bunny 1.3.0 → 2.17.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
-SHA1:
-  metadata.gz: 972f0f0055bbc831144da54e6805c4d5f5e2262c
-  data.tar.gz: ad2f3539fb6946771238e6451b17d33cfdba692b
+SHA256:
+  metadata.gz: 3f3ba4a5565a42eace00f8789c1be953871d9dc57d42028a94a9dd70a586271e
+  data.tar.gz: c7d2fdf832b466a27c7ae884e6195729d31a6bf1cb6accde9818c2f72b46cdc6
 SHA512:
-  metadata.gz: ad93a02f5623437589eb2027259de36a2269df8ab06d251b00e6b28b265f8f562068a1c7e473b2513eff18bdde77902b46a24c114a9a1c9dbe29426c93f67085
-  data.tar.gz: da94e1d8349674cf2819678f2fb18a1fd911017fb7c99104d64d755c9d7f9622068db60a5502411c6442d04fe341372d09d31c6e5885c0cda18293ea04db3e81
+  metadata.gz: 26a352492ca652429975e45ec36fad12ebfd3c94e66babb879afb0f09dcec0158affc820f8a29fd07a25d8db6ac9312fd31ebc5b1a3d4c315ecfeeb880a5a8f2
+  data.tar.gz: 393fa99fb87c7784eccd0f5019fe3668b39064e15935d85f4134c85aa6b138849d0026addb814f49d00f606b9157b78878863c875d2ad6480b17b3053dcb98fb

data/.github/ISSUE_TEMPLATE.md

@@ -0,0 +1,18 @@
+## Does This Really Belong to GitHub issues?
+
+If you find a bug you understand well, poor default, incorrect or unclear piece of documentation,
+or missing feature, please [file an
+issue](http://github.com/ruby-amqp/bunny/issues) on GitHub.
+
+Please use [Bunny's mailing list](http://groups.google.com/group/ruby-amqp) for questions,
+investigations, and discussions. GitHub issues should be used for specific, well understood, actionable
+maintainers and contributors can work on.
+
+When filing an issue, please specify
+
+ * Which Bunny and RabbitMQ versions are used
+ * Recent RabbitMQ log file contents
+ * Full exception stack traces
+ * Steps to reproduce or a failing test case
+
+This would greatly help the maintainers help you.

data/.gitignore

@@ -11,7 +11,7 @@ Gemfile.lock
 .tags
 .tags_sorted_by_file
 .Apple*
-bin/*
+/bin/*
 .bundle/*
 vendor/*
 playground/*
@@ -20,3 +20,9 @@ repl-*
 debug/*
 *.dump
 deploy.docs.sh
+.ruby-version
+.idea
+*.srl
+spec/tls/*.pem
+spec/tls/*.pem~
+spec/tls/*.p12

data/.rspec

@@ -1,3 +1 @@
---color
---format
-progress
+-c -fp

data/.travis.yml

@@ -1,15 +1,21 @@
+dist: bionic
 language: ruby
 bundler_args: --without development
-before_script: "./bin/ci/before_build.sh"
-script: "bundle exec rspec -cfs spec"
+cache: bundler
+before_install:
+  - gem install bundler
+before_script:
+  - "./bin/ci/install_on_debian.sh"
+  - until sudo lsof -i:5672; do echo "Waiting for RabbitMQ to start..."; sleep 1; done
+  - "./bin/ci/before_build.sh"
+script: "bundle exec rake integration_without_recovery"
 rvm:
-  - "2.1.0"
-  - "2.0"
-  - "1.9.3"
-  - "jruby"
-  - "1.9.2"
-  - "rbx"
-  - "1.8.7"
+  - ruby-head
+  - "2.7.1"
+  - "2.6.6"
+  - "2.5.8"
+  - "2.4.10"
+  - "2.3.8"
 notifications:
   email: michael@rabbitmq.com
 services:
@@ -17,10 +23,11 @@ services:
 branches:
   only:
     - master
-    - 1.1.x-stable
+    - 2.14.x-stable
+    - 2.13.x-stable
+env:
+  - CI=true
 matrix:
   allow_failures:
-    - rvm: rbx
-    - rvm: "1.9.2"
-    - rvm: "1.8.7"
-    - rvm: jruby
+    rvm:
+      - ruby-head

