All notable changes to this project will be documented in this file. See standard-version for commit guidelines.
2.1.0 (2018-02-18)
- (#104) improve poison message handling (9d5ccc3)
- (#109) randomize initial connection index based on pid or hostname (2d6f372)
- (#74) correct errors in direct reply queue implementation (cccdf09)
- (#78) make it possible to handle messages from a unique queue using the alias/friendly name (50045a8)
- correct issue where publishes and requests before connection or configuration had completed would get lost (1f58c0b)
- remove remaining lodash references in core that got missed (9aa3b04)
- (#87) add support for scatter-gather pattern via request (db23b48)
- (#93) add support for bulkPublish operation (d2df5ea)
- add support for purging queues (19a678d)
2.1.0 (2018-02-18)
- (#104) improve poison message handling (9d5ccc3)
- (#109) randomize initial connection index based on pid or hostname (2d6f372)
- (#74) correct errors in direct reply queue implementation (cccdf09)
- (#78) make it possible to handle messages from a unique queue using the alias/friendly name (50045a8)
- correct issue where publishes and requests before connection or configuration had completed would get lost (1f58c0b)
- (#87) add support for scatter-gather pattern via request (db23b48)
- (#93) add support for bulkPublish operation (d2df5ea)
- add support for purging queues (19a678d)
- Allow publisher confirms to be turned off per exchange
- Moving towards ES6 features
- Clean up cases where arguments were being re-written (perf issue)
- Reduce cases where properties are introduced to hashes after instantiation (perf issue)
- Rewrite tests into single modules with individual config to make contributing & maintenance easier
- Move build system to Travis
- Use coveralls to track code coverage
- Remove lodash dependency
- Remove when dependency
- Change logging to bole
- Change commit style to conventional commits
- Adopt semistandard as format style
- Improve README badges
- Break docs into multiple docs
- Add code of conduct
- Add contributor guidelines
- #80 - fixes issue where multiple messages were nack'd in nobatch mode
- #42 - fixes issue regarding broken build status
- Fixes for: #120, #107, #97, #42
- feature (#72) - support sending persistent messages to default exchanges (thanks @bmatson!)
- feature (#66) - support passing SSL certs as string (thanks @mrfelton!) making progress callback an optional argument
- #71 - leave error handler attached to connection object to try and prevent errors from being thrown from the AMQP library
- #69 - fixes an issue where the
*.connection.configured
event was not correctly being raised - #64 - added mention of
deadLetterRoutingKey
option to queue declaration - #57 - fixes (with breaking change) by removing dependency on deprecated when feature and
- #52 - serialize arrays as JSON by default
- Change connection timeout to default to 2 seconds
- #49 - bug fix for exchange recreation when reconnecting
- #33, #54 - bug fixes around graceful shutdown
- #50 - deleting queue after starting subscription causes channel error
- bug fix - ensure correct SNI is used with SSL connections when using multiple endpoints
- improvement - switch UUID lib to
uuid
sincenode-uuid
is deprecated
- #38 - Correct race conditions in queueFsm and exchangeFsm causing errors during reconnection
- #37 - Add ability to capture Rabbit generated queue names
- #36 - Make options optional for addExchange and addQueue
- #19 - Add support for publishing arrays directly (as buffers)
- Added Drone build
- Update dependencies to latest
- Defect - queues and exchanges with the same name shared channels causing serious problems
- Improvement - add support for default exchange
- Improvement - changed from jshint to eslint
- Improvement - remove Vagrant in favor of just using Dockerfile, updated instructions
- Bug Fix - Publish no longer throws errors if made before calling addConnection or configure or if exchange does not exist
- Bug Fix - #26 - fixed a bug preventing
bindQueue
from working on unique queues - Bug Fix - #22 - addConnection did not return a promise (thanks @mkozjak)
- Enhancement - publishing a number as a body gets converted to a string (thanks @brandonpsmith)
- Enhancement - routing keys can be changed on binding (thanks @droidenator)
- Bug fix - corrected bug causing connection.unreachable event not to be prefixed with connection name correctly (thanks @Cyri-L)
-
Bug fix - uri parsing should not including leading slash between host specification and vhost
-
Add section to README about logging
-
Feature - add default and custom strategies for returned mandatory messages
- 103 - gulp test no longer throws expect.js errors
- 111 - support exclusive subscriptions when calling
startSubscription
- 112 - add
shutdown
method to allow node to exit - 116 - support multiple serializers
- 119, 115, 107 - resubscribe to all previously subscribed queues after reconnect
- 121 -
rejectUnhandled
should now correctly reject unhandled messages - 122 - guarantee support for type and routing keys set to
''
- Feature -
unique
property on queue creation allows for "single use" exclusive queues with TTL- "hash" adds a numeric value
- "id" uses client id for clearer ownership
- Feature - add machine and process info to connection properties for easier identification in management console
- Feature - consumer tags are now based on "client id"
- Feature - handlers can now be scoped/limited by queue name
- Feature - allow custom limits for deferred published messages
- Feature - add support for AMQP URIs when defining connections
- Feature - add ability to stop a running subscription to the queue
- Improvement - use "client id" to create response queue names
- Improvement - more warning level logging around channel and connection disruption
