99fc7e4e50
Moving the Start/Stop logic into Frame.
...
Also formalizes the relationship between the function's Task and
how it relates to graceful shutdown. Specifically how it relates
to finishing the requests currently in progress.
2015-09-25 12:49:26 -07:00
f89a586b75
Removing boilerplate-generated #region
2015-09-25 12:49:25 -07:00
37f7dfd01f
Typo in comments
2015-09-25 12:49:25 -07:00
c424b8437c
Renaming based on PR feedback
2015-09-25 12:49:24 -07:00
844c791b16
Processing function should return Task
2015-09-25 12:49:23 -07:00
091084cfe2
Refactoring MemoryPool class locations
...
Moving Iterator out into its own file rather than being a nested class
Moving pool classes into Infrastructure namespace instead of Http
2015-09-25 12:49:23 -07:00
13defc5a32
Initialize Memory2 in ListenerContext
...
Better than initializing from multiple other classes
2015-09-25 12:49:22 -07:00
5ba1b9fb3c
Secondary listeners need to allocate their own memory pool
2015-09-25 12:49:21 -07:00
6db3d9e645
Using specific exception types
2015-09-25 12:49:21 -07:00
76b528e1d2
Adding comments to MemoryPool classes
2015-09-25 12:49:20 -07:00
ff0affe34d
Removing _frameTask and fixing Task.Run argument
2015-09-25 12:49:19 -07:00
30ec2cb0b1
Harden framing epilog
...
* Connection no longer needes Frame's processing Task
* Any exceptions from unusual processing is logged
2015-09-25 12:49:19 -07:00
28250d9981
Providing Path to application
2015-09-25 12:49:18 -07:00
04f6446a50
Updates for PR feedback
2015-09-25 12:49:18 -07:00
d3a87c4c14
Removing MessageBodyExchanger base class
2015-09-25 12:49:17 -07:00
52dc37eae7
Fixing a header parsing bug
...
When request header data arrives with \r\n split across packets
2015-09-25 12:49:16 -07:00
e5a3bda3a2
Progress on flow control
...
* Dealing with race conditions
* Reworking iterator methods
* Shutdown and Close need to be passed throught the write-behind mechanism
2015-09-25 12:49:15 -07:00
4193c92ba7
Fixing merge conflicts
2015-09-25 12:49:15 -07:00
1f6aaebeda
Changing flow control for incoming data
...
SocketInput is now awaitable
Input buffers are slab-allocated and pinned from large object heap
Request frame parsing is offloaded from libuv thread
2015-09-25 12:49:14 -07:00
557f6d6993
Refactoring response control flow
...
- Bring through both sync and async execution paths
- Remove callback pattern from from and socketoutput write calls
2015-09-25 12:47:51 -07:00
1e39473047
uv_pipe_connect returns void
...
This fix prevents Libuv.pipe_connect from throwing when the stack
memory that was previously incorrectly interpreted as the int return
value happens to be negative.
When pipe_connect threw, an assertion failure would follow since the
pipe handle would be closed prematurely.
http://docs.libuv.org/en/v1.x/pipe.html#c.uv_pipe_connect
#205
2015-09-25 11:40:40 -07:00
c4bf10d6e9
Add third party notices for libuv
...
Fixes #218
2015-09-23 15:42:58 -07:00
355bc01a15
Hosting#358 Move the default address into Hosting.
2015-09-22 11:56:46 -07:00
891b991a34
Use new packInclude feature to include native assets
...
- Now Kestrel won't break when people overwrite the build.cmd
2015-09-19 03:59:58 -07:00
47323a88e1
Surface errors from Listener.StartAsync better
...
If CreateListenSocket fails, surface the original error instead of throwing a
NullReferenceException from Listener.Dispose().
#185
2015-09-16 18:19:01 -07:00
2d01f2752b
Catch and log uv_accept errors
...
- This is also what is done by node.js
- "tcp, pipe: don't assert on uv_accept() errors (Ben Noordhuis)"
0685707bc6
2015-09-16 18:18:58 -07:00
f14af1f409
Don't swallow too many exceptions
...
