Integrate client with Confluent schema registry and Avro #40

roopahc · 2016-09-07T19:46:47Z

Fixes : #36

ghost · 2016-09-07T19:46:49Z

It looks like @roopahc hasn't signed our Contributor License Agreement, yet.

Appreciation of efforts,

clabot

criccomini · 2016-09-08T15:50:28Z

@edenhill @ewencp Would appreciate a review when you've got the cycles. :)

roopahc · 2016-09-08T18:00:27Z

@ConfluentCLABot Have signed CLA

wkiser · 2016-09-08T20:03:13Z

confluent_kafka/avro/serializer/MessageSerializer.py

+    pass
+
+
+class ContextStingIO(io.BytesIO):


nit: should be String

matthayes · 2016-09-08T23:11:31Z

FYI I tried testing this and was able to consume message using kafka-avro-console-consumer. Thanks for working on this 👍

criccomini · 2016-09-08T23:21:58Z

Hi Matt! :)

matthayes · 2016-09-08T23:24:13Z

Hey! :)

matthayes · 2016-09-09T18:50:06Z

confluent_kafka/avro/AvroProducer.py

+                self.log.error("Schema required for key serialization")
+                raise SerializerError("Avro schema required for key")
+
+        self.producer.produce(topic, value, key)


kwargs is being ignored, so if you pass callback/on_delivery or partition they aren't used.

criccomini · 2016-09-12T14:54:57Z

@edenhill @ewencp ping

ewencp

@roopahc This looks like a great start. I took a pass on all the main code (didn't go through tests thoroughly). I think the main thing that would need to be decided is the composition/delegation approach to the API vs inheriting from the Producer class. There were some other comments, but they are mostly details.

ewencp · 2016-09-15T16:52:14Z

README.md

+schemaRegistryClient = CachedSchemaRegistryClient(url=schemaRegistryUrl)
+serializer = MessageSerializer(schemaRegistryClient)
+
+avroProducer = AvroProducer(producer, serializer)


Interesting -- I was imagining just using a subclass of Producer. In the case of Producer the subclass probably doesn't save too much, but it also means that you only have to override the one method. With the delegation approach used here, you need to re-implement the API to do delegation (e.g. you are currently missing poll() and flush() methods in this implementation.

You also separate out the serializer class. Are you thinking of reusing this elsewhere or could it just be integrated directly into the AvroProducer class? I'm imagining that it would be a lot simpler and more concise for users if they could just do something like:

from confluent_kafka.avro import AvroProducer producer = AvroProducer({'bootstrap.servers': 'mybroker,mybroker2'}, schema_registry_url='https://<host>:<port>') producer.produce(...)

Are there drawbacks to this? Even if you wanted some of the helper classes to still be available separately (CachedSchemaRegistryClient in particular seems useful to have separate), they could be instantiated by the producer itself since it is already specific to the serialization format.

@ewencp I started with inheritance, but it did not let me inherit class Producer.

Traceback (most recent call last): File "test.py", line 4, in <module> class AvroProducer(Producer): TypeError: Error when calling the metaclass bases type 'cimpl.Producer' is not an acceptable base type

Hmm, interesting. I didn't realize that making C classes subclassable apparently requires some extra work. Some brief Googling suggests it is definitely possible, but does require some extra work. @edenhill Any idea what's involved and if it will make sense to get a cleaner design?

I will look into the sub-classing problem, and I agree with @ewencp that it would be preferred to wrapping it.

ewencp · 2016-09-15T17:10:11Z

confluent_kafka/avro/AvroProducer.py

+            Sends message to kafka by encoding with specified avro schema
+            @:param: topic: topic name
+            @:param: value: A dictionary object
+            @:param: value_avro_schema : Avro schema for value


I'd shorten these names to value_schema and key_schema -- we're in an Avro-specific class, so they couldn't really be anything else.

I don't know if people might also find it useful to be able to set the schema as a default in the constructor? In languages like Java where the schema is attached anyway and is handled automatically if you use code generation, this isn't as annoying. It might be a bit more annoying to have to pass the same schema (the normal case) into every produce call. Not sure how valuable this will be since usually you only have a small number of call sites to produce() anyway, but seems annoying to have to carry the schema around separately everywhere.

Do you think a topic->schema mapping in the constructor would suffice?

