Date | Commit message (Collapse) |
|
Unicorn 0.97.0 has a bunch of internal cleanups and small fixes
and this is mainly to resync with those changes.
keepalive_timeout now defaults to 5 seconds (from 2 seconds
previous). This should help out clients on slower connections.
Some small fixes and cleanups:
* Rainbows::Fiber::IO objects may leak if a rare app uses them
explicitly with FiberSpawn/FiberPool-only (not RevFiberSpawn)
* quiet down ENOTCONN handling, there's nothing we can do about
this error so we won't fill our logs with it.
|
|
The previous 2 second default was not enough for folks on slow
connections where our OS socket buffers would've masked the time
it took to write out larger responses.
ref: <20100219220904.GA11377@dcvr.yhbt.net>
|
|
Just create an empty string instead and let Unicorn::HttpParser
allocate it internally to whatever size is needed.
|
|
No point in having extra code around for platforms we
don't care about.
|
|
instance_variable_{set,get} are faster, but equally ugly
|
|
|
|
Since we deal with untrusted/non-local clients, those clients
may disconnect at inopportune times and leave us with ENOTCONN
when we try to call getpeername(2)
|
|
For the very rare apps out there using Rainbows::Fiber::IO, the
FiberSpawn and FiberPool (but not RevFiberSpawn) models could
leak memory if the app-created Rainbows::Fiber::IO objects
were dereferenced without being removed from the RD/WR hashes.
|
|
commit a5f4d11cdb9465b1ffa2892b3d84ee53b8962930 in unicorn.git
switched all ivars to struct members for ease-of-hacking and
object size.
|
|
When available (Ruby 1.9), we can use Hash#compare_by_identity
to improve performance.
|
|
* rack-1.1:
http_response: disallow blank, multi-value headers
|
|
Rev 0.3.2 makes performance with Threads* under Ruby 1.8
tolerable.
|
|
The HeaderHash optimizations in Rack 1.1 interact badly with
Rails 2.3.5 (and possibly other frameworks/apps) which set
multi-value "Set-Cookie" headers without relying on the proper
methods provided by Rack::Utils.
While this is an issue with Rails not using properly, there
may be similar apps that make this mistake and Rack::Lint
does not guard against it.
Rack-ML-Ref: <20100105235845.GB3377@dcvr.yhbt.net>
|
|
This release contains minor bugfixes/compatibility improvements
for ThreadSpawn, ThreadPool and EventMachine users.
Excessive error messages from spurious wakeups using
ThreadSpawn/ThreadPool under most platforms are silenced. Only
Ruby 1.9 users under Linux were unaffected by this bug.
EventMachine users may now use EM::Deferrable objects in
responses, vastly improving compatibility with existing
async_sinatra apps.
|
|
Some async apps rely on more than just "async.callback" and
make full use of Deferrables provided by the EM::Deferrable
module. Thanks to James Tucker for bringing this to our
attention.
|
|
Under all MRI 1.8, a blocking Socket#accept Ruby method (needs
to[1]) translate to a non-blocking accept(2) system call that may
wake up threads/processes unnecessarily. Unfortunately, we
failed to trap and ignore EAGAIN in those cases.
This issue did not affect Ruby 1.9 running under modern Linux
kernels where a _blocking_ accept(2) system call is not (easily,
at least) susceptible to spurious wakeups. Non-Linux systems
running Ruby 1.9 may be affected.
[1] - using a blocking accept(2) on a shared socket with
green threads is dangerous, as noted in
commit ee7fe220ccbc991e1e7cbe982caf48e3303274c7
(and commit 451ca6997b4f298b436605b7f0af75f369320425)
|
|
This release should fix ThreadSpawn green thread blocking issues
under MRI 1.8. Excessive socket closing is avoided when using
Thread* models with Sunshowers (or clients disconnecting
during uploads).
There is a new RevFiberSpawn concurrency model which combines
Rev with the traditional FiberSpawn model.
|
|
No point in becoming the straw that causes a rehash
since hardly anybody uses it.
|
|
|
|
We'll export this across the board to all Rack applications
to sleep with. This provides the optimum method of sleeping
regardless of the concurrency model you choose. This method
is still highly not recommended for pure event-driven models
like Rev or EventMachine (but the threaded/fiber/actor-based
variants are fine).
|
|
|
|
Thanks to Ben Sandofsky for the extra set of eyes
|
|
This is like the traditional FiberSpawn, but more scalable (but
not necessarily faster) as it can use epoll or kqueue.
|
|
Under MRI 1.8, listen sockets do not appear to have the
nonblocking I/O flag on by default, nor does it set the
nonblocking I/O flag when calling #accept (but it does
when using #accept_nonblock, of course).
