Re: How to consume fedora-messaging?

[Adam Williamson]

On Mon, 2019-06-10 at 09:24 +0900, Tristan Cacqueray wrote:
>  On Mon, Jun 10, 2019 at 01:24 Igor Gnatenko wrote:
>  > Hello,
>  > 
>  > I have been trying to write some script which would listen on
>  > generation of new repository / successful build is tagged in Koji and
>  > do some actions locally. Or basically whenever someone pushes commits,
>  > I want to fetch repo locally.
>  > 
>  > I was reading https://fedora-messaging.readthedocs.io/en/latest/consuming.html,
>  > but it is not very clear to me where I can find list of topics and
>  > what data messages will contain... 
As well as the doc, there is a sample consumer:
https://github.com/fedora-infra/fedora-messaging/blob/master/fedora_messa...
and a couple of sample config files:
https://github.com/fedora-infra/fedora-messaging/blob/master/configs/fedo...
https://github.com/fedora-infra/fedora-messaging/blob/master/configs/fedo...
that you will probably find useful.

>  
>  Hi Igor,
>  
>  you can find the list of topics and their associated schema here:
>  https://fedora-fedmsg.readthedocs.io/en/latest/topics.html 
Treat this with caution, because it is...sometimes a lie. It is in fact
auto-generated from the test cases for the
fedmsg_meta_fedora_infrastructure project:

https://github.com/fedora-infra/fedmsg_meta_fedora_infrastructure/

which is where the metadata providers that produce things like the
human-readable summaries you get from FMN live.

If you maintain a thing that publishes fedmsgs, you are *supposed to*
also maintain a provider in fedmsg_meta that can interpret all the
fedmsgs your thing publishes, and a test suite for that provider which
includes tests for every topic your provider publishes on and useful
docstrings for each test. That doc page is then actually built more or
less just by gluing all the test docstrings together.

However, there is no enforcement of any of this. Which means that it
quite often isn't really all in sync. I think there may still be some
fedmsg publishers which don't have a metadata provider or tests at all.
There are definitely some where a provider and tests exist, but the
fedmsg publisher's behaviour has changed a lot since they were written
and no-one has updated the provider or the tests, so the doc no longer
actually lists the topics on which messages are now being published.

I don't really use this list any more because of issues like that...

>  And you can also find samples on this website:
>  https://apps.fedoraproject.org/datagrepper/raw 
...this is much more useful, because it is just a log of all the
*actual* fedmsgs that have ever been published. It can be a bit
unwieldy to work with, but it's not too bad. You can see all Koji
messages from the last two days here:

https://apps.fedoraproject.org/datagrepper/raw?category=buildsys&delt...

You can tweak the 'delta' arg (given in seconds), or use 'start' and
'end' (given as Unix timestamps) instead, to get messages from
different periods.

Note, datagrepper consumes and publishes *fedmsgs*, not fedora-
messaging messages (at least at present). We are currently in the
middle of trying to move things over from fedmsg (the old 0MQ-based
system) to fedora-messaging (the new AMQP-based system). To aid this,
there are bridges in place in both directions: messages published to
fedora-mesaging are re-published as fedmsgs, and messages published as
fedmsgs are re-published to fedora-messaging. Even for messages that
are actually natively published to fedora-messaging, what you see on
datagrepper is the result of the AMQP->0MQ bridge: it's not the actual
original message, but the re-published fedmsg. AFAIK there is no
publicly available 'datagrepper-for-fedora-messaging' ATM, though, so
you can't really look at an archive of the actual messages as they're
seen on the new system.

There is one very important wrinkle you should bear in mind thanks to
this: the message format.

If you're writing an AMQP consumer in Python, what you'll ultimately
get for your consumer to process is a `message` object which is an
instance of a fedora_messaging.api.Message() (or a subclass of it - a
message schema class). This will have a `body` attribute which should
be a dict of the message 'body' - the main meat of the message. If the
message was natively published as a fedora-messaging AMQP message, that
is indeed what it will be. However, if it was published as a fedmsg,
what you get as `message.body` is actually the *entire* fedmsg dict -
the whole dict as you see it when you view a message on datagrepper,
with a bunch of headers and a 'msg' dict with the real body in it, e.g.
https://apps.fedoraproject.org/datagrepper/id?id=2019-633e9879-8974-4ad4-...
.
To get to what is effectively the 'body' of a fedmsg, you need to take
message.body['msg']. This wrinkle is currently under discussion at
https://github.com/fedora-infra/fedmsg-migration-tools/issues/20 .

I came across this while converting fedmsg consumers to fedora-
messaging, and I wrote a little helper to deal with it:

https://pagure.io/fedora-qa/fedora_openqa/blob/fedora-messaging/f/fedora_...