data/CONTRIBUTING.md

@@ -0,0 +1,132 @@
+## Overview
+
+This project **does not** use GitHub issues for questions, investigations, discussions, and so on.
+Issues are appropriate for something specific enough for a maintainer or contributor to work on:
+
+ * There should be enough information to reproduce the behavior observed in a reasonable amount of time
+ * It should be reasonably clear why the behavior should be changed and why this cannot or should not be addressed
+   in application code, a separate library and so on
+
+ All issues that do not satisfy the above properties belong to the [Ruby RabbitMQ clients mailing list](http://groups.google.com/forum/#!forum/ruby-amqp). Pull request that do not satisfy them have a high chance
+ of being closed.
+
+## Submitting a Pull Request
+
+Please read the sections below to get an idea about how to run Bunny test suites first. Successfully
+running all tests, at least with `CI` environment variable exported to `true`, is an important
+first step for any contributor.
+
+Once you have a passing test suite, create a branch and make your changes on it.
+When you are done with your changes and all
+tests pass, write a [good, detailed commit message](http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html) submit a pull request on GitHub.
+
+## Pre-requisites
+
+The project uses Bundler for dependency management and requires RabbitMQ `3.5+` to be running
+locally with the `rabbitmq-management` and `rabbitmq_consistent_hash_exchange` plugins enabled.
+
+### Running the Specs
+
+The specs require RabbitMQ to be running locally with a specific set of virtual hosts
+and users. RabbitMQ can be provisioned and started any way that's convenient to you
+as long as it has a suitable TLS keys configuration and management plugin enabled.
+Make sure you have a recent version of RabbitMQ (> `3.7.10`).
+
+The test suite can either use a locally available RabbitMQ node ([generic binary builds](http://www.rabbitmq.com/install-generic-unix.html)
+are an option that works well) or by running a RabbitMQ server in a Docker container.
+
+### Using a locally installed RabbitMQ node
+
+It is possible to start a local RabbitMQ node from the repository root. It is not necessarily
+optimal but can be a good starting point but is a useful example:
+
+```
+RABBITMQ_NODENAME=bunny RABBITMQ_CONFIG_FILE=./spec/config/rabbitmq.conf RABBITMQ_ENABLED_PLUGINS_FILE=./spec/config/enabled_plugins rabbitmq-server
+```
+
+The specs need the RabbitMQ management plugin to be enabled and include TLS connectivity tests,
+so the node must be configured to use a [certificate and key pair](http://www.rabbitmq.com/ssl.html#certificates-and-keys).
+The config and enabled plugin files in the spec/config directory take care of that
+but certificates must be provisioned locally. By default there's a set of CA, server, and client certificates pre-generated at `spec/tls`.
+
+The `BUNNY_CERTIFICATE_DIR` environment variable can be used to a directory containing a CA certificate
+and a certificate/key pair to be used by the server. The directory can be generated using
+[tls-gen](https://github.com/michaelklishin/tls-gen)'s basic profile. This option is recommended.
+
+`BUNNY_RABBITMQ_HOSTNAME` can be used to override the expected server hostname for [peer verification](http://www.rabbitmq.com/ssl.html#peer-verification) in the TLS test suite:
+
+```
+BUNNY_CERTIFICATE_DIR="/path/to/tls-gen/basic/result" BUNNY_RABBITMQ_HOSTNAME="mayflower" bundle exec rspec
+
+```
+
+Certificates can be generated with [tls-gen](https://github.com/michaelklishin/tls-gen)'s basic profile.
+In that case they include a Subject Alternative Name of `localhost` for improved portability.
+
+
+### Node Setup
+
+There is also a script that preconfigured the node for Bunny tests. It is sufficient to run
+it once but if RabbitMQ is reset it has to be executed again:
+
+```
+RABBITMQ_NODENAME=bunny ./bin/ci/before_build
+```
+
+The script uses `rabbitmqctl` and `rabbitmq-plugins`
+to set up RabbitMQ in a way that Bunny test suites expect. Two environment variables,
+`RABBITMQCTL` and `RABBITMQ_PLUGINS`, are available to control what `rabbitmqctl` and
+`rabbitmq-plugins` commands will be used. By default they are taken from `PATH`
+and prefixed with `sudo`.
+
+And then run the core integration suite:
+
+```
+RABBITMQ_NODENAME=bunny CI=true rspec
+```
+
+#### Running a RabbitMQ server in a Docker container
+
+First off you have to [install Docker Compose](https://docker.github.io/compose/install/) (and by proxy Docker).
+Version >= 1.6.0+ is required for compose version 2 syntax.
+
+After those have been installed (and the `docker-compose` command is available on your command line path), run
+
+```
+docker-compose build && docker-compose run --service-ports rabbitmq
+```
+
+The first time you do this, it will take some time, since it has to download everything it needs
+to build the Docker image.
+
+The RabbitMQ server will run in the foreground in the terminal where you started it. You can stop
+it by pressing CTRL+C. If you want to run it in the background, pass `-d` to `docker-compose`.
+
+### Toxiproxy
+
+If Toxiproxy is running locally on standard ports or started via Docker:
+
+```
+docker-compose run --service-ports toxiproxy
+```
+
+then Bunny will run additional resiliency tests.
+
+### Running Test Suites
+
+Prior to running the tests, configure the RabbitMQ permissions by running `./bin/ci/before_build`
+if you have RabbitMQ locally installed, if you are running RabbitMQ via Docker as above this step
+is not required as the setup is baked in.
+
+Make sure you have those two installed and then run integration tests:
+
+    bundle install
+    rake integration
+
+It is possible to run all tests:
+
+    bundle exec rspec
+
+It is possible to run only integration and regression tests but exclude unit and stress tests:
+
+    CI=true bundle exec rspec spec/higher_level_api/ spec/lower_level_api spec/issues spec/higher_level_api/integration/connection_recovery_spec.rb