We'd have to do two: one for keys and one for values

@criccomini I like that, but there are definitely use cases where people intentionally publish data with different schemas to the same topic. So I guess it kind of depends on how opinionated we want to be...

but there are definitely use cases where people intentionally publish data with different schemas to the same topic

Won't the schema registry reject that (if compatibility is turned on)?

I guess I agree it's possible. It's just inconvenient, lol. :)

Another annoying option would be to provide a wrapper class for key/value to emulate Java. This would allow us to exactly mirror the parent's produce() method. The wrapper would have like payload and schema fields. AvroProducer would take the schema field out and use it, and pass the serialized payload into the parent producer.

Yeah, people turn off compatibility for those cases. I find the pattern questionable, but people do it...

The most convenient way for users might be to just use produce(*args, **kwargs) and sort out the details in the header of the method (i.e. 1 positional arg = value, 2 position args = (key, value), kwargs just work by name). The obvious drawback is lack of documentation in the method signature.

ewencp · 2016-09-15T17:10:13Z

confluent_kafka/avro/AvroProducer.py

+                value = self.serializer.encode_record_with_schema(topic, value_avro_schema, value)
+            else:
+                self.log.error("Schema required for value serialization")
+                raise SerializerError("Avro schema required for value")


In the Java serializer we auto-generate schemas for primitive types. Should we consider doing that here to make it easier to use? I think this is especially common for keys.

Would it be OK to leave this as a separate Github issue/PR? I agree it's useful, but I don't think this PR needs to be exhaustive, and I think you should be able to bolt it on without backwards incompatible changes right?

Yeah, sure that's fine. Just a gap I saw.

Personally, I think this PR should go ahead with supplying schemas for primitive types. Right now, this code only accepts records.

ewencp · 2016-09-15T17:12:37Z

confluent_kafka/avro/AvroProducer.py

+        self.serializer = message_serializer
+        self.log = logging.getLogger(__name__)
+
+    def produce(self, topic, value=None, value_avro_schema=None, key=None, key_avro_schema=None, *args, **kwargs):


Hmm, this argument order is kind of interesting. With only one signature, it's annoying to support everything cleanly. The main thing that's unusual here is that key comes after value. I get the reason for this is so you can produce values without keyword arguments, but would it make sense to try parsing a *args list so you don't have to use keyword arguments to get the natural ordering when using both keys and values? Or does that get too messy (or maybe not possible to always distinguish the right mapping)?

Personally I prefer required keyword arguments in this case (i.e. def produce(self, topic, *, value=None, value_avro_schema=None, etc.) It's hard to remember the argument order and easy to get it wrong (e.g. people may differ on whether they think key or value first is more natural).

@roopahc just pointed out that the parent (wrapped) class' produce() method has nearly an identical order:

http://docs.confluent.io/3.0.1/clients/confluent-kafka-python/index.html#confluent_kafka.Producer.produce

Seems like keeping it as-is actually matches best with what's already there...

I still think keyword args would be clearer. The order is similar except with schema params inserted. It's also not obvious whether the callback/on_delivery and partitioner args from producer will work as positional arguments after key_avro_schema and in what order they need to be or if they need to be keyword args. It's probably best to have these explicitly declared as keyword args to reduce confusion and mistakes.

Hah, it's possible we had this exact same discussion when reviewing the original Python API proposal and I have just completely forgotten about it...

I have tried aligning the order of arguments with confluent_kafka.Producer().produce() method documentation.
http://docs.confluent.io/3.0.1/clients/confluent-kafka-python/index.html. It might be misleading if we change the order in AvroProducer class. let me know your views on the same.

At this point, I yield. :) I don't have a super strong opinion. Can you guys give an exact method header for what you think is best? My Python foo is weak. Is this it?

def produce(self, *args, **kwargs):

@matthayes mentioned required keyword args. It looks like this is only 3.0 compatible PEP-3102. Based on @ewencp's comment below, I still think:

produce(*args, **kwargs)

Is the best we can do right now.

ewencp · 2016-09-15T17:16:34Z

confluent_kafka/avro/CachedSchemaRegistryClient.py

+        self.subject_to_schema_versions = {}
+
+        # below classes were not hashable. hence defining them explicitely as a quick fix
+        schema.RecordSchema.__hash__ = self.hash_func


Do this once at the module level so it only happens on import?

ewencp · 2016-09-15T17:27:03Z

confluent_kafka/avro/CachedSchemaRegistryClient.py

+            url += '/' + subject
+
+        result, code = self._send_request(url)
+        if (code == 200):


Other status codes should be valid here too

ewencp · 2016-09-15T17:36:33Z

confluent_kafka/avro/serializer/MessageSerializer.py

+        return self.encode_record_with_schema_id(schema_id, record)
+
+    # subject = topic + suffix
+    def encode_record_for_topic(self, topic, record, is_key=False):


This seems like an odd use case -- it's basically assuming that the latest schema will be compatible with the data you pass in? Are you using this in practice?