- Swallowing too many exceptions can end up hiding issues in
Kestrel itself. It's better to let the process die.
- If we want to handle certain exceptions we should be as
specific as possible with our try/catch blocks.
2015-09-16 18:18:55 -07:00
2e225b0db6
Hosting#331 Add IServerAddressesFeature.
2015-09-15 12:47:18 -07:00
78de14d248
Adding NeutralResourcesLanguageAttribute
2015-09-10 18:15:27 -07:00
958cc3eca3
Add missing arguments to LogDebug arguments in KestrelTrace
2015-09-10 12:40:16 -07:00
813127e7f6
Merge branch 'benaadams/negative-threads' into dev
2015-09-10 11:58:09 -07:00
ec88e57f67
Remove commented code in KestrelTrace
2015-09-10 11:41:27 -07:00
1ebf98d40d
Log message style changes
...
- Disabled some log messages for possibly being too verbose.
- Don't log ConnectionReadFin when there is an error.
2015-09-10 11:41:25 -07:00
0ef096b41c
Increment connection id for logging
2015-09-10 11:41:22 -07:00
dda5774a8b
Throw exception if ThreadCount negative
2015-09-10 08:03:42 +01:00
887cf2c7be
Negative thread count fix
2015-09-10 02:57:49 +01:00
dc08062545
Merge branch 'drussilla/add-logging' into dev
...
Conflicts:
src/Microsoft.AspNet.Server.Kestrel/Http/Connection.cs
src/Microsoft.AspNet.Server.Kestrel/Http/ListenerSecondary.cs
2015-09-09 11:28:09 -07:00
ac77c11211
Make ConnectionCallback static again.
2015-09-09 02:13:49 +02:00
2b2943d5b0
Restore static delegates
2015-09-09 01:27:44 +02:00
2041e4d08b
Prevent pipes from being closed prematurely on OS X and Linux
...
This change fixes the in-process marshalling of TCP handles on Mac and Linux
that is used to support having multiple threads accepting connections from
multiple loops.
On these two platforms, the ReadStart callback somtimes gets called with a
status and pipe_pending_count equal to zero. Now when the status is zero
just exit the callback without closing the pipe.
This change more closely follows the example at
https://nikhilm.github.io/uvbook/processes.html#sending-file-descriptors-over-pipes
2015-09-08 16:21:45 -07:00
a93a66fe7c
Replace Trace.WriteLine with ITraceLogger
2015-09-09 00:26:26 +02:00
6d47227975
KestrelTrace refactored and added to the ServiceContext. Close aspnet/KestrelHttpServer#141
2015-09-05 18:17:17 +02:00
5ee80e0155
Fix bug #191 introduced by b25d2d9
2015-09-05 12:29:00 +02:00
f10c989d90
Add ServiceContext to make it easier to flow new services through ctors
2015-09-04 15:04:25 -07:00
1584d70e1f
Clean up if an exception is thrown in the middle of ServerFactory.Start
2015-09-03 15:26:30 -07:00
c50aec1729
Adding comments to meaningless field names
2015-09-02 21:14:18 -07:00
dffd977c3f
Fixing field names
2015-09-02 20:50:21 -07:00
5c7007a4e9
Fixing member field name
2015-09-02 20:37:11 -07:00
0859d82d6b
Adding private keyword to private enums
2015-09-02 20:33:51 -07:00
3bbb77f9d0
Alphabetizing using statements
2015-09-02 20:31:11 -07:00
b25d2d9772
Continued updates for #184
...
* Reordering members
- Fields
- Constructors
- Properties
- EverythingElse
- NestedTypes
* Removing commented code
2015-09-02 20:08:54 -07:00
2467cf2ade
Nested types must be last
2015-09-02 00:07:58 -07:00
46604d68b3
Initial commit for style cleanup #184
...
* adding private keyword to fields
* one type per file
2015-09-01 22:25:51 -07:00
c2192d7bd1
Change IServerInformation to IFeatureCollection.
