raise_failure=False):
                 self.add(key, value, sanitize=True)
@@ -304,18 +305,32 @@ def deep_copy(self):
                                           harmonization={self.__class__.__name__.lower(): self.harmonization_config})
 
     def __str__(self):
-        return self.serialize()
+        return self.serialize(use_packer="json")
 
-    def serialize(self):
-        self['__type'] = self.__class__.__name__
-        json_dump = utils.decode(json.dumps(self))
-        del self['__type']
-        return json_dump
+    def serialize(self, use_packer: str = "msgpack"):
+        delete_type = False
+        if '__type' not in self:
+            delete_type = True
+            self['__type'] = self.__class__.__name__
+
+        if use_packer == "json":
+            packed = json.dumps(self)
+        else:
+            packed = msgpack.packb(self)
+
+        if delete_type:
+            del self['__type']
+        return packed
 
     @staticmethod
-    def unserialize(message_string: str):
-        message = json.loads(message_string)
-        return message
+    def unserialize(message: bytes, use_packer: str = "msgpack"):
+        try:
+            if use_packer == "json":
+                return json.loads(message)
+            else:
+                return msgpack.unpackb(message)
+        except Exception as exc:
+            raise exceptions.UnserializationError(exception=exc, object=message)
 
     def __is_valid_key(self, key: str):
         try:
@@ -462,14 +477,18 @@ def to_dict(self, hierarchical: bool = False, with_type: bool = False,
                 json_dict_fp = json_dict_fp[subkey]
 
         for key, value in jsondicts.items():
-            new_dict[key] = json.dumps(value, ensure_ascii=False)
+            new_dict[key] = json.dumps(value)
 
         return new_dict
 
     def to_json(self, hierarchical=False, with_type=False, jsondict_as_string=False):
         json_dict = self.to_dict(hierarchical=hierarchical, with_type=with_type)
         return json.dumps(json_dict, ensure_ascii=False, sort_keys=True)
 
+    def to_msgpack(self, hierarchical=False, with_type=False):
+        msgpack_dict = self.to_dict(hierarchical=hierarchical, with_type=with_type)
+        return msgpack.packb(msgpack_dict)
+
     def __eq__(self, other: dict) -> bool:
         """
         Wrapper is necessary as we have additional members

intelmq/lib/pipeline.py

@@ -117,14 +117,14 @@ def send(self, message: str, path: str = "_default",
              path_permissive: bool = False):
         raise NotImplementedError
 
-    def receive(self) -> str:
+    def receive(self) -> bytes:
         if self._has_message:
             raise exceptions.PipelineError("There's already a message, first "
                                            "acknowledge the existing one.")
 
         retval = self._receive()
         self._has_message = True
-        return utils.decode(retval)
+        return retval
 
     def _receive(self) -> bytes:
         raise NotImplementedError

intelmq/lib/test.py

@@ -8,6 +8,7 @@
 import io
 import inspect
 import json
+import msgpack
 import os
 import re
 import unittest
@@ -150,8 +151,7 @@ def setUpClass(cls):
         elif cls.bot_type != 'collector' and cls.default_input_message == '':
             cls.default_input_message = {'__type': 'Event'}
         if type(cls.default_input_message) is dict:
-            cls.default_input_message = \
-                utils.decode(json.dumps(cls.default_input_message))
+            cls.default_input_message = msgpack.packb(cls.default_input_message)
 
         if cls.use_cache and not os.environ.get('INTELMQ_SKIP_REDIS'):
             password = os.environ.get('INTELMQ_TEST_REDIS_PASSWORD') or \
@@ -168,10 +168,10 @@ def setUpClass(cls):
     harmonization = utils.load_configuration(pkg_resources.resource_filename('intelmq',
                                                                              'etc/harmonization.conf'))
 
-    def new_report(self, auto=False, examples=False):
+    def new_report(self, auto=False, examples=False) -> message.Report:
         return message.Report(harmonization=self.harmonization, auto=auto)
 
-    def new_event(self):
+    def new_event(self) -> message.Event:
         return message.Event(harmonization=self.harmonization)
 
     def get_mocked_logger(self, logger):
@@ -240,7 +240,7 @@ def prepare_bot(self, parameters={}, destination_queues=None):
             self.input_queue = []
             for msg in self.input_message:
                 if type(msg) is dict:
-                    self.input_queue.append(json.dumps(msg))
+                    self.input_queue.append(message.MessageFactory.serialize(msg))
                 elif issubclass(type(msg), message.Message):
                     self.input_queue.append(msg.serialize())
                 else:
@@ -315,17 +315,17 @@ def run_bot(self, iterations: int = 1, error_on_pipeline: bool = False,
 
