Remove all `include ActiveSupport::Testing::MethodCallAssertions`
in actioncable's tests since we can do it only in `ActionCable::TestCase`
in order to prevent code duplication.
We use the same approach for other modules of Rails.
We have defined `ActionCable::TestCase` in `actioncable/test/test_helper.rb`
that we can use in order to prevent code duplication and build common
interface for actioncable's test.
Simplify our dev testing and CI story since we're also testing against
Redis for the Active Support cache store.
Directly test whether db, host, password, etc are passed through as
config instead of spinning up a Redis server with a password set on it.
- Adds RedisAdapterTest::AlternateConfiguration to account
for a relatively common alternative setup, as it’s used
as the first example in the
[Redis rubygem](https://github.com/redis/redis-rb#getting-started)
- Supplies original RedisAdapterTest with more complete
redis:// url format by adding a ‘userinfo’ (blank user),
so that it resembles the alternate configuration
- Supplies original EventedRedisAdapterTest with more complete
redis:// url as well
- Adds before_script to start redis-server with password as a daemon
and with explicit defaults copied from the default redis.conf
(Instead of using Travis' default init/upstart scripts for `redis` service)
Before this commit, the database connection used in Action Cable's
PostgreSQL adapter was "owned" by `ActiveRecord::Base.connection_pool`.
This meant that if, for example, `#clear_reloadable_connections!` was called on the pool, Active
Record would "steal" the database connection from Action Cable, and
would cause all sorts of issues. This became evident during file
reloads; despite Action Cable trying its hardest to return its borrowed
database connection to Active Record via `@pubsub.shutdown`, Active Record calls
`#clear_reloadable_connections!` on the connection pool, and due to the order of callbacks, Active
Record's callback was being executed first. This meant that if you tried
to rerender a view after a file was reloaded, you would have to wait
through Active Record's timeout and such.
Now, Action Cable takes direct ownership of the database connection it
uses. It removes the connection from the pool to avoid the situation
described above. Action Cable also makes sure to call `#disconnect!` on
the connection when appropriate, to match the previous behavior of
Active Record.
[ Jon Moss & Matthew Draper]
reactor_running? will be true just after the thread enters
EventMachine.run; reactor_thread only gets set after the internal
initialize_event_machine method has been called, the C extension is set
up, and it is entering its run loop.
* Introduce a connection coder responsible for encoding Cable messages
as WebSocket messages, defaulting to `ActiveSupport::JSON` and duck-
typing to any object responding to `#encode` and `#decode`.
* Consolidate encoding responsibility to the connection. No longer
explicitly JSON-encode from channels or other sources. Pass Cable
messages as Hashes to `#transmit` and rely on it to encode.
* Introduce stream encoders responsible for decoding pubsub messages.
Preserve the currently raw encoding, but make it easy to use JSON.
Same duck type as the connection encoder.
* Revert recent data normalization/quoting (#23649) which treated
`identifier` and `data` values as nested JSON objects rather than as
opaque JSON-encoded strings. That dealt us an awkward hand where we'd
decode JSON strings… or not, but always encode as JSON. Embedding
JSON object values directly is preferably, no extra JSON encoding,
but that should be a purposeful protocol version change rather than
ambiguously, inadvertently supporting multiple message formats.
`skip` raises an exception to abort the execution of the test, so
`super` would never be called and thus `@rx_adapter` and `@tx_adapter`
would never have been defined at the time of teardown.
Define them just before skipping and zap the warnings.
EM::Hiredis were spewing screenfuls of warnings when running the Action Cable tests.
Copied over the technique that shushes up faye-websocket in the client tests, so
we can reduce the noise ratio.
Note: there's still warnings spewed after tests have finished when EM::Hiredis shuts
down. I haven't been able to shush them up yet.
The `WorkerTest`'s `Receiver` is imporsonating an `ActionCable::Connection::Base`, but
just delegates the logger to `ActionCable.logger`.
This creates a mismatch as the connection requires the logger to be a
`TaggedLoggerProxy`'ied logger, while the server doesn't.
Thus to ensure an exception isn't raised when the worker tries to call `tag`
other tests have to assign a proxied logger to their test server.
Instead of forcing change on other tests, have Receiver adhere to the connection
contract and use a `TaggedLoggerProxy`.
As a consequence remove more setup from the tests.
Instead of depending on ApplicationCable::Connection being defined at initialize
we should inject it in the Railtie.
Thus we can kill more setup in the tests too.
We were explicitly referencing Rails.root in ActionCable::Server::Configuration.initialize,
thereby coupling ourselves to Rails.
Instead add `app/channels` to Rails' app paths and assign the existent files
to `channel_paths`.
Users can still append to those load paths with `<<` and `push` in `config/application.rb`.
This means we can remove the custom `Dir` lookup in `channel_paths` and the Rails
and root definitions in the tests.
This new adapter does get a little more intimate with the redis-rb gem's
implementation than I would like, but it's the least bad of the
approaches I've come up with.
We're no longer doing our work in the EM event loop, so errors are quite
unlikely, and if they do occur, they're not really our responsibility to
handle.
