RabbitMQ management api can't connect

[gentunian]

Describe the bug

When opening the "broker" tab, nothing is shown.

To Reproduce

1. Use this docker-compose.yml:

version: '3'

services:

  broker:
    image: rabbitmq:3.8-management

  monitor:
    image: mher/flower:0.9.5
    command:  flower --address=0.0.0.0 --port=5555 --broker=$CELERY_BROKER_URL --broker_api=$CELERY_BROKER_API_URL
    environment:
      CELERY_BROKER_URL: amqp://broker:5672
      CELERY_BROKER_API_URL: http://guest:guest@broker:15672/api
    ports:
      - 5555:5555

2. run it: docker-compose up
3. Open http:localhost:5555/broker

Expected behavior

Flower showing rabbitmq management api connection.

Actual behavior

broker_1   | {bad_header,<<"GET /api">>}
monitor_1  | [E 201030 16:57:13 broker:84] RabbitMQ management API call failed: [Errno 104] Connection reset by peer

[pinoatrome]

+1

[pinoatrome]

Hi,
I managed to avoid the 'bad_header' error in rabbitMQ changing the name of the BROKER_API_URL on the docker-compose.yml file.

from

CELERY_BROKER_API_URL: http://guest:guest@broker:15672/api

to

FLOWER_BROKER_API_URL: http://guest:guest@broker:15672/api

HTH
GG

[marcelolima]

@pinoatrome it's just a variable name change, would you mind sharing how it would fix the problem? (the URL is still the same)

[pinoatrome]

@marcelolima
Hi Marcelo,
you are correct: the URL is the same and uses 15672 (management port) instead of 5672 (applicative port), using 'FLOWER_BROKER_API_URL' instead of 'CELERY_BROKER_API_URL' results in a app.options.broker_api being populated (with the management URL passed) instead of building the api_url in RabbitMQ(BrokerBase) constructor with a default format (that is using the applicative port).

This is what I got from a debugging session:

in class BrokerView(BaseHandler) (views/broker.py:14)
in case the transport is 'amqp' (true when RabbitMQ is used as broker) this assignment

http_api = app.options.broker_api

results in a not null value for http_api only if FLOWER_BROKER_API_URL is specified.

This avoids to build a api_url using the default pattern in class RabbitMQ(BrokerBase) (utils/broker.py: 41)
that fails when asking for queues because it is using the port 5672 instead of the management port 15672.

class BrokerView(BaseHandler):
    @web.authenticated
    @gen.coroutine
    def get(self):
        app = self.application
        broker_options = self.capp.conf.BROKER_TRANSPORT_OPTIONS

        http_api = None
        if app.transport == 'amqp' and app.options.broker_api:
            http_api = app.options.broker_api

if Broker is instantiated with a null http_api argument (not specifying FLOWER_BROKER_API_URL) the default value assigned to http_api is used:

'http://guest:guest@localhost:5672/api//' <-- is not using the 'management' port (15672) but the applicative port (5672)

class RabbitMQ(BrokerBase):
        ....
        self.vhost = quote(self.vhost, '') or '/' if self.vhost != '/' else self.vhost
        ....
        if not http_api:
            http_api = "http://{username}:{password}@{host}:{port}/api/{vhost}".format(
                username=self.username, password=self.password,
                host=self.host, port=self.port, vhost=self.vhost
            )

Then in queues this http_api results in url with port 5672 instead of 15672; when invoked generates a HTTPClientError with message:
"RabbitMQ management API call failed: Stream closed"

    @gen.coroutine
    def queues(self, names):
        url = urljoin(self.http_api, 'queues/' + self.vhost)

Hope this Helps
Ciao

[marcelolima]

thank you, @pinoatrome !
It wasn't the problem I was facing but after checking your response I also went to check the source code and found the reason:

I was passing: http://guest:guest@broker:15672/api instead of http://guest:guest@broker:15672/api/vhost (vhost at the end). That's why I was getting a 404 before.

[eddyojb88]

I think that there is a significant series of inconsistencies with using a dockerized flower.

1. Pointing the command in the docker-compose file to an $env_var seems to command flower to default to Redis

2. When setting environment variables, there seems to be a random success rate as to whether a celery worker is picked up or not in the Flower UI

3. There is a random success rate upon first firing up docker-compose and it successfuly calling the API and displaying something in the UI. Refreshing the page results in a stalling of the API.

Can someone point me to a working configuration of django, database, celery, rabbitmq and flower in docker? Flower is looking to be coming out of my stack at this point, having spent a day trying to resolve this one.

[0mars]

I started seeing all of these issues, since I added a vhost

[JayDarji]

Those who're still wondering and running flower with celery flower command directly, ie like

celery --broker=amqp://guest:guest@localhost:5672// flower -A project_name --basic_auth=guest:guest

and facing this problem that isn't able to get queues in broker tab of flower, can add a little extra argument, and then it'll work.

--broker_api=http://guest:guest@localhost:15672/api/vhost

So now, full command should be:

celery --broker=amqp://guest:guest@localhost:5672// flower -A project_name --basic_auth=guest:guest --broker_api=http://guest:guest@localhost:15672/api/vhost

Here, if your rabbitmq and flower running on same server, you can build urls with localhost in broker and broker_api parameters.
If that is not a case, then you can pass the public ip of the server where rabbitmq is running.

Also, in broker we're passing rabbitmq-server port and in broker_api we're passing rabbitmq management plugins port.
If rabbitmq management plugin is not enabled, please enable that because flower gets queues info from rabbitmq management plugin API.

[adv27]

Thank @JayDarji 🙏
The current documentation on the latest(2.0.0) still seems wrong
https://flower.readthedocs.io/en/2.0/config.html#broker-api

$ celery flower broker-api="http://username:password@rabbitmq-server-name:15672/api/"

The correct one should be

$ celery flower --broker_api="http://username:password@rabbitmq-server-name:15672/api/vhost"

[redliu312]

@JayDarji

wow, you answer really solved my issue.
Thank you

[ismaelzsilva]

@adv27 answer solved it, thank you!