data/ChangeLog.md

@@ -1,3 +1,889 @@
+## Changes between Bunny 2.17.x and 2.18.0 (in development)
+
+No changes yet.
+
+
+## Changes between Bunny 2.16.x and 2.17.0 (Sep 11th, 2020)
+
+### Easier to Specify a Client-Propvided Connection Name
+
+It is now easier to provide a client-provided (custom) connection
+name that will be displayed in the RabbitMQ management UI and mentioned in
+[server logs](https://www.rabbitmq.com/logging.html).
+
+Instead of
+
+``` ruby
+conn = Bunny.new(client_properties: {connection_name: 'app ABC #{rand}'})
+conn.start
+```
+
+a new top-level connection option now can be used:
+
+``` ruby
+conn = Bunny.new(connection_name: 'app ABC #{rand}')
+conn.start
+```
+
+Contributed by @brerx.
+
+GitHub issue: [ruby-amqp/bunny#600](https://github.com/ruby-amqp/bunny/pull/600)
+
+
+## Changes between Bunny 2.15.0 and 2.16.0 (Aug 14th, 2020)
+
+### Asynchronous Exception Delegate
+
+Bunny now can delete asynchronous connection (`Bunny::Session`) exception to an arbitrary
+delegate object. Use the `:session_error_handler` connection setting to pass it.
+The value defaults to `Thread.current`.
+
+Contributed by @bbascarevic.
+
+GitHub issue: [ruby-amqp/bunny#597](https://github.com/ruby-amqp/bunny/issues/597)
+
+
+## Changes between Bunny 2.14.0 and 2.15.0 (Apr 8th, 2020)
+
+### More Defensive Thread Join Operations
+
+Bunny is now more defensive around thread join operations which it performs
+when stopping its consumer work pool.
+
+`Thread#join` can cause an unhandled exception to be re-raised at
+a very surprising moment. This behavior can also be affected by 3rd party
+libraries, e.g. those that do connection pooling. While Bunny cannot
+fully avoid every possible surprising failure, it now avoids at least
+one such problematic interaction triggered by a custom [interrupt handler](https://ruby-doc.org/core-2.5.1/Thread.html#method-c-handle_interrupt)
+in a 3rd party library.
+
+GitHub issue: [#589](https://github.com/ruby-amqp/bunny/issues/589)
+
+Contributed by @fuegas.
+
+### Dependency Updates
+
+`amq-protocol` dependency has been bumped to `2.3.1` to support `connection.update-secret`
+protocol extension.
+
+### Gem Installation Fixed on Windows
+
+`bin/ci`, a directory with symlinks, is no longer included into the gem.
+
+Contributed by Jack Xiaosong Xu.
+
+### Lazy Peer Certificate Chain Information Logging
+
+Peer certificate chain information is now logged lazily, which prevents
+an obscure exception originating ASN.1 parser and makes the logging
+code evaluate only when it is really necessary.
+
+GitHub issue: [#578](https://github.com/ruby-amqp/bunny/pull/578)
+
+Contributed by Garrett Thornburg.
+
+
+## Changes between Bunny 2.13.0 and 2.14.0 (Feb 20th, 2019)
+
+### Improved Peer Verification Failure Logging
+
+When [peer verification](https://www.rabbitmq.com/ssl.html#peer-verification) fails, the connection will now log
+some relevant peer certificate chain details. If Bunny
+log level is set to `debug`, the same information will be logged
+unconditionally.
+
+### Closing Connections without Waiting for Response
+
+`Bunny::Session#close` now accepts a parameter that controls whether
+it waits for a `connection.close-ok` frame. Not waiting is useful
+when it is known for a fact that the node might not respond
+(it might be shutting down, connection is known to be interrupted
+or unrecoverable and so on) or waiting is irrelevant to the caller.
+
+### Successful Connection Recovery Notification
+
+`Bunny::Session#after_recovery_completed` (accepts a block)
+and a new connection option, `:recovery_completed` (a callable object)
+can be used to react to successful connection and topology recovery.
+
+GitHub issue: [#573](https://github.com/ruby-amqp/bunny/pull/573).
+
+Contributed by Ionut Popa.
+
+### effin_utf8 Dependency Dropped
+
+This library no longer supports Ruby 1.8 and thus
+doesn't need to depend on the `effin_utf8` gem.
+
+Contributed by Luciano Sousa.
+
+
+## Changes between Bunny 2.12.0 and 2.13.0 (Dec 25th, 2018)
+
+### More Defensive `Bunny::Channel` Method(s)
+
+`Bunny::Channel#queue` will now throw an `ArgumentError` if a `nil`
+is passed for queue name.
+
+GitHub issue: [#570](https://github.com/ruby-amqp/bunny/issues/570)
+
+### Correct Logging of Recovery Attempts Left
+
+During connection recovery, if `recover_attempts` is not set (is `nil`)
+connection could produce confusing log messages.
+
+GitHub issue: [#569](https://github.com/ruby-amqp/bunny/issues/569)
+
+
+## Changes between Bunny 2.11.0 and 2.12.0 (Sep 22nd, 2018)
+
+### More Defensive Treatment of `queue.declare-ok` Responses
+
+Responses for `queue.declare` are now checked against a memoized
+queue name (but only if the queue is not server-named). This helps
+avoids scenarios with overlapping/concurrent requests due to high
