- Added an optional handler callback to request a last will message from the handler on each connect. If none is provided by the handler, the last will message, if any, set when initializing Connection is used as before.
- Switch to using
:ssl.handshakeinstead of:ssl.ssl_acceptfor OTP 24 compatibility.
- Ease the requirement on the gen_state_machine dependency such that is accept gen_state_machine 3.0.0
- Accepted some open PRs that has been waiting for a long while. These include allowing the user to pass in options to connections, and will categorize a connection error as "connectivity", such that tortoise will try to reconnect with back-off.
- Allow a timeout on async publish; this will allow the user to override the default timeout set when retrieving the connection; Note that this timeout will not work for the publish itself.
- A connection timeout is now categorized as a connectivity issue. Thanks to Riccardo Binetti and Frank Hunleth for this contribution.
- Pass on connection errors so we can back-off on connection timeouts.
- A use-macro has been added to
Tortoise.Handlerwhich implements default behaviour for the callbacks so the user only have to implement the callbacks they care about. This should improve usability of the project quite a bit.
- Avoid a possible race condition that could happen when a connection was requested while the connection was in the process of being established. This update should avoid hanging processes when a publish is attempted in the small window between going from connecting to connected.
- A usage example was added to the README-file. Thanks, Tanweer Shahzaad.
- The
Tortoisemodule does no longer delegate thesubscribe,unsubscribe, and their_syncvariants to theTortoise.Connectionmodule. They should just be accessed directly on theTortoise.Connectionmodule instead. This is done to make the documentation efforts easier, because conceptually the subscription belongs to the connection.
- Improvements has been made to the type specs and the documentation of various modules.
- A
ping/1andping_sync/2function has been added to theTortoise.Connectionmodule. This allow the user to ping the broker to check the connection. Notice that the connection process will still send ping messages to fulfill the connection keep alive timeout interval. There is no need to manually ping an open connection.
- Upgraded to a more recent version of dialyxir and made the project pass the dialyzer.
-
As according to the specification, an unsubscribe, and a subscribe package can no longer be encoded if it has no topic filters defined.
-
Got rid of TravisCI in favor of SemaphoreCI. Hopefully this will provide more reliable CI, and a funner experience for people working on Tortoise.
disconnect/1has been added to theTortoise.Connectionmodule. Given aclient_idit will close down the connection cleanly; in flight messages will get canceled and a disconnect package will be send to the broker.
-
When a connection drop the inflight manager will now re-transmit packages with quality of service that was inflight when the regain a connection.
-
Improved the logic for handling incoming publishes with Quality of Service 2 (exactly once delivery). The inflight manager will ensure we only handle the message once (as we should); possible duplicate messages caused by retransmisions should get trapped and not forwarded to the user defined publish handler.
-
It is now possible to pass in a binary/string as the host for a connection using the
Tortoise.Transport.SSLmodule; it will be coerced to a charlist.
- The documentation on how to connect to a broker using the
Tortoise.Transport.SSLmethod has been improved by Bram Verburg.
-
The controller will stop with an
{:protocol_violation, {:unexpected_package_from_remote, package}}if the server send a package of the typepackage. This is the right thing to do because servers should implement the protocol proper, and clients should not accept unexpected packages. -
Instead of using the Logger module to log the ping response times from the connection keep alive messages the ping response times are not dispatched using a the ETS based PubSub. This will allow the user to log, or do whatever they want with the ping response time.
-
Expose connection status changes via the
Tortoise.Eventspubsub making it possible to implement custom behavior when the connection goes down and when it becomes available. Great for setting up alarms. -
The receiver will now broadcast a "status down" message on the
Tortoise.Eventspubsub, and the connection process will listen for that message and enter the reconnect on status change.
Tortoise.Connection.reconnect/1has been removed as it is has been replaced with the pubsub based status-down message approach described in the "changed"-section. It might make a return at some point, but for now it has been removed.
-
Keep the active connection in a ETS based registry, allowing processes that publish messages to the wire, using either
Tortoise.publish/4orTortoise.Pipe.publish/4to obtain the network connection and shoot the message directly on the network socket. With this change we can also error out if a process tries to send on a non-existing connection. -
The implementation for
Tortoise.publish/4andTortoise.publish_sync/4has been moved to theTortoisemodule itself and is therefore no longer delegated to theTortoise.Connectionmodule. This changes the interface a bit but makes for a cleaner interface.
-
This release introduces a
Tortoise.Eventsmodule that implements aRegistrybased PubSub. Tortoise will use this to dispatch system events to listeners. For now it is used to dispatch new network connections, which is currently used byTortoise.publish, theTortoise.Connection.Inflightprocess, and theTortoise.Pipes. In the future we might add more message types toTortoise.Events. -
Tests for
Tortoise.publish/4andTortoise.publish_sync/4
-
The
Tortoise.Connection.Transmitterprocess is no longer needed, so it has been removed. -
Some dead code in the
Tortoise.Connection.Inflightmodule has been removed. This should not change anything user facing.
-
A server sending a ping request to a client is now considered a protocol violation, as specified by both the MQTT 3.1.1 and MQTT 5 specifications.
-
The connection process will now cancel the keep alive timer if it goes offline. Previously it would terminate itself because it would not get the ping response from the server.
-
Regression: The receiver will no longer crash on an assertion when it request a reconnect from the
Tortoise.Connectionprocess. -
The specified Elixir version in the mix.exs file should now allow more versions of Elixir without warnings.