2015-09-01 15:27:57 -07:00
d0bd2b3dd0
Configure unix pipes via the hostname instead of the scheme.
2015-08-31 10:46:48 -07:00
a919ea8d0a
Remove unnecessary listener interfaces
2015-08-31 10:46:46 -07:00
a9b8cfa582
Remove unnecessary use of generics in listeners
2015-08-31 10:46:40 -07:00
803ec38073
React to string[] -> StringValues changes.
2015-08-28 12:59:00 -07:00
40ca61b640
Merge branch 'release' into dev
2015-08-26 16:32:19 -07:00
2642c84bf9
Don't automatically set Content-Length: 0 in some circumstances
...
- When in response to a HEAD Request
- For 101, 204, 205 and 304 responses
- For non keep-alive connections
2015-08-26 16:16:35 -07:00
69759231ff
Set Content-Length: 0 when an AppFunc completes without a write
...
- Previously an incomplete chunked response would be written instead.
- Add test to verify Content-Length: 0 is set automatically.
- Add test to verify HTTP/1.0 keep-alive isn't used if no Content-Length
is set for the response.
- Add tests to verify errors are handled properly after chunked writes.
#173
2015-08-25 23:17:41 -07:00
ca0a42a01e
Merge branch 'Daniel15/unix-sockets' into dev
2015-08-24 12:53:31 -07:00
7cdfdddaa0
Unpinnng Dnx.Runtime.Abstractions package version
2015-08-18 21:34:43 -07:00
9ade227abb
Implement support for UNIX sockets.
...
The common use-case for Kestrel in production will be behind a reverse proxy such as Nginx. In cases where the reverse proxy is located on the same machine as the application, connecting via a UNIX socket is more efficient than a TCP socket, as it avoids going through the network layer. Accessing 127.0.0.1 through TCP still needs to initiate a TCP connection and perform handshaking, checksumming, etc, all of which is avoided by using UNIX sockets.
- Moved TCP-specific stuff from Listener into new TcpListener class (same with ListenerPrimary and ListenerSecondary)
- Made Listener abstract
- Created new PipeListener. Note that while the use case is for UNIX sockets, this is called "Pipe" in uv, so I've called this "PipeListener" so the terminology is consistant
- Uses "unix" URL scheme to determine whether to use socket. "http://127.0.0.1:5000 " is for listening via TCP while "unix:///var/run/kestrel-test.sock" is for listening via UNIX socket
#156
2015-08-15 15:50:34 -07:00
753d64660d
Ignore zero length writes when automatically chunking responses
...
- Zero length writes would previously be interpreted as the end of response
- Optimize writing the chunked response suffix
- Add tests for automatic response chunking
2015-08-14 15:59:47 -07:00
3fb33119ee
Initial non-optimized support for automatically chunking responses
2015-08-14 15:59:39 -07:00
32c4f314b6
Lock around Connection.End to make it thread-safe
2015-08-14 12:17:06 -07:00
4f0480a4d0
Gracefully handle exceptions thrown from OnStarting callbacks
...
- If OnStarting is being called after the app func has completed, return a 500.
- If Onstarting is being called due to a call to write, throw from write.
2015-08-14 12:05:43 -07:00
6a01043e1a
Fix ListenerSecondary so it reliably accepts new connections on Windows
...
Calling uv_read_start on a named pipe with a larger than necessary buffer
would cause pieces of the next uv_ipc_frame_uv_stream struct to be read into
the uv_read_start buffer when a lot of tcp handles were passed quickly over
the pipe.
This prevented the struct from properly being queued for the next call to
uv_accept to consume. The empty queue caused the call to uv_accept in
ListenerSecondary to fail and return WSAEWOULDBLOCK leaving the connection
in a zombie state.
2015-08-14 11:57:21 -07:00
c0cc511b5b
Prevent access to closed socket in in Connection.End
2015-08-14 11:49:45 -07:00
b162202519
Properly close keep alive connections
2015-08-14 11:49:43 -07:00
43ebf710ab
Fix regression in reading config
2015-08-14 09:33:02 -07:00
c7535f127c
fix build break due to aspnet\configuration #246
2015-08-13 22:42:48 -07:00
097fb35ddf
Fix startup when you reference the assembly directly instead of Hosting
...