+network latency as demonstrated in [#558](https://github.com/ruby-amqp/bunny/issues/558).
+
+"Mismatched" responses will be ignored: Bunny channel API would throw
+an exception for such declarations and there would be no way to "return to"
+even if a matching response arrived and was matched with one of the pending
+requests in a reasonable period of time.
+
+As part of this work a new Toxiproxy-based test suite was introduced
+to Bunny.
+
+GitHub issue: [#558](https://github.com/ruby-amqp/bunny/issues/558)
+
+Reproduction steps contributed by Brian Morton and Scott Bonebraker.
+
+### I/O Exceptions from Heartbeat Sender are Now Silent
+
+Heartbeat sender's purpose is to notify the peer, not so much
+to detect local connectivity failures; those will be detected
+by the I/O loop and transport.
+
+For single threaded connection users that prefer to roll their own
+recovery strategies getting exceptions from the heartbeat sender
+was counterproductive and painful to deal with.
+
+As part of this work a new Toxiproxy-based test suite was introduced
+to Bunny.
+
+GitHub issue: [#559](https://github.com/ruby-amqp/bunny/issues/559)
+
+Contributed by Scott Bonebraker.
+
+### Correct Connection State on Connections that Experienced Missed Heartbeat
+
+Connections that experienced connection closure did not always correctly transition to the closed state.
+`Bunny::ConnectionClosedError` will now be thrown when an operation is attempted on such
+connections.
+
+GitHub issue: [#561](https://github.com/ruby-amqp/bunny/issues/561)
+
+Contributed by Scott Bonebraker.
+
+### Connection Recovery Will Fail When Max Retry Attempt Limit is Exceeded
+
+GitHub issue: [#549](https://github.com/ruby-amqp/bunny/issues/549)
+
+Contributed by Arlandis Word.
+
+### Squashed Warnings
+
+Many warnings have been eliminated.
+
+GitHub issue: [#563](https://github.com/ruby-amqp/bunny/issues/563)
+
+Contributed by @dacto.
+
+
+### API Reference Corrections
+
+GitHub issue: [#557](https://github.com/ruby-amqp/bunny/pull/557)
+
+Contributed by Bruno Costa.
+
+
+## Changes between Bunny 2.10.0 and 2.11.0 (Jun 21st, 2018)
+
+### More Reliable System-wide Trusted Certificate Directory Detection
+
+Bunny no longer tries to compile a list of trusted CA certificates on its own.
+Instead it uses an OpenSSL API method that makes OpenSSL set the path(s),
+which should cover more platforms and be forward- and backward-compatible.
+
+GitHub issue: [#555](https://github.com/ruby-amqp/bunny/issues/555).
+
+Contributed by Ana María Martínez Gómez.
+
+
+
+## Changes between Bunny 2.9.0 and 2.10.0 (Jun 5th, 2018)
+
+`2.10.0` is a maintenance release that introduces a couple of
+**minor potentially breaking changes**.
+
+
+### Disabling Heartbeats Also Disables TCP Socket Read Timeouts
+
+Disabling heartbeats will now disable TCP socket read timeouts.
+
+They go hand in hand and users who prefer TCP keepalives via
+kernel configuration previously had to also explicitly configure
+a zero read timeout.
+
+GitHub issue: [#551](https://github.com/ruby-amqp/bunny/pull/551).
+
+Contributed by Carl Hörberg.
+
+
+### `verify_peer: false` Has the Expected Effect Again
+
+Make sure `verify_peer: false` has the expected effect again.
+
+Default value of connection's `:verify_peer` option to `true` only when
+all of `:verify_ssl`, `:verify_peer`, and `:verify` are `nil`.
+
+GitHub issue: [#541](https://github.com/ruby-amqp/bunny/issues/541).
+
+Contributed by Howard Ding.
+
+
+### Maximum Number of Channels Limited to 2K by Default
+
+Default maximum number of channels is limited to 2047 to reduce the probability
+of severe channel leaks. See [rabbitmq/rabbitmq-server#1593](https://github.com/rabbitmq/rabbitmq-server/issues/1593) for details.
+
+Applications that want to use more channels per connection can still configure a higher value
+using the `channel_max` setting (for both Bunny and RabbitMQ server).
+
+GitHub issue: [#553](https://github.com/ruby-amqp/bunny/pull/553).
+
+
+
+### Squashed Some Warnings
+
+GitHub issue: [#552](https://github.com/ruby-amqp/bunny/pull/552).
+
+Contributed by @utilum.
+
+
+
+### Disabling Heartbeats Disables TCP Socket Read Timeouts
+
+Disabling heartbeats will also disable TCP socket read timeouts,
+since the two are effectively interconnected. In this case a mechanism
+such as [TCP keepalives](http://www.rabbitmq.com/heartbeats.html#tcp-keepalives) is assumed to be used.
+
+See [RabbitMQ heartbeats guide](http://www.rabbitmq.com/heartbeats.html) for a more
+detailed overview of the options.
+
+GH issue: [#519](https://github.com/ruby-amqp/bunny/issues/519).
+
+Contributed by Carl Hörberg.
+
+
+## Changes between Bunny 2.8.0 and 2.9.0 (Jan 8th, 2018)
+
+### Ruby 2.2 Requirement
+
+Bunny now requires Ruby 2.2.