Whenever you want the body of a message you just run it through that
and it should give you the 'true' body. That also means you won't have
to worry if the bridge gets changed to fix the bug, that function will
still give the correct result.

If you want to look at some real-world fedora-messaging consumers, that
file contains three consumers for scheduling openQA jobs and reporting
their results. Also see the reference config files (the '.toml' files)
and the README, at top level of the fedora-messaging branch:

https://pagure.io/fedora-qa/fedora_openqa/tree/fedora-messaging

You can also look at Bodhi, which has also been converted to fedora-
messaging:

https://github.com/fedora-infra/bodhi/blob/fde8eefc016b55be76ec6a5298b8b4...

the config file for Bodhi is in infra ansible:

https://infrastructure.fedoraproject.org/cgit/ansible.git/tree/roles/bodh...

Oh, one more wrinkle I had fun with: if you want to use the 'public'
credentials for the brokers, be aware that the amqp_url setting and
[tls] block settings in the sample config files are correct, but the
necessary certificate and key files for the staging broker are not
included in the `fedora-messaging` package at present. If you look in
/etc/fedora-messaging after installing it you'll see only the files for
the production broker are there. If you want to listen to the staging
broker, you have to grab the files from a git checkout (they're in the
'configs' dir) and copy them to /etc/fedora-messaging .
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

How to consume fedora-messaging?

[Dusty Mabe]

On 6/10/19 12:06 PM, Adam Williamson wrote:
>  On Mon, 2019-06-10 at 09:24 +0900, Tristan Cacqueray wrote:
> > On Mon, Jun 10, 2019 at 01:24 Igor Gnatenko wrote:
> >> Hello,
> >>
> >> I have been trying to write some script which would listen on
> >> generation of new repository / successful build is tagged in Koji and
> >> do some actions locally. Or basically whenever someone pushes commits,
> >> I want to fetch repo locally.
> >>
> >> I was reading https://fedora-messaging.readthedocs.io/en/latest/consuming.html,
> >> but it is not very clear to me where I can find list of topics and
> >> what data messages will contain...
>  
>  As well as the doc, there is a sample consumer:
>  https://github.com/fedora-infra/fedora-messaging/blob/master/fedora_messa...
>  and a couple of sample config files:
>  https://github.com/fedora-infra/fedora-messaging/blob/master/configs/fedo...
>  https://github.com/fedora-infra/fedora-messaging/blob/master/configs/fedo...
>  that you will probably find useful.
>  
> >
> > Hi Igor,
> >
> > you can find the list of topics and their associated schema here:
> > https://fedora-fedmsg.readthedocs.io/en/latest/topics.html
>  
>  Treat this with caution, because it is...sometimes a lie.  

related: https://github.com/fedora-infra/fedora-messaging/issues/180

Dusty

How to consume fedora-messaging?

[Pavel Raiskup]

Hi Adam, or anyone,

On Monday, June 10, 2019 6:06:42 PM CEST Adam Williamson wrote:
>  If you're writing an AMQP consumer in Python, what you'll
> ultimately
>  get for your consumer to process is a `message` object which is an
>  instance of a fedora_messaging.api.Message() (or a subclass of it - a
>  message schema class). 
I fail to see an example of this, I mean ...  [1] says

    Message bodies are JSON objects, that adhere to a schema. Message schemas
    live in their own Python package, so they can be installed on the producer
    and on the consumer.

do we have any such package so I (as a consumer) can install the package
with schema class, and use it to parse the message body?  I have seen
the example of producing the message using the schema [2], but not
consuming - only the toy example which is not really using the schema
class.

>  This will have a `body` attribute which should be a dict of the
> message
>  'body' - the main meat of the message. 
... I can use the "meat" to instantiate the message object manually, by
Message(body=body) perhaps, but I still have to first check the topic,
etc, I hoped there's something like:

  from MYAPP import MessageConsumer
  from fedora_messaging.api import consume
  ...

  class Consumer(MessageConsumer):
      def consume(self, message):
          """ message _is_ instance of the schema class """
          message.do_some_stuff()
          if message.some_property:
              do_something()

  consume(Consumer)

Well, if not that far -- I'd at least like to see something like
`python3-foo-fedora-messaging` package which I could install and give it a
try (this sounds like it is common pattern from the docs).  I'd like to design
the copr.git/fedora-messaging directory according to that.

[1] https://fedora-messaging.readthedocs.io/en/latest/tutorial/schemas.html
[2]
https://github.com/fedora-infra/fedora-messaging/blob/master/docs/tutoria...
[3]
https://github.com/fedora-infra/fedora-messaging/blob/master/docs/tutoria...

Pavel

How to consume fedora-messaging?

