Commits on May 27, 2018
  1. Merge pull request #607 from FortnoxAB/patch-2

    jamesgorman2 committed May 27, 2018
    PooledConnection.close must be subscribed to to be executed
Commits on May 23, 2018
  1. Fixed imports

    JonasHallFnx committed May 23, 2018
Commits on May 6, 2018
  1. Merge pull request #604 from TurboProgramming/patch-1

    jamesgorman2 committed May 6, 2018
    Replace compile with implementation in gradle getting started. It will be deprecated end 2018.
  2. Merge pull request #603 from davidmoten/patch-1

    jamesgorman2 committed May 6, 2018
    Fix headers in markdown that now require an intervening space
Commits on Apr 24, 2018
  1. Implementation Replacement

    TurboProgramming committed Apr 24, 2018
    Replaced compile with implementation as it will be deprecated end 2018.
Commits on Mar 9, 2018
  1. Fix headers in markdown that now require an intervening space

    davidmoten committed Mar 9, 2018
    github markdown changed a few months ago and now require an intervening space after ### headers
Commits on Feb 13, 2018
  1. Update README status info

    jamesgorman2 committed Feb 13, 2018
    * Add project status to clarify status of 1.0/0.5/0.4
    * Mark 0.5.x as the stable release branch
Commits on Nov 9, 2017
  1. Merge pull request #595 from forresthopkinsa/patch-1

    jamesgorman2 committed Nov 9, 2017
    Fix links in getting started
Commits on Aug 8, 2017
  1. Fix links

    forresthopkinsa committed Aug 8, 2017
Commits on Feb 7, 2017
  1. Refactor spectator metrics

    NiteshKant committed Feb 6, 2017
    Currently the listener does not expose metrics to specific events but instead does internal calculations and expose custom gauges. This is limiting as it gives less information.
    Modified the metrics to represent the actual events.
    More flexible insights.
Commits on Nov 17, 2016
  1. `RequestN` reduces with completed subscribers.

    NiteshKant committed Nov 17, 2016
    #### Problem
    Re-calculation of max `requestN` per subscriber did not take into account removed subscribers correctly. This causes the `requestN` value to decrease over time when subscribers are completed.
    #### Modification
    - Modified `recalculateMaxPerSubscriber` to take old and new subscriber count instead of trying to determine the correct value from the current subscriber queue size.
    - `subscribers.remove()` does not happen on unsubscribe but only from the task that recalculates the max `requestN` values. This also coalesces multiple removes into a single task run.
    #### Result
    Consistent `requestN` values with new subscribes and completions.
Commits on Nov 4, 2016
  1. Upgrade RxJava to 1.2.+

    jamesgorman2 authored and NiteshKant committed Nov 4, 2016
Commits on Sep 12, 2016
Commits on Aug 31, 2016
Commits on Aug 23, 2016
  1. Fixes Issue #530 (#543)

    NiteshKant committed Aug 23, 2016
    `ServerSentEventDecoder` was not ignoring whitespaces before the event name, this caused the event name to be invalid (since the parser only checks for the first character).
    Now, ignoring all whitespaces before the event name.
Commits on Jul 5, 2016
  1. Fixed client connection level events (#527)

    NiteshKant committed Jul 5, 2016
    #### Problem
    Since the pipelines for client are setup after the connection is established, callbacks like `connect` are not send to the handlers. This was the reason for missing the connection related events for clients.
    `PooledConnection` acquire events were not published because the events were getting published before the event listeners were setup.
    #### Modification
    This change moves the work done for client on `ChannelHandler.connet()` to `ChannelActivityBufferingHandler` which is added on channel creation.
    Now, publishing events for pool acquire are published using the event publisher configured on the channel.
Commits on Jul 2, 2016
Commits on Jul 1, 2016
  1. Collect sequences of ByteBufs into a single ByteBuf (#524)

    jamesgorman2 authored and NiteshKant committed Jul 1, 2016
    * Collect sequences of ByteBufs into a single ByteBuf
    Two methods are provided - collect all and collect up to max bytes.
  2. Empty body for zero length payloads (#522)

    jamesgorman2 authored and NiteshKant committed Jul 1, 2016
    * Empty body for zero length payloads (#470, #520)
    Update `RawRequest` and `HttpServerResponseImpl` so they return `content-length: 0` and an empty body
    when no writes are made to `HttpClientRequest` and `HttpServerResponse` respectively.
    This also adds some regression level tests against the bytes sent and received from the server.
Commits on Jun 23, 2016
  1. Disable `LocalEchoTest` (#519)

    NiteshKant committed Jun 23, 2016
    Travis does not seem to be happy about this test, although it passed for the PR.
    Disabling this for now as this is not really critical functionality.
Commits on Jun 22, 2016
  1. Fixes #501 (#518)

    NiteshKant committed Jun 22, 2016
    Removing JMH dependency as no JMH tests are present today.
  2. Delayed write streams race condition. (#517)

    NiteshKant committed Jun 22, 2016
    `BackpressureManagingHandler` was not correctly enqueuing the writes to the eventloop when done from outside the eventloop.
    Fixed the problem and also added test.
Commits on Jun 19, 2016
  1. Local transport.

    NiteshKant committed Jun 19, 2016
    Added example for using netty's local (in jvm) transport.
    This required a modification in `TcpClient` to add a capability of providing `EventLoopGroup` and `Channel` class.
  2. Named wire loggers. (#516)

    NiteshKant committed Jun 19, 2016
    Wire Logging available for all clients and servers is used to debug network traffic.
    In an application that contains multiple clients (or servers), it will be useful if these logs can be enabled per client/server instead of the current model of enabling logs for all the clients.
    This change deprecates the existing `enableWireLogging` methods and add a replacement to also include the name of the logger.
Commits on Jun 15, 2016
  1. Better write backpressure (#515)

    NiteshKant committed Jun 15, 2016
    Today any writes to the channel has a `request(1)` semantic that the subscriber always ever just request 1 item from the source.
    This is sub-optimal if the source that is writing isn't in-memory. For out-of-process sources, it is much better to ask for more than 1 item to get batching behavior.
    This change introduces the following scheme:
    - Have a max buffer size (64 now) per channel.
    - This demand is equally split among all active subscribers (for more than one `write()` on the channel)
    - All subscribers request up to the maximum limit allowed for them (64/n).
    - The subscribers then request more when the pending items go below half the max limit (64/2*n)
Commits on Jun 9, 2016
  1. Upgrade to netty 4.1.0-Final (#513)

    NiteshKant committed Jun 9, 2016
    * Upgrade to netty 4.1.0-Final
Commits on Jun 1, 2016
  1. README update

    NiteshKant committed Jun 1, 2016
    Removed the binary references to snapshot and added the latest-release badge
  2. Merge pull request #510 from NiteshKant/0.5.x

    NiteshKant committed Jun 1, 2016
    Fixes #509 and other minor fix
Commits on May 31, 2016
  1. Bug fixes

    NiteshKant committed May 31, 2016
    - Issue 509: `HttpServerToConnectionBridge` was calling `ctx.fireUserEvent..` method instead of `super.userEv..` which will make the second subscriber unregistered. Now calling `super.userEven..`. Also added a test for this.
    - `AbstractHttpConnectionBridge` was setting the status as complete before invoking the subscribers. This will cause issues for concatenated subscribers as the second subscriber to the content will see that the content is already completed.
Commits on May 23, 2016
  1. Fix project release (#508)

    rdegnan authored and NiteshKant committed May 23, 2016
  2. Update gradle wrapper

    rdegnan committed May 23, 2016