+
+
+### Connection Recovery Now Retries on Timeouts
+
+Connection recovery now will retry on TCP connection timeouts.
+
+GitHub issue: [#537](https://github.com/ruby-amqp/bunny/pull/537).
+
+
+### More URI Query Parameters
+
+Bunny now supports more URI query parameters plus aliases
+that are identical to those of the server.
+
+Contributed by Andrew Babichev.
+
+GitHub issue: [#534](https://github.com/ruby-amqp/bunny/pull/534)
+
+
+
+## Changes between Bunny 2.7.0 and 2.8.0 (Dec 18th, 2018)
+
+This release has **minor breaking public API changes**.
+
+### `Bunny::Channel#close` on a Closed Channel Now Raises a Sensible Exception
+
+`Bunny::Channel#close` on an already closed channel will now raise a sensible exception.
+If the channel was closed due to a channel-level protocol exception, that exception will
+be mentioned.
+
+GitHub issue: [#528](https://github.com/ruby-amqp/bunny/issues/528), see [9df7cb](https://github.com/ruby-amqp/bunny/commit/9df7cb04d9ff12b1af62a11e239fd81e5472c872) for
+details.
+
+### JRuby 9K Compatibility
+
+A JRuby 9K compatibility issue was corrected by Marian Posăceanu.
+Note that JRuby users are recommended to use [March Hare](http://rubymarchhare.info/), a JRuby-oriented client, instead
+of Bunny.
+
+GitHub issue: [#529](https://github.com/ruby-amqp/bunny/pull/529)
+
+### Connection Exceptions are Logged as Warning with Automatic Recovery
+
+When automatic recovery is enabled, connection errors are now logged as warnings
+and not errors.
+
+Contributed by Merten Falk.
+
+GitHub issue: [#531](https://github.com/ruby-amqp/bunny/pull/531)
+
+### Server Heartbeat Value as a String
+
+It is now possible to specify a server-defined heartbeat value as a string (`"server"`), not just
+a symbol. This makes it easier to load settings from YAML files.
+
+Contributed by Tyrone Wilson.
+
+GitHub issue: [#524](https://github.com/ruby-amqp/bunny/pull/524)
+
+
+## Changes between Bunny 2.7.0 and 2.7.1 (Sep 25th, 2017)
+
+### Sensible Socket Read Timeouts When RabbitMQ is Configured to Disabled Heartbeats
+
+Bunny now correctly handles scenarios where server is configured
+to disable heartbeats (which is a terrible idea, don't do it!)
+
+GitHub issue: [#519](https://github.com/ruby-amqp/bunny/issues/519).
+
+### Bunny::Channel#basic_get Usability
+
+`Bunny::Channel#basic_get` invoked with a non-existent queue now
+throws a channel exception instead of a generic operation timeout.
+
+GitHub issue: [#518](https://github.com/ruby-amqp/bunny/issues/518).
+
+### Spec Suite Improvements
+
+`BUNNY_CERTIFICATE_DIR` environment variable now can be used
+to override local CA and client certificate/key pair directory.
+The directory is expected to be the result directory generated
+by the basic [tls-gen](http://github.com/michaelklishin/tls-gen) profile.
+
+TLSv1.0 is no longer used in tests because it's being disabled by default
+by more and more installations as it has known vulnerabilities
+and is no longer considered to be acceptable by several compliance
+standards (e.g. PCI DSS).
+
+### Improved Synchronisation for channel.close Handlers
+
+`channel.close` handler will now acquire a lock . This avoids concurrency
+hazards in some rare scenarios when a channel is closed due a protocol
+exception by the server and concurrently opened by user code
+at the same time.
+
+### More Meaningful Error Messages in Bunny::Session#create_channel
+
+Sometimes users attempt to open a channel on a connection that
+isn't connected yet because `Bunny::Session#start` was never invoked.
+
+`Bunny::Session#create_channel` will now provide a more sensible exception message
+in those cases.
+
+
+## Changes between Bunny 2.6.0 and 2.7.0 (May 11th, 2017)
+
+### amq-protocol Update
+
+Minimum `amq-protocol` version is now [`2.2.0`](https://github.com/ruby-amqp/amq-protocol/blob/master/ChangeLog.md#changes-between-210-and-220-may-11th-2017) which includes
+a change in [how timestamps are encoded](https://github.com/ruby-amqp/amq-protocol/issues/64).
+
+
+### `Bunny::ContinuationQueue#poll` Less Prone to Race Conditions
+
+`Bunny::ContinuationQueue#poll` was reworked with feedback from Joseph Wong.
+
+GitHub issue: [#462](https://github.com/ruby-amqp/bunny/issues/462)
+
+
+### Recovery Attempt Counting Strategy Changed
+
+Previous behehavior is not unreasonable but is not what many users and
+even RabbitMQ team members come to expect. Therefore it can be
+considered a bug.
+
+Previously a reconnection counter was preserved between successful
+recoveries. This made the integration test that uses server-sent
+connection.close possible.
+
+With this change, the counter is reset after successful reconnection
+but there's an option to go back to the original behavior. We also do
+a hell of a lot more logging.
+