- The receiver could report an invalid header length protocol violation when the server send too few bytes, and thus taking too long sending the complete package header. This patch will ensure we only report invalid header length when we have pulled more than 40 bits in.
-
Documentation. Lot's of documentation. It might be a bit wordy, and it might repeat itself here and there, but it is a start that can be improved upon and good enough for a 0.4.4 release.
-
Added some articles that describe usage of Tortoise and MQTT.
-
rename
renew/1toreconnect/1onTortoise.Connection-module. -
Some type specs has been altered to better fit reality.
-
The
Tortoise.Transport.Tcptransport will now accept a binary as a host name. Internally it will cast it to a char list.
- The wrong field was pulled out of the configuration options when the
last will message was set, so it was impossible to configure a last
will message. The last will message can now be set by passing in a
%Tortoise.Package.Publish{}struct as the:willwhen starting a connection.
- The
Tortoise.Connection.renew/1will now return:okon success; allowing theTorotise.Connection.Receiverto not crash on its assertion when it request a reconnect.
- Tortoise should now survive the server it is connected to being
restarted.
{:error, :econnrefused}and{:error, :closed}has been added to the errors that make tortoise attempt a reconnect.
-
Incremental backoff has been added to the connector, allowing us to retry reconnecting to the broker if the initial (or later reconnect) attempt fails. The backoff will per default start retrying after 100 ms and it will increment in multiples of 2 up until 30 seconds, at which point it will flip back to 100 ms and start over. This should ensure that we will be able to connect fairly quickly if it is a network fluke (or the network devise is not ready yet), and still not try too often or too quickly.
The backoff can be configured by passing
backoffcontaining a keyword list to the connection specification. Examplebackoff: [min_interval: 100, max_interval: 30_000]. Both times are in milliseconds.
-
The code for establishing a connection and eventually reconnecting has been combined into one. This makes it easier to test and verify, and it will make it easier to handle connection errors.
Because the initial connection is happening outside of the
init/1function the possible return values of theTortoise.Connection.start_link/1-function has changed a bit. A fatal error will cause the connection process to exit instead because the init will always return{:ok, state}. This might break some implementation using Tortoise.For now it is only
{:error, :nxdomain}that is handled with connection retries. Error categorization has been planned so we can error out if it is a non-recoverable error reason (such as no cacert files specified), instead of retrying the connection. In the near future more error reasons will be handled with reconnect attempts. -
A protocol violation from the server during connection will be handled better; previously it would error with a decoding error, because it would attempt to decode 4 random bytes. The error message should be obvious now.
-
Thanks to Troels Brødsgaard Tortoise now implement a module for its registry. This is found in
Tortoise.Registry. -
The user defined controller handler callback module now accept "next actions" in the return tuple; this allow the user to specify that a topic should get subscribed to, or unsubscribed from, by specifying a return like
{:ok, new_state, [{:subscribe, "foo/bar", qos: 3}, {:unsubscribe, "baz/quux"}]}.This is needed as the controller must not be blocked, and the user defined callback module run in the context of the controller. By allowing next actions like this the user can subscribe and unsubscribe to topics when certain events happen.
-
The test coverage tool will now ignore modules found in the lib/tortoise/handlers/-folder. These modules implement the
Tortoise.Handler-behaviour, so they should be good.
-
Tortoise.subscribe/3is now async, so a message will get sent to the mailbox of the calling process. The old behavior can be found in the newly createdTortoise.subscribe_sync/3that will block until the server has acknowledged the subscribe. -
Tortoise.unsubscribe/3is now also async, so like the subscribe a message will get sent to the mailbox of the calling process. The old behavior can be found in the newly addedTortoise.unsubscribe_sync/3that will block until the server has acknowledged the subscribe. -
A major refactorization of the code handling the logic running the user defined controller callbacks has been lifted from the
Tortoise.Connection.Controllerand put into theTortoise.Handlermodule. This change made it possible to support the next actions, and makes it much easier test and add new next action types in the future.
- Fix an issue where larger messages would crash the receiver. It has been fixed and tested with messages as large as 268435455 bytes; which is a pretty big MQTT message.
-
The
Tortoise.Transport.SSLwill now pass in[verify: :verify_peer]as the default option when connecting. This will guide the user to pass in a list of trusted CA certificates, such as one provided by the Certifi package, or opt out of this by passing theverify: :verify_noneoption; this will make it hard for the user to make unsafe choices.Thanks to Bram Verburg for this improvement.
-
Experimental SSL support, please try it out and provide feedback.
-
Abstract the network communication into a behaviour called
Tortoise.Transport. This behaviour specify callbacks needed to connect, receive, and send messages using a network transport. It also specify setting and getting options, as well as listening using that network transport; the latter part is done so they can be used in integration tests. -
A TCP transport has been created for communicating with a broker using TCP. Use
Tortoise.Transport.Tcpwhen specifying the server in the connection to use the TCP transport. -
A SSL transport
Tortoise.Transport.SSLhas been added to the project allowing us to connect to a broker using an encrypted connection.
- The
{:tcp, 'localhost', 1883}connection specification has been removed in favor of{Tortoise.Transport.Tcp, host: 'localhost', port: 1883}. This is done because we support multiple transport types now, such as theTortoise.Transport.SSLtype (which also takes akeyand acertoption). The format is{transport, opts}.
-
The project is now on Hex which will hopefully broaden the user base. Future changes will be logged to this file.
-
We will from now on update the version number following Semantic Versioning, and major changes should get logged to this file.