Normally this is not a problem even when using green threads
since MRI will internally select(2) on the file descriptor
before attempting a blocking (and immediately successful)
accept(2).
However, when sharing a listen descriptor across multiple
processes, spurious wakeups are likely to occur, causing
multiple processes may be woken up when a single client
connects.
This causes a problem because accept(2)-ing on multiple
threads/processes for a single connection causes blocking accepts in
multiple processes, leading to stalled green threads.
This is not an issue under 1.9 where a blocking accept() call
unlocks the GVL to let other threads run.
|
|
TeeInput may explicitly close on client disconnects to
avoid error messages being written to the socket, likewise
with "hack.io" users.
|
|
|
|
|
|
|
|
This makes them easier to override in subclasses.
|
|
It gets in the way of Rev/EM-based models that won't use EvCore.
It doesn't actually do anything useful except making an extra
layer of indirection to follow.
|
|
Make RACK_DEFAULTS == Unicorn::HttpRequest::DEFAULTS
and LOCALHOST == Unicorn::HttpRequest::LOCALHOST
No point in having a duplicate objects, and it also makes it
easier to share runtime constant modifications between servers.
|
|
This release introduces compatibility with Sunshowers, a library
for Web Sockets, see http://rainbows.rubyforge.org/sunshowers
for more information. Several small cleanups and fixes.
Eric Wong (20):
add RevThreadPool to README
rev: do not initialize a Rev::Loop in master process
rainbows.1: update headers
do not log IOError raised during app processing
move "async.callback" constant to EvCore
larger thread pool default sizes ({Rev,}ThreadPool)
ev_core: no need to explicitly close TmpIOs
EventMachine: allow usage as a base class
NeverBlock: resync with recent our EM-related expansion
RevThread*: move warning message to a saner place
EventMachineDefer: preliminary (and) broken version
TODO: add EM Deferrables
RevThread*: remove needless nil assignment
README: HTML5 Web Sockets may not be supported, yet...
env["hack.io"] for Fiber*, Revactor, Thread* models
EventMachineDefer is experimental
README: add Sunshowers reference
Rakefile: resync with Unicorn
doc/comparison: add Web Sockets to comparison
README updates
|
|
Not enough time or interest at the moment to get this
fully working...
|
|
This exposes a client IO object directly to the underlying
application.
|
|
We no longer explicitly close @input
|
|
There's a good chunk of tests that fail with this, still.
Worse, I haven't been able to figure out what's wrong since
it looks like it would involve looking at C++ code...
|
|
Don't expect RevThreadPool to work with Rev <= 0.3.1, either.
|
|
The last change to our EventMachine support code broke
our (lightly-tested) NeverBlock support badly.
|
|
We'll be adding EventMachine-based concurrency models.
|
|
Just let the GC deal with it
|
|
This matches what EM sets for its built-in thread pool
|
|
Rev/Packet-based models may support it in the future
|
|
A client disconnect could possibly trigger IOError on
close whereas EOFError does not occur.
|
|
It may make it harder to switch between concurrency models with
SIGHUP this way...
|
|
This release fixes a memory leak in our existing Revactor
concurrency model. A new RevThreadPool concurrency model has
been added as well as small cleaups to exit handling in workers.
|
|
This should be like RevThreadSpawn except with more predictable
performance (but higher memory usage under low load).
|
|
We now correctly exit!(2) if our master can't kill us.
|
|
This model has basically been rewritten to avoid unbounded
memory growth (slow without keepalive) due to listeners
not properly handling :*_closed messages.
Performance is much more stable as a result, too.
|
|
Just die naturally here if threads don't die on
their own.
|
|
keepalive_timeout (default: 2 seconds) is now supported to
disconnect idle connections. Several new concurrency models
added include: NeverBlock, FiberSpawn and FiberPool; all of
which have only been lightly tested. RevThreadSpawn loses
streaming input support to become simpler and faster for the
general cases. AppPool middleware is now compatible with all
Fiber-based models including Revactor and NeverBlock.
A new document gives a summary of all the options we give you:
http://rainbows.rubyforge.org/Summary.html
If you're using any of the Rev-based concurrency models, the
latest iobuffer (0.1.3) gem will improve performance. Also,
RevThreadSpawn should become usable under MRI 1.8 with the next
release of Rev (0.3.2).
|