An ASGI channel layer that uses Redis as its backing store, and supports both a single-server and sharded configurations, as well as group support.
You'll need to instantiate the channel layer with at least hosts
,
and other options if you need them.
Example:
channel_layer = RedisChannelLayer(
host="redis",
db=4,
channel_capacity={
"http.request": 200,
"http.response*": 10,
}
)
The server(s) to connect to, as either URIs or (host, port)
tuples. Defaults to ['localhost', 6379]
. Pass multiple hosts to enable sharding, but note that changing the host list will lose some sharded data.
Prefix to add to all Redis keys. Defaults to asgi:
. If you're running
two or more entirely separate channel layers through the same Redis instance,
make sure they have different prefixes. All servers talking to the same layer
should have the same prefix, though.
Message expiry in seconds. Defaults to 60
. You generally shouldn't need
to change this, but you may want to turn it down if you have peaky traffic you
wish to drop, or up if you have peaky traffic you want to backlog until you
get to it.
Group expiry in seconds. Defaults to 86400
. Interface servers will drop
connections after this amount of time; it's recommended you reduce it for a
healthier system that encourages disconnections.
Default channel capacity. Defaults to 100
. Once a channel is at capacity,
it will refuse more messages. How this affects different parts of the system
varies; a HTTP server will refuse connections, for example, while Django
sending a response will just wait until there's space.
Per-channel capacity configuration. This lets you tweak the channel capacity based on the channel name, and supports both globbing and regular expressions.
It should be a dict mapping channel name pattern to desired capacity; if the
dict key is a string, it's intepreted as a glob, while if it's a compiled
re
object, it's treated as a regular expression.
This example sets http.request
to 200, all http.response!
channels
to 10, and all websocket.send!
channels to 20:
channel_capacity={
"http.request": 200,
"http.response!*": 10,
re.compile(r"^websocket.send\!. "): 20,
}
If you want to enforce a matching order, use an OrderedDict
as the
argument; channels will then be matched in the order the dict provides them.
Pass this to enable the optional symmetric encryption mode of the backend. To
use it, make sure you have the cryptography
package installed, or specify
the cryptography
extra when you install asgi_redis
:
pip install asgi_redis[cryptography]
symmetric_encryption_keys
should be a list of strings, with each string
being an encryption key. The first key is always used for encryption; all are
considered for decryption, so you can rotate keys without downtime - just add
a new key at the start and move the old one down, then remove the old one
after the message expiry time has passed.
Data is encrypted both on the wire and at rest in Redis, though we advise you also route your Redis connections over TLS for higher security; the Redis protocol is still unencrypted, and the channel and group key names could potentially contain metadata patterns of use to attackers.
Keys should have at least 32 bytes of entropy - they are passed through the SHA256 hash function before being used as an encryption key. Any string will work, but the shorter the string, the easier the encryption is to break.
If you're using Django, you may also wish to set this to your site's
SECRET_KEY
setting via the CHANNEL_LAYERS
setting:
CHANNEL_LAYERS = {
"default": {
"BACKEND": "asgi_redis.RedisChannelLayer",
"ROUTING": "my_project.routing.channel_routing",
"CONFIG": {
"hosts": ["redis://:[email protected]:6379/0"],
"symmetric_encryption_keys": [SECRET_KEY],
},
},
}
Optional extra arguments to pass to the redis-py
connection class. Options
include socket_connect_timeout
, socket_timeout
, socket_keepalive
,
and socket_keepalive_options
. See the
redis-py documentation for more.
A "local and remote" mode is also supported, where the Redis channel layer
works in conjunction with a machine-local channel layer (asgi_ipc
) in order
to route all normal channels over the local layer, while routing all
single-reader and process-specific channels over the Redis layer.
This allows traffic on things like http.request
and websocket.receive
to stay in the local layer and not go through Redis, while still allowing Group
send and sends to arbitrary channels terminated on other machines to work
correctly. It will improve performance and decrease the load on your
Redis cluster, but it requires all normal channels are consumed on the
same machine.
In practice, this means you MUST run workers that consume every channel your application has code to handle on the same machine as your HTTP or WebSocket terminator. If you fail to do this, requests to that machine will get routed into only the local queue and hang as nothing is reading them.
To use it, just use the asgi_redis.RedisLocalChannelLayer
class in your
configuration instead of RedisChannelLayer
and make sure you have the
asgi_ipc
package installed; no other change is needed.
"Sentinel" mode is also supported, where the Redis channel layer will connect to a redis sentinel cluster to find the present Redis master before writing or reading data.
Sentinel mode supports sharding, but does not support multiple Sentinel clusters. To run sharding of keys across multiple Redis clusters, use a single sentinel cluster, but have that sentinel cluster monitor multiple "services". Then in the configuration for the RedisSentinelChannelLayer, add a list of the service names. You can also leave the list of services blank, and the layer will pull all services that are configured on the sentinel master.
Redis Sentinel mode does not support URL-style connection strings, just tuple-based ones.
Configuration for Sentinel mode looks like this:
CHANNEL_LAYERS = {
"default": {
"BACKEND": "asgi_redis.RedisSentinelChannelLayer",
"CONFIG": {
"hosts": [("10.0.0.1", 26739), ("10.0.0.2", 26379), ("10.0.0.3", 26379)],
"services": ["shard1", "shard2", "shard3"],
},
},
}
The "shard1", "shard2", etc entries correspond to the name of the service configured in your
redis sentinel.conf file. For example, if your sentinel.conf says sentinel monitor local 127.0.0.1 6379 1
then you would want to include "local" as a service in the RedisSentinelChannelLayer configuration.
You may also pass a sentinel_refresh_interval
value in the CONFIG
, which
will enable caching of the Sentinel results for the specified number of seconds.
This is recommended to reduce the need to query Sentinel every time; even a
low value of 5 seconds will significantly reduce overhead.
Redis >= 2.6 is required for asgi_redis. It supports Python 2.7, 3.4, 3.5 and 3.6.
Please refer to the main Channels contributing docs. That also contains advice on how to set up the development environment and run the tests.
To report security issues, please contact [email protected]. For GPG signatures and more security process information, see https://docs.djangoproject.com/en/dev/internals/security/.
To report bugs or request new features, please open a new GitHub issue.
This repository is part of the Channels project. For the shepherd and maintenance team, please see the main Channels readme.