         """ Test if report has required fields. """
         if self.bot_type == 'collector':
-            for report_json in self.get_output_queue():
-                report = message.MessageFactory.unserialize(report_json,
+            for report_data in self.get_output_queue():
+                report = message.MessageFactory.unserialize(report_data,
                                                             harmonization=self.harmonization)
                 self.assertIsInstance(report, message.Report)
                 self.assertIn('raw', report)
                 self.assertIn('time.observation', report)
 
         """ Test if event has required fields. """
         if self.bot_type == 'parser':
-            for event_json in self.get_output_queue():
-                event = message.MessageFactory.unserialize(event_json,
+            for event_data in self.get_output_queue():
+                event = message.MessageFactory.unserialize(event_data,
                                                            harmonization=self.harmonization)
                 self.assertIsInstance(event, message.Event)
                 self.assertIn('classification.type', event)
@@ -383,7 +383,7 @@ def get_output_queue(self, path="_default"):
         """Getter for items in the output queues of this bot. Use in TestCase scenarios
             If there is multiple queues in named queue group, we return all the items chained.
         """
-        return [utils.decode(text) for text in chain(*[self.pipe.state[x] for x in self.pipe.destination_queues[path]])]
+        return [text for text in chain(*[self.pipe.state[x] for x in self.pipe.destination_queues[path]])]
         # return [utils.decode(text) for text in self.pipe.state["%s-output" % self.bot_id]]
 
     def test_bot_name(self, *args, **kwargs):
@@ -514,9 +514,9 @@ def assertMessageEqual(self, queue_pos, expected_msg, compare_raw=True, path="_d
         given queue position.
         """
         event = self.get_output_queue(path=path)[queue_pos]
-        self.assertIsInstance(event, str)
+        self.assertIsInstance(event, bytes)
 
-        event_dict = json.loads(event)
+        event_dict = msgpack.unpackb(event)
         if isinstance(expected_msg, (message.Event, message.Report)):
             expected = expected_msg.to_dict(with_type=True)
         else:

intelmq/tests/bots/collectors/tcp/test_collector.py

@@ -119,7 +119,10 @@ def test_intelmq_exchange(self):
         for i, msg in enumerate(self.get_output_queue()):
             report = MessageFactory.unserialize(msg, harmonization=self.harmonization, default_type='Event')
 
-            output = MessageFactory.unserialize(utils.base64_decode(report["raw"]), harmonization=self.harmonization, default_type='Event')
+            output = MessageFactory.unserialize(utils.base64_decode(report["raw"]),
+                                                harmonization=self.harmonization,
+                                                default_type='Event',
+                                                use_packer="json")
             self.assertDictEqual(output, INPUT1)
 
             del report['time.observation']

intelmq/tests/bots/experts/cymru_whois/test_expert.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-import json
+import msgpack
 import unittest
 
 import intelmq.lib.test as test
@@ -89,7 +89,7 @@ def test_6to4_result(self):
         """
         self.input_message = EXAMPLE_6TO4_INPUT
         self.run_bot()
-        actual = json.loads(self.get_output_queue()[0])
+        actual = msgpack.loads(self.get_output_queue()[0])
         self.assertDictContainsSubset(EXAMPLE_6TO4_INPUT, actual)
         self.assertIn("source.asn", actual)
         self.assertIn("source.as_name", actual)

intelmq/tests/bots/experts/idea/test_expert.py

@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 import unittest
 import json
+import msgpack
 
 import intelmq.lib.test as test
 from intelmq.bots.experts.idea.expert import IdeaExpertBot
@@ -82,8 +83,8 @@ def test_conversion(self):
         # The ID in the generated Idea event is random, so we have to extract
         # the data from the "output" field and compare after removing ID's
         event = self.get_output_queue()[0]
-        self.assertIsInstance(event, str)
-        event_dict = json.loads(event)
+        self.assertIsInstance(event, bytes)
+        event_dict = msgpack.loads(event)
         self.assertIsInstance(event_dict, dict)
         self.assertTrue("output" in event_dict)
         idea_event = json.loads(event_dict["output"])