ChangeLog

5.3.0           2012/06/25
        * finagle-core: ClientBuilder and ServerBuilder's MonitorFilter to report exception sources
        * finagle-test: Add TaskTrackingTimer benchmark
        * finagle-thirft: SeqIdFilter: helpful exceptions
        * Dispatcher: all write exceptions are wrapped with WriteException

5.1.0		2012/06/14

	* finagle: Upgrade util to 5.2.0, ostrich to 8.1.0
	* finagle-redis: extend codec
	* finagle-memcache: first phase of serverset support
	* finagle-b3: we are using zipkin now
	* finagle-kestrel: Kill com.twitter.concurrent.Channel
	* finagle-core: JvmFiler, trace jvm events

5.0.2		2012/06/01

	* finagle-core: Refactor the new way of managing resources
	* finagle-core: Add capability to drain dispatcher on expire
	* finagle-core: ZookeeperServersetCluster notify client when underlying Cluster is ready
	* finagle-exception: Upgrade jerkson to 0.5.0 and streamyj to 0.4.1

5.0.0		2012/05/25

	* Upgrade to scala 2.9.2
	* finagle-core: FailFast, do not admit requests when marked failed
	* finagle-redis: now return tuples and doubles
	* finagle-memcached: record client recv annotation after hit/miss
	* finagle-core: expand exception chain in StatsFilter

4.0.0		2012/05/08

	* http: add AddResponseHeadersFilter, CorsFilter
	* finagle-redis: Add ZCard, ZRem, ZRevRange, ZRevRangeByScore commands
	* finagle-memcached: Added b3 annotations for key hits/misses
	* finagle-memcached: getResult/getsResult for java
	* finagle-http: fix client codec
	* HeapBalancer: close services that are removed from cluster
	* finagle-redis: add tracing and stats support
	* finagle-http: Add the client address for http traces. Also record uri after service name has been set
	* finagle-memcached: more efficient implementation of GetResult.merged
	* finagle-redis: return pairs for hgetall, zrangebyscores, zrevrangebyscores cmds
	* finagle-core: introduce Transports and Dispatchers
	* tracing: annotate client connection when tracing
	* update to netty 3.4.1.Final
	* prevent tenuring of response objects for responses from previously idle connections
	* finagle-redis: add quit command
	* finagle-http: add service multiplexer
	* friendlier java interface for retries

3.0.0		2012/03/14

	* ServiceToChannelHandler: keep track of shutdowns with
	pending replies, don't monitor errors that occur after
	cancellation.
	* Add two received_bytesm sent_bytes to per-channel stats.
	* http: Add wwwAuthenticate(=) and isXmlHttpRequest
	* http: CancelledRequestExceptions become 499s
	* clients: experimental Fail-Fast mode
	(ClientBuilder.expFailFast(true))
	* b3: Only log that a span was unfinished if we actually found
	one
	* thrift: better support for one-way calls
	* server: option for cancellation on hangup (default=true)
	* codecs: improved support for multiplexed protocols
	* upgrade to netty 3.4.0.alpha1 which fixes connect timer
	resolution issues. Trustin says this should be as stable as
	3.3.1.Final
	* serversets: allow for additional endpoints
	* tests: use true ephemeral ports instead of RandomSocket
	* memcache stress: use new command line flag library from
	commons
	* thrift: concatenate arrays with System.arraycopy
	* memached: fix bug wherein we made requests with empty keys
	on retries

2.0.1		2012/03/01

	* commons-stats: memoize counters
	* cluster: don't reuse adds; keep many values per key
	* clientbuilder: buffer handlers to avoid deadlock

2.0.0		2012/02/27

	* unify client and server factories
	* memcache stress test: one connection per process
	* b3: make tests deterministic
	* upgrade to newest util (and to Guava r11)

1.11.1	2012/02/15

	* HeapBalancer: safely remove services from cluster when
	they have requests in flight
	* http: speed up logging
	* core: keep track of handler walltimes