[Jeremy Cline]

Hi,

On 6/20/19 4:11 AM, Pavel Raiskup wrote:
>  Hi Adam, or anyone,
>  
>  On Monday, June 10, 2019 6:06:42 PM CEST Adam Williamson wrote:
> > If you're writing an AMQP consumer in Python, what you'll ultimately
> > get for your consumer to process is a `message` object which is an
> > instance of a fedora_messaging.api.Message() (or a subclass of it - a
> > message schema class).
>  
>  I fail to see an example of this, I mean ...  [1] says
>  
>       Message bodies are JSON objects, that adhere to a schema. Message schemas
>       live in their own Python package, so they can be installed on the producer
>       and on the consumer.
>  
>  do we have any such package so I (as a consumer) can install the package
>  with schema class, and use it to parse the message body?  I have seen
>  the example of producing the message using the schema [2], but not
>  consuming - only the toy example which is not really using the schema
>  class. 
Only if the publisher of the message has created a schema. There are a
couple at this point, but not many. The consumer will log (at warning
level) if a message arrived that indicates it has a schema and you don't
have it installed. It doesn't give you a ton of information how to find 
that schema, though, so I filed an issue[0] to fix that.

You can try it out with Bodhi's[1] messages, for example.

>  
> > This will have a `body` attribute which should be a dict of the message
> > 'body' - the main meat of the message.
>  
>  ... I can use the "meat" to instantiate the message object manually, by
>  Message(body=body) perhaps, but I still have to first check the topic,
>  etc, I hoped there's something like:
>  
>     from MYAPP import MessageConsumer
>     from fedora_messaging.api import consume
>     ...
>  
>     class Consumer(MessageConsumer):
>         def consume(self, message):
>             """ message _is_ instance of the schema class
> """
>             message.do_some_stuff()
>             if message.some_property:
>                 do_something()
>  
>     consume(Consumer)
>   
This example wouldn't work because you're not passing the actual
callable, but that issue aside, your consumer gets an instance of
the Message class or one of its sub-classes. The Message class is
used if the publisher used it or if you don't have the schema
available.


[0] https://github.com/fedora-infra/fedora-messaging/issues/187
[1] https://pypi.org/project/bodhi-messages/

- Jeremy

How to consume fedora-messaging?

[Adam Williamson]

On Thu, 2019-06-20 at 10:11 +0200, Pavel Raiskup wrote:
>  Hi Adam, or anyone,
>  
>  On Monday, June 10, 2019 6:06:42 PM CEST Adam Williamson wrote:
>  > If you're writing an AMQP consumer in Python, what you'll ultimately
>  > get for your consumer to process is a `message` object which is an
>  > instance of a fedora_messaging.api.Message() (or a subclass of it - a
>  > message schema class).
>  
>  I fail to see an example of this, I mean ...  [1] says
>  
>      Message bodies are JSON objects, that adhere to a schema. Message schemas
>      live in their own Python package, so they can be installed on the producer
>      and on the consumer.
>  
>  do we have any such package so I (as a consumer) can install the package
>  with schema class, and use it to parse the message body? 
Yeah, Bodhi. Write a consumer that listens to Bodhi messages, then run
it without python3-bodhi installed. When it gets a message, it'll just
show up as an instance of Message(), and your system log will have a
warning like this:

Jun 20 14:44:40 openqa-stg01.qa.fedoraproject.org fedora-
messaging[11082]: [WARNING fedora_messaging.message] The schema
"bodhi.update.request.testing.v1" is not in the schema registry! Either
install the package with its schema definition or define a schema.
Falling back to the default schema...

Now install python3-bodhi, run the consumer again, and this time the
message should show up as an instance of the appropriate schema class.
(I have not actually tested this as the consumers I'm writing don't
need to use any of the schema methods really...but that's how it's
supposed to work).

>    I have seen
>  the example of producing the message using the schema [2], but not
>  consuming - only the toy example which is not really using the schema
>  class.
>  
>  > This will have a `body` attribute which should be a dict of the message
>  > 'body' - the main meat of the message.
>  
>  ... I can use the "meat" to instantiate the message object manually, by
>  Message(body=body) perhaps, but I still have to first check the topic,
>  etc, I hoped there's something like:
>  
>    from MYAPP import MessageConsumer
>    from fedora_messaging.api import consume
>    ...
>  
>    class Consumer(MessageConsumer):
>        def consume(self, message):
>            """ message _is_ instance of the schema class
> """
>            message.do_some_stuff()
>            if message.some_property:
>                do_something()
>  
>    consume(Consumer) 
Yeah, so long as the package that provides the schema class is
installed, this actually should work. I guess whether you write your
consumer to work with or without the schema class, or just have it blow
up if the schema class is not installed, is kinda up to you.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net