This method is not being used anywhere. We can take it off .

ewencp · 2016-09-15T17:40:53Z

confluent_kafka/avro/serializer/MessageSerializer.py

+
+HAS_FAST = False
+try:
+    from fastavro.reader import read_data


Should we also get the fast writer?

@ewencp May be this can be considered as performance improvement and be part of separate PR?

Sure, that's fine. Was just surprised to see reader but not writer :)

ewencp · 2016-09-15T17:41:41Z

confluent_kafka/avro/serializer/Util.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python


Module names shouldn't be capitalized (i.e. the file name should just be util.py, not Util.py. In fact, this is true for all the filenames under this module (CachedSchemaRegistryClient, AvroProducer, etc).

Indeed, try to use from confluent_kafka.avro.serializer import util leads to ImportError: cannot import name 'util'.

@blbradley Fixed this.

ewencp · 2016-09-15T17:45:14Z

README.md

+
+producer = Producer({'bootstrap.servers': 'mybroker,mybroker2'})
+
+schemaRegistryUrl = 'https://<host>:<port>'


nit: these variables should use underscores instead of camel case

criccomini · 2016-09-15T21:05:32Z

I think the main thing that would need to be decided is the composition/delegation approach to the API vs inheriting from the Producer class

@roopahc initially had the class try to extend Producer. This didn't work because it's a C class, not a Python class underneath. I'm open to suggestions on how to fix this, but Google hasn't yielded anything. It appears you can't extend it. It gives an error (which I no longer have, sigh).

Edit: I see that @roopahc has pasted the proper stack trace above.

criccomini · 2016-09-19T16:33:13Z

confluent_kafka/avro/avro_producer.py

+                raise SerializerError("Avro schema required for key")
+
+        self._producer.produce(topic, value, key, *args, **kwargs)
+    def poll(self, timeout):


Nit: newline above def poll

criccomini · 2016-09-21T20:34:48Z

README.md

+
+schema_registry_url = 'https://<host>:<port>'
+
+avroProducer = AvroProducer(producer, schema_registry_url, value_schema=value_avro_schema, key_schema=key_avro_schema)


Either make the names default_value_schema and default_key_schema or switch to maps from topic to schema:

default_key_schemas=None, default_value_schemas=None

If you go with the latter approach, you'd get the default schema based on the topic name.

criccomini · 2016-09-21T20:35:31Z

confluent_kafka/avro/avro_producer.py

        self._producer = producer
-        self._serializer = message_serializer
+        _cash_client= CachedSchemaRegistryClient(url=schema_registry_url)


nit: _cache_client

criccomini · 2016-09-21T20:36:15Z

confluent_kafka/avro/avro_producer.py

+    '''
+    def __init__(self, producer, schema_registry_url, key_schema = None, value_schema = None):  # real signature unknown; restored from __doc__
+        self._producer = producer
+        _cash_client= CachedSchemaRegistryClient(url=schema_registry_url)


nit: _cache_client

Also add a space between t=

criccomini · 2016-09-21T20:37:22Z

confluent_kafka/avro/avro_producer.py

+        # get schemas from  kwargs if defined
+        key_schema = kwargs.pop('key_schema', None)
+        value_schema = kwargs.pop('value_schema', None)
+        topic= kwargs.pop('topic', None)


Giving up on commenting on this, but you have a bunch of different space styles:

style= one style = two style=three

Please just use style = two everywhere except params. For params, use style=three. Please update everywhere.

criccomini

Overall, LGTM. Left a bunch of nits.