+GitHub issue: [#408](https://github.com/ruby-amqp/bunny/issues/408)
+
+
+### Absolute Windows File Paths are No Longer treated as Inline Certs
+
+Contributed by Jared Smartt.
+
+GitHub issue: [#492](https://github.com/ruby-amqp/bunny/issues/492).
+
+
+### Opening a Channel on an Intentionally Closed Connection Immediately Raises an Exception
+
+Contributed by Alessandro Verlato.
+
+GitHub issue: [#465](https://github.com/ruby-amqp/bunny/issues/465)
+
+
+### Bunny::ConsumerWorkPool#shutdown Terminates Early When It's Safe to Do So
+
+`Bunny::ConsumerWorkPool#shutdown(true)` waited for consumer shutdown
+even if the pool wasn't active (there were no consumers on its
+channel).
+
+GitHub issue: [#438](https://github.com/ruby-amqp/bunny/issues/438).
+
+
+### Retry on new Ruby 2.1+ variations of `EAGAIN`, `EWOULDBLOCK`
+
+GitHub issue: [#456](https://github.com/ruby-amqp/bunny/issues/456)
+
+
+### Do Not Modify Host Arrays
+
+Bunny now can work with frozen host arrays.
+
+GitHub issue: [#446](https://github.com/ruby-amqp/bunny/issues/446)
+
+
+
+## Changes between Bunny 2.5.0 and 2.6.0 (October 15th, 2016)
+
+### Graceful Shutdown of Consumers
+
+Consumer work pool will now allow for a grace period before stopping
+pool threads so that delivery processing in progress can have a chance to finish.
+
+GitHub issue: [#437](https://github.com/ruby-amqp/bunny/pull/437)
+
+Contributed by Stefan Sedich.
+
+### `Bunny::Channel#wait_for_confirms` Now Throws When Used on a Closed Channel
+
+GitHub issue: [#428](https://github.com/ruby-amqp/bunny/pull/428)
+
+Contributed by Dimitar Dimitrov.
+
+### Race Condition Eliminated in `Bunny::Channel#wait_for_confirms`
+
+GitHub issue: [#424](https://github.com/ruby-amqp/bunny/issues/424)
+
+Contributed by Dimitar Dimitrov.
+
+### More Defensive Consumer Work Pool
+
+`Bunny::ConsumerWorkPool#join` and `Bunny::ConsumerWorkPool#pause`
+no longer fails with a `NoMethodError` on nil when executed
+on a work pool that doesn't have active threads (consumers).
+
+This change is largely cosmetic and won't affect the majority
+of of projects in any way.
+
+
+## Changes between Bunny 2.4.0 and 2.5.0 (July 20th, 2016)
+
+### Exchange Bindings are Now Correctly Recovered
+
+GitHub issue: [#410](https://github.com/ruby-amqp/bunny/issues/410)
+
+Contributed by Andrew Bruce.
+
+
+### `Bunny::Channel#wait_for_confirms` Awaits While There're Outstanding Unconfirmed Messages
+
+GitHub issue: [#424](https://github.com/ruby-amqp/bunny/issues/424)
+
+Contributed by Dimitar Dimitrov.
+
+
+### Queue Recovery Respects the `:no_declare` Option
+
+Queue recovery now respects the `:no_declare` option.
+
+
+### `Bunny::Channel#wait_for_confirms` Throws Early
+
+`Bunny::Channel#wait_for_confirms` now throws an exception
+early when invoked on a closed channel.
+
+GitHub issue: [#428](https://github.com/ruby-amqp/bunny/pull/428).
+
+Contributed by Dimitar Dimitrov.
+
+
+
+## Changes between Bunny 2.3.0 and 2.4.0 (June 11th, 2016)
+
+**This release includes minor breaking API changes**.
+
+### Unconfirmed Delivery Tag Set Reset on Network Recovery
+
+Channels will now reset their unconfirmed delivery tag set after
+recovery.
+
+GitHub issue: [#406](https://github.com/ruby-amqp/bunny/pull/406)
+
+Contributed by Bill Ruddock.
+
+### Support (Quoted) IPv6 Addresses in Address Lists
+
+GitHub issue: [#383](https://github.com/ruby-amqp/bunny/issues/383).
+
+Contributed by Jeremy Heiler.
+
+### Transport#read_fully Doesn't Try to Recover
+
+Since transport is replaced by a recovering connection
+anyway, and this produces confusing errors up the stack.
+
+GitHub issue: [#359](https://github.com/ruby-amqp/bunny/issues/359)
+
+Contributed by Donal McBreen.
+
+### Client-Provided Session `:properties` Merged with Defaults
+
+Client-Provided Session `:properties` will now be merged with defaults
+instead of replacing them. This makes it much more convenient to
+override a single key.
+
+### More Predictable RABBITMQ_URL Handling
+
+**This is a breaking API change**.
+
+`RABBITMQ_URL` no longer will be used if any other
+connection options are provided. This makes it possible
+to use `RABBITMQ_URL` for some connections and options
+for others in a single OS process.
+
+GitHub issue: [#403](https://github.com/ruby-amqp/bunny/pull/403)
+
+Contributed by Jimmy Petersen.
+
+
+## Changes between Bunny 2.2.0 and 2.3.0 (Feb 26th, 2016)
+
+### Thread#abort_on_exception Setting for Consumer Work Pool Threads
+
+`Bunny::Session#create_channel` now supports a 3rd argument that,
+when set to `true`, makes consumer work pool threads to have
+`Thread#abort_on_exception` set on them.
+
+GH issue: [#382](https://github.com/ruby-amqp/bunny/pull/382)
+
+Contributed by Seamus Abshere.
+
+### Explicit Transport Closure on Recovery
+