1.11.0	2012/02/13

	* finagle-redis: publish.
	* http: Message.mediaType=: no longer sets charset to utf-8
	* http: Message.setContentType: defaults charset to utf-8
	* http: added Message.charset, Message.charset=
	* Refactor resource management inside of the builders, making lifetimes
	explicit
	* Generic cluster API, support in the HeapBalancer for dynamic hosts list;
	remove other load balancers
	* redis: memcache.Client-style interface
	* netty: update to 3.3.1.Final, and add NPN TLS extension support
	* b3: flush span if timeout occurs
	* b3: java friendly tracer API
	* stress/example: "topo", a minimal appserver-like topology for testing
	* memcache: tunable example for benchmarking, testing
	* ServerBuilder: major refactor & simplification
	* SummarizingStatsReceiver: fix bug where sort didn't operate over a
	stable collection

1.10.0	2012/01/24

	* http: Message.withX methods now have parameterized return types
	* memcached: (Ketama) host ejection
	* Noisier, more robust monitoring.
	* zk cluster: avoid potential infinite loop
	* http streaming: better detection, handling of dead channels
	* http: always encode output as UTF-8
	* stream: use offer/broker for duplex stream
	* redis: imported finagle redis client from Tumblr. not yet published.

1.9.12	2012/01/05

	* SSLEngine: do not reuse between connections
	* ServerBuilder.Server: localAddress to get ephemeral port values
	* More detailed client stats
	* kestrel: parse parseStatsLines
	* FailFastFactory: new fast-fail strategy with maxOutstandingConnections
	* exceptions: add peer information to connection failures, add service name to all exceptions
	* fixes: space leak in stats, ReliableDuplexFilter lazy load ordering
	* b3: allow the user to specify sample rate
	* exception reporter: share scribe connection across instances
	* finagle-ostrich: remove, it's obsoleted by finagle-ostrich4

1.9.10	2011/12/02

	* update to util 1.12.7

1.9.9		2011/12/01

	* update to util 1.12.6

1.9.8		2011/11/30

	* upgrade to netty 3.2.7.Final
	* streaming: support HTTP/1.0
	* native: fix content-length header

1.9.7		2011/11/29


	* http: Propagate CancelledRequestException rather than
	continuing
	* thrift: deprecated LookupContext, using
	finagle.thrift.ClientId
	* use monitors where applicable (composable widgets for
	handling exceptions)
	* memcache: PHPMemCacheClient
	* make RetryingFilter usable from Java
	* thrift: convert uncaught application exceptions into
	TApplicationExceptions on the wire
	* NoStacktrace.getStackTrace() returns non-empty stacktrace
	array
	* kill off openConnectionsHealthThresholds (unused)

1.9.6		2011/11/08

	* SSL: make native provider bootstrapping code work; refactor
	'Ssl' singleton
	* tracing: Update configurations and libraries to use tracer
	factory introduced to make it possible to release resources
	properly in tracers
	* finagle-thrift: connection options + client id
	* a more flexible retry filter

1.9.5		2011/11/04

	* ExpiringService: add stats for idle and lifetime expiration.
	Deactivate() regardless of latch counter
	* Filters: don't apply refcounts on composition, do it only
	(and always) in the builder
	* Count pending timeouts, and export them
	* ChannelService: release services when `prepareChannel`
	fails.  This fixes a connection leak
	* WatermarkPool: flush waiters on creation failure.  This
	fixes a condition wherein we'd keep waiters unnecessarily.
	* Timeout exceptions: include service name
	* Suppress excessive logging of the exceptions that mostly
	caused by unreliable client connections.
	* memcached: Add generic type support (via Client.adapt), and
	Array[Byte] and String instances
	* finagle-stream: do not admit concurrent requests
	* Tracing: Add release method to Tracer trait, add
	tracerFactory to builders, deprecate tracer on builders

1.9.4		2011/10/24

	* Service: response Future will now be cancelled if the client
          disconnects
	* use static exceptions for TimedOutRequestException
	* kestrel: Cancel request Future on close in kestrel ReadHandle
	* exception cleanup: wrap all Channel exceptions with their
	underlying exception.  This makes UnknownChannelExceptions in
	particular easier to debug
	* Friendlier exceptions: In order to give very specific error
	messages to timeouts in finagle (so beginners know which
	settings in the client and server builder caused them), I'm
	adding lots of human-readable messages to the various
	TimeoutExceptions.
	* some timer bug fixes, particularly: TimerToNettyTimer:
	actually cancel underlying task.