criccomini · 2016-09-21T20:41:59Z

confluent_kafka/avro/avro_producer.py

+            key_schema=self.key_schema
+
+        # if value_schema is not initialized, fall back on default value_schema passed as construction param.
+        if value_schema is None:


if not value_schema:

criccomini · 2016-09-21T20:42:16Z

confluent_kafka/avro/avro_producer.py

+        if value_schema is None:
+            value_schema=self.value_schema
+
+        if value is not None:


if value:

criccomini · 2016-09-21T20:42:22Z

confluent_kafka/avro/avro_producer.py

+            value_schema=self.value_schema
+
+        if value is not None:
+            if value_schema is not None:


if value_schema

criccomini · 2016-09-21T20:42:45Z

confluent_kafka/avro/avro_producer.py

+                log.error("Schema required for value serialization")
+                raise SerializerError("Avro schema required for value")
+
+        if key is not None:


criccomini · 2016-09-21T20:42:54Z

confluent_kafka/avro/avro_producer.py

+                raise SerializerError("Avro schema required for value")
+
+        if key is not None:
+            if key_schema is not None:


if key_schema:

criccomini · 2016-09-21T20:43:21Z

confluent_kafka/avro/avro_producer.py

+        key_schema = kwargs.pop('key_schema', None)
+        value_schema = kwargs.pop('value_schema', None)
+        topic= kwargs.pop('topic', None)
+        if topic is None:


if not topic:

criccomini · 2016-09-21T20:43:30Z

confluent_kafka/avro/avro_producer.py

+        key= kwargs.pop('key', None)
+
+        # if key_schema is not initialized, fall back on default key_schema passed as construction param.
+        if key_schema is None:


if not key_schema:

criccomini · 2016-09-21T20:44:16Z

README.md

@@ -85,6 +106,9 @@ Install
 **Install from PyPi:**

    $ pip install confluent-kafka
+
+    #for Avroproducer


Nit: space between # and f

criccomini · 2016-09-23T14:42:31Z

README.md

 value = {"name": "Value"}
 key = {"name": "Key"}

 producer = Producer({'bootstrap.servers': 'mybroker,mybroker2'})

 schema_registry_url = 'https://<host>:<port>'

-avroProducer = AvroProducer(producer, schema_registry_url, value_schema=value_avro_schema, key_schema=key_avro_schema)
+avroProducer = AvroProducer(producer, schema_registry_url, value_schema=default_value_schema, key_schema=default_key_schema)


sorry, I meant name the pram default_value_schema and default_key_schema, not the values you pass in.

avroProducer = AvroProducer(producer, schema_registry_url, default_value_schema=value_schema, default_key_schema=key_schema)

criccomini · 2016-09-23T14:43:37Z

@ewencp @matthayes look good?

matthayes · 2016-09-23T17:44:06Z

LGTM. I also tested the new code and it's working in my codebase, including parameters passed to produce.

criccomini · 2016-09-23T22:03:35Z

Awesome, thanks Matt! @ewencp we've also moved forward internally with this code. Could you or @edenhill please merge? When that's done, we can open follow-up issues, and start on consumer.

criccomini · 2016-09-27T17:59:06Z

ping

blbradley · 2016-09-30T00:45:37Z

Thoughts:

Schema generation for primitives should be included in the first patch. It would reduce the design changes required to implement it.
AvroProducer.produce requires keyword arguments and Producer.produce does not. These should be consistent.

Anything we can do to get this moving? I'd gladly do the work.

criccomini · 2016-09-30T15:55:02Z

@blbradley no one from Confluent is responding. We've been going back and forth on the produce() method for over a week. At the end of the day, we need to make forward progress.

Anything we can do to get this moving? I'd gladly do the work.

Go for it! :) We're starting to look at the consumer side. If you want to forge ahead with the producer, I say 👍 .

criccomini · 2016-11-08T18:47:10Z

@edenhill what work remains? Are you just referring to eliminating the need for a wrap vs extend?

ewencp · 2016-11-09T00:24:52Z

@criccomini Mainly the wrap vs extend -- the concern is that folks pick this stuff up from master and, whether we think it is reasonable or not given an unreleased version, start depending on it and complain loudly when you change things. The only other concern I would have is if anything pops up in how we have to handle the deserialization for consumers (though I don't really envision that being an issue).

criccomini · 2016-11-09T20:55:15Z

I guess my counter-concern is just that this goes into a branch, diverges from master, and ends up in the dust bin after two months of work. We also have a consumer PR incoming today.

Do you or @edenhill have a suggestion about how to fix the wrap/extend issue?

edenhill · 2016-11-09T20:59:08Z

Oh that certainly wont happen, despite the fact that we've been extremely slow on this we really want your Avro clients to be part of this project.

