New 4.3.0 release breaks Redis Sentinel connection

[jobec]

The new 4.3.0 breaks the Redis Sentinel connection, triggering an error like when queueing a task:

redis.sentinel.MasterNotFoundError: No master found for 'celery-master'

While starting a celery worker gives this error

 ERROR - celery.worker.consumer.consumer - consumer: Cannot connect to sentinel://localhost:26379/0: No master found for 'celery-master'.

kombu==4.2.2-post1 did not have this issue.

Redis library = 3.0.1

[auvipy]

can you help to bisect the commit causing this?

[jobec]

Looks like it happened in this commit: a59b44e

When running though a debugger, self.connection.client.alt is empty on this code line

kombu/kombu/transport/redis.py

Line 1104 in b4be5cf

for url in self.connection.client.alt:

This causes the next part to connect to an empty list of sentinels:

kombu/kombu/transport/redis.py

Lines 1107 to 1111 in b4be5cf

	sentinel_inst = sentinel.Sentinel(
	connection_list,
	min_other_sentinels=getattr(self, 'min_other_sentinels', 0),
	sentinel_kwargs=getattr(self, 'sentinel_kwargs', None),
	**additional_params)

[jobec]

Celery (Django) setting:

CELERY_BROKER_URL = "sentinel://localhost:26379/0"
CELERY_BROKER_TRANSPORT_OPTIONS = {
    'master_name': 'celery-master',
}

[auvipy]

@badcure could you check?

[lithammer]

I'm thinking that maybe something like this ought to fix it. But it's just guesses.

-        connection_list = []
+        connection_list = [(connparams['host'], connparams['port'])]
         for url in self.connection.client.alt:
             url = _parse_url(url)
             if url.scheme == 'sentinel':
                 connection_list.append([url.hostname, str(url.port)])

Or:

         connection_list = []
         for url in self.connection.client.alt:
             url = _parse_url(url)
             if url.scheme == 'sentinel':
                 connection_list.append([url.hostname, str(url.port)])
+        if not connection_list:
+            connection_list = [(connparams['host'], connparams['port'])]

[lithammer]

Tried to fix this in #1010.

[hajarada]

Hello All, I know this is issue is for 4.3, but it seems that 4.2.1 has that issue as well. and the same workaround applies.
In my setup we have 8 sentinels sitting behind two loadbalancers that use GSLB, so originally all consumers were configured to talk to the GSLB hostname address which translates to one of the two LB IP addresses. two weeks ago following a routine libraries upgrade, all celery applications stopped working with the following trace:

==> /apps/run/logs/app_name.internal.worker.log <==
self.send_packed_command(self.pack_command(*args))
File "/apps/app_name/lib/python3.6/site-packages/redis/connection.py", line 594, in send_packed_command
self.connect()
File "/apps/app_name/lib/python3.6/site-packages/redis/sentinel.py", line 45, in connect
self.connect_to(self.connection_pool.get_master_address())
File "/apps/app_name/lib/python3.6/site-packages/redis/sentinel.py", line 101, in get_master_address
self.service_name)
File "/apps/app_name/lib/python3.6/site-packages/redis/sentinel.py", line 224, in discover_master
raise MasterNotFoundError("No master found for %r" % (service_name,))
kombu.exceptions.OperationalError: No master found for 'app_name'

I changed the broker_url to be the two LB names instead of the GSLB name and that seems to have fixed it.

Here is the pip list
ackage Version

---

amqp 2.4.1
asn1crypto 0.24.0
bcrypt 3.1.6
billiard 3.6.0.0
celery 4.2.1
certifi 2018.11.29
cffi 1.12.2
chardet 3.0.4
Click 7.0
cryptography 2.5
Flask 1.0.2
Flask-Auth 0.85
Flask-HTTPAuth 3.2.4
Flask-Session 0.3.1
idna 2.8
itsdangerous 1.1.0
Jinja2 2.10
kombu 4.3.0
ldap3 2.5.2
MarkupSafe 1.1.1
mod-wsgi 4.6.5
packaging 19.0
paramiko 2.4.2
pip 19.0.3
pip-review 1.0
pyasn1 0.4.5
pycparser 2.19
pycrypto 2.6.1
pymongo 3.7.2
PyNaCl 1.3.0
pyparsing 2.3.1
pytz 2018.9
redis 3.2.0
requests 2.21.0
scp 0.13.0
setuptools 40.8.0
six 1.12.0
urllib3 1.24.1
vine 1.2.0
Werkzeug 0.14.1