+Bunny now will explicitly close previosly used transport before starting
+connection recovery.
+
+GitHub issue: [#377](https://github.com/ruby-amqp/bunny/pull/377).
+
+Contributed by bkanhoopla.
+
+### No TLS Socket Double-init
+
+Makes sure that TLS sockets are not double-initialized.
+
+GH issue: [#345](https://github.com/ruby-amqp/bunny/issues/345).
+
+Contributed by Carl Hörberg.
+
+### Lazily Evaluated Debug Log Strings
+
+GH issue: [#375](https://github.com/ruby-amqp/bunny/pull/375)
+
+Contributed by Omer Katz.
+
+
+
+## Changes between Bunny 2.1.0 and 2.2.0 (Sep 6th, 2015)
+
+### Add :addresses to connect options
+
+Before this the connection options only allowed multiple hosts, an
+address is a combination of a host and a port. This makes it possible to
+specify different hosts with different ports.
+
+Contributed by Bart van Zon (Tele2).
+
+### Recover from connection.close by default
+
+Bunny will now try to reconnect also when server sent connection.close is
+received, e.g. when a server is restarting (but also when the connection is
+force closed by the server). This is in-line with how many other clients behave.
+The old default was `recover_from_connection_close: false`.
+
+Contributed by Carl Hörberg (CloudAMQP).
+
+
+## Changes between Bunny 2.0.0 and 2.1.0
+
+Bunny 2.1.0 has an **important breaking change**. It is highly
+advised that 2.1.0 is not mixed with earlier versions of Bunny
+in case your applications include **integers in message headers**.
+
+### Integer Value Serialisation in Headers
+
+Integer values in headers are now serialised as signed 64-bit integers. Previously
+they were serialised as 32-bit unsigned integers, causing both underflows
+and overflows: incorrect values were observed by consumers.
+
+It is highly
+advised that 2.1.0 is not mixed with earlier versions of Bunny
+in case your applications include integers in message headers.
+
+If that's not the case, Bunny 2.1 will integeroperate with any earlier version
+starting with 0.9.0 just fine. Popular clients in other languages
+(e.g. Java and .NET) will interoperate with Bunny 2.1.0 without
+issues.
+
+
+### Explicit Ruby 2.0 Requirement
+
+Bunny now requires Ruby 2.0 in the gemspec.
+
+Contributed by Carl Hörberg.
+
+### JRuby Fix
+
+Bunny runs again on JRuby. Note that
+JRuby users are strongly advised to use March Hare instead.
+
+Contributed by Teodor Pripoae.
+
+
+
+## Changes between Bunny 1.7.0 and 2.0.0
+
+Bunny `2.0` doesn't have any breaking API changes
+but drops Ruby 1.8 and 1.9 (both EOL'ed) support,
+hence the version.
+
+### Minimum Required Ruby Version is 2.0
+
+Bunny `2.0` requires Ruby 2.0 or later.
+
+## Non-Blocking Writes
+
+Bunny now uses non-blocking socket writes, uses a reduced
+number of writes for message publishing (frames are batched
+into a single write), and handles TCP back pressure from
+RabbitMQ better.
+
+Contributed by Irina Bednova and Michael Klishin.
+
+### Reduced Timeout Use
+
+`Bunny::ContinuationQueue#poll` no longer relies on Ruby's `Timeout` which has
+numerous issues, including starting a new "interruptor" thread per operation,
+which is far from efficient.
+
+Contributed by Joe Eli McIlvain and Carl Hörberg.
+
+### Capped Number of Connection Recovery Attempts
+
+`:recovery_attempts` is a new option that limits the number of
+connection recovery attempts performed by Bunny. `nil` means
+"no limit".
+
+Contributed by Irina Bednova.
+
+### Bunny::Channel#basic_ack and Related Methods Improvements
+
+`Bunny::Channel#basic_ack`, `Bunny::Channel#basic_nack`, and `Bunny::Channel#basic_reject`
+now adjust delivery tags between connection recoveries, as well as have a default value for
+the second argument.
+
+Contributed by Wayne Conrad.
+
+### Logger Output Remains Consistent
+
+Setting the `@logger.progname` attribute changes the output of the logger.
+This is not expected behaviour when the client provides a custom logger.
+Behaviour remains unchainged when the internally initialized logger is used.
+
+Contributed by Justin Carter.
+
+### prefetch_count is Limited to 65535
+
+Since `basic.qos`'s `prefetch_count` field is of type `short` in the protocol,
+Bunny must enforce its maximum allowed value to `2^16 - 1` to avoid
+confusing issues due to overflow.
+
+### Per-Consumer and Per-Channel Prefetch
+
+Recent RabbitMQ versions support `basic.qos` `global` flag, controlling whether
+`prefetch` applies per-consumer or per-channel. Bunny `Channel#prefetch` now
+allows flag to be set as optional parameter, with the same default behaviour as
+before (per-consumer).
+
+Contributed by tiredpixel.
+
+
+## Changes between Bunny 1.6.0 and 1.7.0
+
+### TLS Peer Verification Enabled by Default
+
+When using TLS, peer verification is now enabled by default.