I'm currently working on making the C classes sub-classable and I will have a branch ready within a day that I'd like you try out.

criccomini · 2016-11-09T21:06:13Z

👍

edenhill · 2016-11-09T22:35:43Z

PR #63 provides sub-classable Producer and Consumer classes, please try it out.

edenhill · 2016-11-14T21:55:31Z

confluent_kafka/avro/__init__.py

    """
        Kafka Producer client which does avro schema encoding to messages.
        Handles schema registration, Message serialization.

        Constructor takes below parameters

-        @:param: producer: confluent_kafka.Producer object
-        @:param: schema_registry_url: Schema registry URL
+        @:param: config: dict object with config parameters.


Mention schema.registry.url and that it is required.

edenhill · 2016-11-14T21:56:02Z

confluent_kafka/avro/__init__.py

@@ -1,6 +1,8 @@
 import logging
 import sys

+from confluent_kafka.cimpl import Producer


Are you sure it should import from .cimpl directly and not from confluent.kafka?

Possibly related to this:
https://travis-ci.org/confluentinc/confluent-kafka-python/jobs/175863152#L584

edenhill · 2016-11-14T21:58:42Z

confluent_kafka/avro/__init__.py

-        self._producer = producer
-        self._serializer = MessageSerializer(CachedSchemaRegistryClient(url=schema_registry_url))
+        if ('schema.registry.url' not in config.keys()):
+            raise ClientError("Missing parameter: schema.registry.url")


Might want to use a standard ValueError

roopahc · 2016-11-14T22:48:03Z

@edenhill I am getting ImportError: dlopen(/Users/roopac/development/confluent-kafka-python/confluent_kafka/cimpl.so, 2): Symbol not found: __Py_ZeroStruct Possibly not related to my code change

ewencp · 2016-11-14T23:03:50Z

@roopahc Are you possibly building against a different python devel package than you are running it with? Py_ZeroStruct exists in 2.7 but not in python 3+.

roopahc · 2016-11-14T23:16:59Z

@ewencp Thanks.. tests pass in my system... Errors seem unrelated.

edenhill · 2016-11-15T06:40:49Z

Since the tests on master are okay I think this failure is related to your code changes.
I see that avro is included unconditionally in __init__.py, is this correct?

The Travis installer does 'pip install .' to install the module, will that make it include or miss the avro extension?

edenhill · 2016-11-15T08:38:43Z

There's a bunch of errors like these:
E IOError: [Errno 2] No such file or directory: 'basic_schema.avsc'

The py.test (or tox) runs from the top directory, so you'll need to reference the .avsc either by tests/ prefix or preferably by dirname()ing the script to get its directory.

I also feel it would be cleaner if the avro tests were in its own sub-directory, e.g., tests/avro, or avro/tests
py.test et.al are pretty good (too good) at finding tests so this shouldnt be a problem.

edenhill · 2016-11-15T08:39:23Z

There's no docstring for the avro module

edenhill · 2016-11-15T10:11:24Z

I updated docstring verification tests on master, so please rebase on latest master again as you fix the other issues. Thanks

edenhill · 2016-11-15T19:16:00Z

@roopahc Check pip install --help, pip install just takes a single module, you need to split that up into multiple lines

edenhill · 2016-11-15T19:17:54Z

Also, please fixup/squash some commits, e.g., the review-fix stuff commits should be joined with their parent commits:

 $ git rebase -i master
 # Mark "fixes-of-fixes" as "f" and move them to the line below their logical parent commit,
 # repeat as necessary

 # When happy with the results (no wip/fix/temporary-tests commit), do a force push:
 $ git push --dry-run --force <remote> <branch>

  # Looks okay? Drop --dry-run

edenhill · 2016-11-15T19:39:01Z

.travis.yml

@@ -7,6 +7,7 @@ before_install:
  - pip install pytest-timeout
 install:
  - pip install -v --global-option=build_ext --global-option="-Itmp-build/include/" --global-option="-Ltmp-build/lib" .
+  - pip install --upgrade pip


Is the .avro install requiring a newer pip version?
If so, we probably need to document what version is required.
And also move this to before_install prior to any other pip commands.

I've found that a lot of older pips don't properly support extra_requires

(We're just shooting in the dark)

criccomini · 2016-11-18T22:20:49Z

Tests are passing now! Please have a look.

edenhill

LGTM, just two questions

edenhill · 2016-11-21T10:39:11Z

confluent_kafka/__init__.py

@@ -1,2 +1,2 @@
-__all__ = ['cimpl','kafkatest']
+__all__ = ['cimpl', 'avro', 'kafkatest']


Should this be conditional based on install type?
What happens if people dont install the avro submodule?