1.9.2		2011/10/14

	* tracing: Change SpanId to use RichU64String
	* tracing: Trace.{enable,disable}
	* http: http.RequestBuilder
	* http: add common media types

1.9.1		2011/09/28

	* memcached: new interface for partial results (ie. retrieval
	failures + successes)
	* tracing: set service name in the endpoint instead of the
	span
	* more idiomatic KetamaClientBuilder constructor
	* tracing: convenience methods to add binary annotations
	* finagle-http: misc. fixes
	* HeapBalancer - a new lg n load balancer (asymptotically
	more efficient!)
	* upgrade to ostrich 4.9.0

1.9.0		2011/08/29

	* http: new http library

1.8.4		2011/08/23

	* tracing: fixed space leak
	* ssl: integrate openssl sslengine
	* thrift: update to sbt-thrift 2.0.1
	* stream: always release server when we close
	* serversets: upgraded to guava-r09
	* core: expose stats on failed retries

1.8.3		2011/08/12

	* thrift: add a server-side buffered codec; fix tracing issue
	in buffered codec
	* http: workaround hole in netty logic to guarantee
	maxRequestSize
	* serversets: don't block on ClientBuilder construction
	* ostrich4: update to ostrich 4.8.2
	* streaming: reimplement HTTP chunked streaming in terms of
	offer/broker
	* kestrel: buffered & merging of ReadHandles

1.8.1		2011/08/05

	* upgrade to netty 3.2.5.Final
	* kestrel: fix visibility of ResultWithCAS
	* kestrel: Client.{read,write}, Client.readReliably,
	MultiReader: suite of kestrel streaming readers (including
	grabby-hands replacements) using Offer/Broker
	* kestrel: don't render timeouts > 2^31-1
	* tracing: fix sampling bug
	* tracing: trace the kestrel client
	* quench unecessary error reporting

1.8.0		2011/08/02

	* SSL: perform hostname validation in the client.  this is the
	reason for the minor bump: we change the .tls() ClientBuilder
	signature
	* introduce util-codec dependency for use of the apache
	commons base 64 codec

1.7.5		2011/07/29

	* make it easier to use the NullStatsReceiver from java
	* convert WriteException to a case class (github issue 25)
	* add _root_. to java.util.Map import. reorder imports
	* Upgrade finagle-common-stats to twitter-common-stats 0.0.16
	* encode more HTTP codec errors
	* adding cardinality to ServiceException
	* finagle CachingPool: limit size to highWatermark - lowWatermark
	* exception reporting: gzipping trace strings
	* set ostrich version of finagle-ostrich4 to 4.7.3
	* ChannelServiceFactory: encode exceptions from bootstrap.connect()

1.7.4		2011/07/22

	* http: convert codec exceptions into 4xx errors; report them
	as such upstream
	* memcache: `gets' and `cas' support.

1.7.3		2011/07/19

	* fix: 1.7.2 introduced a regression in the pool that may,
	under certain circumstances, admit connections beyond the high
	watermark.
	* quench exception for setting multiple Promise values when
	racing with write errors.
	* Adds tracing support for the Memcached client.  This only
	creates the client side annotations, no support for passing
	along ids and such to the memcached server.


1.7.2		2011/07/18

	* support end-to-end cancellation in finagle clients; fixes
	possible pool starvation conditions, and allows clients to
	cancel requests and reclaim resources
	* end-to-end timeouts: introduce, .connectTimeout and
	.timeout.  .connectTimeout is the end-to-end connection
	timeout (includes acquisition & TCP time), .timeout is an
	end-to-end request time: no requests will take longer
	* some doc fixes
	* misc. bug fixes