+It is still possible to [disable verification](http://rubybunny.info/articles/tls.html), e.g. for convenient
+development locally.
+
+Peer verification is a means of protection against man-in-the-middle attacks
+and is highly recommended in production settings. However, it can be an inconvenience
+during local development. We believe it's time to have the default to be
+more secure.
+
+Contributed by Michael Klishin (Pivotal) and Andre Foeken (Nedap).
+
+
+### Higher Default Connection Timeout
+
+Default connection timeout has been increased to 25 seconds. The older
+default of 5 seconds wasn't sufficient in some edge cases with DNS
+resolution (e.g. when primary DNS server is down).
+
+The value can be overriden at connection time.
+
+Contributed by Yury Batenko.
+
+
+### Socket Read Timeout No Longer Set to 0 With Disabled Heartbeats
+
+GH issue: [#267](https://github.com/ruby-amqp/bunny/pull/267).
+
+
+### JRuby Writes Fixes
+
+On JRuby, Bunny reverts back to using plain old `write(2)` for writes. The CRuby implementation
+on JRuby suffers from I/O incompatibilities. Until JRuby
+
+Bunny users who run on JRuby are highly recommended to switch to [March Hare](http://rubymarchhare.info),
+which has nearly identical API and is significantly more efficient.
+
+
+### Bunny::Session#with_channel Synchornisation Improvements
+
+`Bunny::Session#with_channel` is now fully synchronised and won't run into `COMMAND_INVALID` errors
+when used from multiple threads that share a connection.
+
+
+
+## Changes between Bunny 1.5.0 and 1.6.0
+
+### TLSv1 by Default
+
+TLS connections now prefer TLSv1 (or later, if available) due to the recently discovered
+[POODLE attack](https://www.openssl.org/~bodo/ssl-poodle.pdf) on SSLv3.
+
+Contributed by Michael Klishin (Pivotal) and Justin Powers (Desk.com).
+
+GH issues:
+
+ * [#259](https://github.com/ruby-amqp/bunny/pull/259)
+ * [#260](https://github.com/ruby-amqp/bunny/pull/260)
+ * [#261](https://github.com/ruby-amqp/bunny/pull/261)
+
+
+### Socket Read and Write Timeout Improvements
+
+Bunny now sets a read timeout on the sockets it opens, and uses
+`IO.select` timeouts as the most reliable option available
+on Ruby 1.9 and later.
+
+GH issue: [#254](https://github.com/ruby-amqp/bunny/pull/254).
+
+Contributed by Andre Foeken (Nedap).
+
+### Inline TLS Certificates Support
+
+TLS certificate options now accept inline certificates as well as
+file paths.
+
+GH issues: [#255](https://github.com/ruby-amqp/bunny/pull/255), [#256](https://github.com/ruby-amqp/bunny/pull/256).
+
+Contributed by Will Barrett (Sqwiggle).
+
+
+## Changes between Bunny 1.4.0 and 1.5.0
+
+### Improved Uncaught Exception Handler
+
+Uncaught exception handler now provides more information about the exception,
+including its caller (one more stack trace line).
+
+Contributed by Carl Hörberg (CloudAMQP).
+
+
+### Convenience Method for Temporary (Server-named, Exclusive) Queue Declaration
+
+`Bunny::Channel#temporary_queue` is a convenience method that declares a new
+server-named exclusive queue:
+
+``` ruby
+q = ch.temporary_queue
+```
+
+Contributed by Daniel Schierbeck (Zendesk).
+
+### Recovery Reliability Improvements
+
+Automatic connection recovery robustness improvements.
+Contributed by Andre Foeken (Nedap).
+
+### Host Lists
+
+It is now possible to pass the `:hosts` option to `Bunny.new`/`Bunny::Session#initialize`.
+When connection to RabbitMQ (including during connection recovery), a random host
+will be chosen from the list.
+
+Connection shuffling and robustness improvements.
+
+Contributed by Andre Foeken (Nedap).
+
+### Default Channel Removed
+
+Breaks compatibility with Bunny 0.8.x.
+
+`Bunny:Session#default_channel` was removed. Please open channels explicitly now,
+as all the examples in the docs do.
+
+
+## Changes between Bunny 1.3.0 and 1.4.0
+
+### Channel#wait_for_confirms Returns Immediately If All Publishes Confirmed
+
+Contributed by Matt Campbell.
+
+### Publisher Confirms is In Sync After Recovery
+
+When a connection is recovered, the sequence counter resets on the
+broker, but not the client. To keep things in sync the client must store a confirmation
+offset after a recovery.
+
+Contributed by Devin Christensen.
+
+### NoMethodError on Thread During Shutdown
+
+During abnormal termination, `Bunny::Session#close` no longer tries
+to call the non-existent `terminate_with` method on its origin
+thread.
+
+
 ## Changes between Bunny 1.2.0 and 1.3.0
 
 ### TLS Can Be Explicitly Disabled
@@ -389,7 +1275,7 @@ Bunny now will use the following TLS/SSL CA's paths on Linux by default:
 and will log a warning if no CA files are available via default paths
 or `:tls_ca_certificates`.
 
-Contributed by Carl Hörberg.
+Contributed by Carl Hörberg.
 
 ### Consumers Can Be Re-Registered From Bunny::Consumer#handle_cancellation