- Improvement - ensure UTC timestamps are put on published messages
- Improvemet - reject
reply
promise with error if noreplyTo
address is provided - Improvement - set appId to "client id" on publish
- Improvement - introduce limit to the number of messages an exchange will cache waiting for a connection
- Improvement - handle blocked and unblocked broker events internally
- Improvement - limit the number of messages stored per exchange while waiting on a connection
- Bug fix - do not resolve close on connection until all publishes have been confirmed
- Bug fix - use reject instead of nack when queues are in noBatch
- Bug fix - noop nacks and rejections when consuming in no-ack mode
- Bug fix - channels closed by the broker should still be re-acquired with redefined primitives
- Bug fix - check for existence of headers before checking for direct-reply-to header. Thanks Matt Young (@mashu-daishi)
- Breaking - no longer recover automatically from a user terminated connection
- Breaking - limit number/duration of connection retries
- Breaking - messages are not auto-re-published when a connection or channel fails
- reply signature has changed to support control over serialziation
- response queues are named very differently, shouldn't break code, but worth noting
wascally.iomonad
logging namespace changed towascally.io
- timeouts added to requests
- 107 - Correct improper connection timeout configuration and update README
- Update dependencies to latest versions
- Use latest whistlepunk log api
- 91 - Provide publish timeout
- 98 - No-op duplicate exchange, queue and binding declarations
- 96 - Fall back to routing key if no type is provided for routing
- 94 - Update amqplib dependnecy to 0.4.0 to support Node 4
- 77 - Add explanation of connection events
- 95 - Add test coverage for wildcards in handle
- Change log namespaces to use '.' delimiter instead of ':'
- Add example for VirtualBox in Vagrantfile.sample - thanks, @josephfrazier
- Correct configuration and README regarding ports, credentials, and clustering - thanks again, @josephfrazier
- Bug fix - Resolve issue #64 working with node-config no longer throws exceptions by avoiding
get
call - Minor improvement to README to better explain publish promises
- Bug fix - correct memory leak in req/res by leveraging new postal feature
- Bug fix - calling destroy on a queue or exchange should defer until they are in ready state.
- Update machine and monologue versions
- Remove when.promise and event handles in favor of deferred promises and transitions
- 65 - Bug Fix: setting replyQueue to false caused publish to fail silently.
- 63 - Add
uri
property to connection object emitted for 'connected' event. - 61 - Bug Fix: correct SSL URIs - thanks, @longplay
- Improvements to connection clean up (specifically around resolving outstanding messages on queues)
- ConnectionFSM - Only emit 'connected' when establishing a new connection, use 'already-connected' otherwise.
Thanks to @dvideby0 and @neverfox for identifying and providing code to help reproduce bugs 39 and 57.
- 39, 57 - Bug fix: acking responses did not resolve them causing them to pile up in a response queue
- Bug fix: setting replyQueue to false throws exceptions
- Bug fix: replying from a service with replyQueue set to false fails
- Spec update - request spec in integration tests failed to ack messages causing closeAll to hang
- 47 - Added support for
noBatch
to queues, thanks @derickbailey! - 45 - Bug fix: port option was being ignored, thanks @esatterwhite
Bug fix - bindExchanges
flat out broken. Thanks to @leobispo for the catch & fix.
Special thanks to @neverfox for finding and reportig 38 & 39 - both serious problems and very difficult to find/reproduce in any kind of automated test.
- 38, 39 - I/O getting blocked when publishing at high frequencies (think for/while loops).
- Removed one-time
failed
event handler from publish call - Cache reject callbacks from publish
- On publish confirmation, remove reject from deferred array
- On exchange connection failure, invoke all rejects in deferred array
- Removed one-time
- 37 - document use of close and closeAll calls
- Correct improper use of .then( null, ... ) which was creating additional promises.
- Update whistlepunk version
- Include biggulp to simplify the gulpfile (yay?)
- Add logging support via whistlepunk
- Add logging statements to assist with troubleshooting/debugging
- 24 - Connection should not close until after all queues have completed batch processing (only applies to user initiated connection shutdown)
- 30 - Escape passwords to be connection URI safe
- 17, 19 - Unhandled messages
- Nack unhandled messages by default
- Provide configurable strategies for handling unhandled messages
- 26 - Support custom reply queue definitions per connection
- Add behavioral specs to improve coverage and testing story
- Fix bug in
reject
batching implementation - Refactor of exchange and queue implementation into channel behavior and FSM
- Reject exchange and queue creation promises on failure
- Reject publish and subscribe calls on failed exchanges and queues
- Bug fix - closing a connection didn't reliably clean up channels, exchanges and queues
- Bug fix - a failed connection that had been closed would continue to attempt reconnecting in the background
- Bug fix - configure doesn't reject the promise if a connection cannot be established
- Add connection timeout
- Add @derickbailey to contributor list
- Add demos and documentation to better explain handlers
- Allow replies to provide a
replyType
without specifyingmore
parameter - Add support for per-message expiration
- Add support for reject (nack without re-queue)
- Code clean-up / addressing linting errors
- Fix README issues
- Correct typo in spec
- Code clean-up / addressing linting errors