@edenhill This should not affect using the client normally, just with basic producer and consumer.
The only catch is you would not be able to use AvroProducer unless the dependencies are installed.

edenhill · 2016-11-21T10:39:58Z

confluent_kafka/avro/__init__.py

+        value_schema = kwargs.pop('value_schema', self._value_schema)
+        topic = kwargs.pop('topic', None)
+        if not topic:
+            log.error("Topic name not specified.")


Why both log and exception?

roopahc · 2016-11-21T19:04:00Z

@edenhill @ewencp is it good for merging ? :)

edenhill · 2016-11-21T19:03:34Z

confluent_kafka/avro/__init__.py

            raise ClientError("Topic name not specified.")
        value = kwargs.pop('value', None)
        key = kwargs.pop('key', None)
        if value:
            if value_schema:
                value = self._serializer.encode_record_with_schema(topic, value_schema, value)
+                log.debug("Value encoded successfully!")


I dont think there is much value (pun!) to these log messages, they dont include any actual information, and there is an exception to counter-represent their coverage, so I think you should remove all the just-before-exception logs, and logs like this.

I just commented on one of these log occurences in the previous review, but all of them should be removed.

edenhill · 2016-11-21T19:05:04Z

Yeah, I think we should just get rid of the excessive/redundant log messages.
after that we're good to go!

roopahc · 2016-11-21T20:16:29Z

@edenhill done :)

edenhill · 2016-11-21T20:28:46Z

LGTM!

edenhill · 2016-11-21T20:32:37Z

Finally!

roopahc mentioned this pull request Sep 7, 2016

Integrate client with Confluent schema registry and Avro #36

Closed

wkiser reviewed Sep 8, 2016
View reviewed changes

confluent_kafka/avro/serializer/MessageSerializer.py

pass

class ContextStingIO(io.BytesIO):

Copy link

wkiser Sep 8, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: should be String

matthayes reviewed Sep 9, 2016
View reviewed changes

ewencp suggested changes Sep 15, 2016

View reviewed changes

criccomini reviewed Sep 19, 2016

View reviewed changes

criccomini reviewed Sep 21, 2016

View reviewed changes

criccomini reviewed Sep 23, 2016

View reviewed changes

edenhill reviewed Nov 14, 2016

View reviewed changes

edenhill reviewed Nov 15, 2016

View reviewed changes

criccomini mentioned this pull request Nov 18, 2016

Client segfaults when threads are used #78

Closed

edenhill reviewed Nov 21, 2016

View reviewed changes

edenhill suggested changes Nov 21, 2016

View reviewed changes

Producer client for handling avro schemas

6dbb509

edenhill approved these changes Nov 21, 2016

View reviewed changes

edenhill merged commit 406a2bd into confluentinc:master Nov 21, 2016

odemkovych mentioned this pull request Jun 8, 2023

Consumer close() function hungs and prevent app exit #1573

Closed


		producer = Producer({'bootstrap.servers': 'mybroker,mybroker2'})

		schemaRegistryUrl = 'https://<host>:<port>'


		schema_registry_url = 'https://<host>:<port>'

		avroProducer = AvroProducer(producer, schema_registry_url, value_schema=value_avro_schema, key_schema=key_avro_schema)

		@@ -1,2 +1,2 @@
		__all__ = ['cimpl','kafkatest']
		__all__ = ['cimpl', 'avro', 'kafkatest']

Integrate client with Confluent schema registry and Avro #40

Integrate client with Confluent schema registry and Avro #40

Conversation

roopahc commented Sep 7, 2016

ghost commented Sep 7, 2016

criccomini commented Sep 8, 2016

roopahc commented Sep 8, 2016

Choose a reason for hiding this comment

matthayes commented Sep 8, 2016

criccomini commented Sep 8, 2016

matthayes commented Sep 8, 2016

Choose a reason for hiding this comment

criccomini commented Sep 12, 2016

ewencp left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

roopahc Sep 20, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ewencp Sep 15, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

criccomini commented Sep 15, 2016 • edited Loading

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

criccomini left a comment

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

criccomini Sep 21, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

criccomini commented Sep 23, 2016

matthayes commented Sep 23, 2016

criccomini commented Sep 23, 2016

roopahc Sep 20, 2016 •

edited

Loading

ewencp Sep 15, 2016 •

edited

Loading

criccomini commented Sep 15, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading

criccomini Sep 21, 2016 •

edited

Loading