- I missed this change in Program.cs when doing the initial package merging
2015-08-12 15:14:01 -07:00
d97b02696f
Merge Kestrel into Microsoft.AspNet.Server.Kestrel
2015-08-11 14:56:20 -07:00
0ac3c3dad6
More ConfigureAwait(false)
2015-08-11 12:46:08 -07:00
688ad9bee7
Merge branch 'benaadams/configure-awaits' into dev
2015-08-11 12:34:53 -07:00
5dfca955b3
Move GeneratedCode project so no NuGet package gets created
2015-08-05 12:22:59 -07:00
7dd256f26e
Add a "dnu restore" prepare step before generating code
...
- This ensures the generated code project's project.lock.json is created
2015-08-05 11:54:17 -07:00
51693304c1
Update dependencies after rename to Microsoft.Dnx.Compilation.CSharp
2015-08-05 11:07:13 -07:00
b8f930bb4c
Update CoreCLR versions
2015-08-04 10:15:28 -07:00
e5144e3139
Include Server and Date in the initial response header dictionary
2015-08-03 16:20:34 -07:00
978dd39924
Update the prepare script used to generate code
...
- The syntax to run dnx with a modified appbase has recently changed
2015-08-03 16:20:32 -07:00
7446fe4cc7
Update SocketOutputTests to account for the "immediate" Write param
2015-08-03 16:20:30 -07:00
6ff894bb19
Using masked byte sequences to recognize well-known header names
2015-08-03 16:20:28 -07:00
b999b47218
String concatination showed up in profiler
2015-08-03 16:20:26 -07:00
f6dc72544c
Response headers don't need to be queued before subsequent write
2015-08-03 16:20:24 -07:00
c0728edda7
Linq .Any() was causing enumeration
2015-08-03 16:20:22 -07:00
f9cf9f1936
Improvements to appending request header
2015-08-03 16:20:20 -07:00
96b03ee212
Perf - enumerate with struct
...
Conflicts:
src/Microsoft.AspNet.Server.Kestrel/project.json
2015-08-03 16:20:18 -07:00
ad089be477
Removing "I like pie" diagnostics
2015-08-03 16:20:16 -07:00
963f086eb0
Prototypeing a fast header dictionary
...
Conflicts:
src/Microsoft.AspNet.Server.Kestrel/project.json
2015-08-03 16:20:14 -07:00
83b2c95385
React to IFeatureCollection changes.
2015-08-03 15:13:58 -07:00
07304640ed
react to DNX renames
2015-07-31 11:54:52 -07:00
4a9515d2e0
Gracefully handle uncaught exceptions in user code when possible
2015-07-30 16:01:33 -07:00
07de3cafd1
Clean build warnings
2015-07-30 12:14:13 -07:00
343d698286
React to DNX renames
2015-07-29 01:04:59 -07:00
47d7f73bdc
Fix SocketOutput so that it can now complete large writes
...
- Complete in this context means that the callback gets invoked.
- Previously, calls to write would never complete if the buffer contained
more than 64 KB (_maxBytesPreCompleted).
- This is tested by SocketOutputTests.CanWrite1MB.
2015-07-28 16:57:24 -07:00
cce9d8f09c
Make SocketOutput more testable
...
- Added a MockLibUv class
- Create a SocketOutputTests class
2015-07-28 16:57:22 -07:00
c345849707
Don't use QueueUserWorkItem to trigger write callbacks immediately
...
- In this case we are off the event loop, so we can invoke the callback directly.
- Increase _maxBytesBufferedBeforeThrottling
2015-07-28 16:57:21 -07:00
Louis DeJardin
Stephen Halter
Eilon Lipton
Chris R
David Fowler
Pranav K
Ben Adams
Ivan Derevyanko
Kai Ruhnau
Daniel Lo Nigro
Kirthi Krishnamraju
Troy Dai
anurse
Victor Hurdugaci