RepoMirrors
/
haproxy
mirror of http://git.haproxy.org/git/haproxy.git/
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
haproxy public development tree
6,625 Commits 20 Branches 367 Tags 200 MiB
C 98.1%
Shell 0.8%
Makefile 0.6%
Python 0.2%
Lua 0.1%
Go to file
Bin Wang 95fad5ba4b BUG/MAJOR: stream-int: don't re-arm recv if send fails 
When
    1) HAProxy configured to enable splice on both directions
    2) After some high load, there are 2 input channels with their socket buffer
       being non-empty and pipe being full at the same time, sitting in `fd_cache`
       without any other fds.

The 2 channels will repeatedly be stopped for receiving (pipe full) and waken
for receiving (data in socket), thus getting out and in of `fd_cache`, making
their fd swapping location in `fd_cache`.

There is a `if (entry < fd_cache_num && fd_cache[entry] != fd) continue;`
statement in `fd_process_cached_events` to prevent frequent polling, but since
the only 2 fds are constantly swapping location, `fd_cache[entry] != fd` will
always hold true, thus HAProxy can't make any progress.

The root cause of the issue is dual :
  - there is a single fd_cache, for next events and for the ones being
    processed, while using two distinct arrays would avoid the problem.

  - the write side of the stream interface wakes the read side up even
    when it couldn't write, and this one really is a bug.

Due to CF_WRITE_PARTIAL not being cleared during fast forwarding, a failed
send() attempt will still cause ->chk_rcv() to be called on the other side,
re-creating an entry for its connection fd in the cache, causing the same
sequence to be repeated indefinitely without any opportunity to make progress.

CF_WRITE_PARTIAL used to be used for what is present in these tests : check
if a recent write operation was performed. It's part of the CF_WRITE_ACTIVITY
set and is tested to check if timeouts need to be updated. It's also used to
detect if a failed connect() may be retried.

What this patch does is use CF_WROTE_DATA() to check for a successful write
for connection retransmits, and to clear CF_WRITE_PARTIAL before preparing
to send in stream_int_notify(). This way, timeouts are still updated each
time a write succeeds, but chk_rcv() won't be called anymore after a failed
write.

It seems the fix is required all the way down to 1.5.

Without this patch, the only workaround at this point is to disable splicing
in at least one direction. Strictly speaking, splicing is not absolutely
required, as regular forwarding could theorically cause the issue to happen
if the timing is appropriate, but in practice it appears impossible to
reproduce it without splicing, and even with splicing it may vary.

The following config manages to reproduce it after a few attempts (haproxy
going 100% CPU and having to be killed) :

  global
      maxpipes 50000
      maxconn 10000

  listen srv1
      option splice-request
      option splice-response
      bind :8001
      server s1 127.0.0.1:8002

  server$ tcploop 8002 L N20 A R10 S1000000 R10 S1000000 R10 S1000000 R10 S1000000 R10 S1000000
  client$ tcploop 8001 N20 C T S1000000 R10 J
 		2017-10-05 11:20:16 +02:00
contrib BUG/MINOR: contrib/halog: fixing small memory leak 2017-10-03 13:52:45 +02:00
doc MINOR: cli: add socket commands and config to prepend informational messages with severity 2017-09-13 13:37:59 +02:00
ebtree
…
examples [RELEASE] Released version 1.8-dev2 2017-06-02 15:59:51 +02:00
include MINOR: ist: add a macro to ease const array initialization 2017-09-21 15:32:31 +02:00
scripts SCRIPTS: create-release: enforce GIT_COMMITTER_{NAME|EMAIL} validity 2017-06-16 12:43:53 +02:00
src BUG/MAJOR: stream-int: don't re-arm recv if send fails 2017-10-05 11:20:16 +02:00
tests TESTS: checks: add a simple test config for tcp-checks 2017-10-04 16:29:19 +02:00
.gitignore MINOR: spoe: add random ip-reputation service as SPOA example 2016-11-09 22:57:02 +01:00
CHANGELOG [RELEASE] Released version 1.8-dev2 2017-06-02 15:59:51 +02:00
CONTRIBUTING DOC: update CONTRIBUTING regarding optional parts and message format 2017-07-18 06:56:40 +02:00
LICENSE
…
MAINTAINERS DOC: add Ben Shillito as the maintainer of 51d 2016-01-13 12:10:23 +01:00
Makefile BUILD: Makefile: improve detection of support for compiler warnings 2017-09-14 19:05:45 +02:00
README DOC: update the list of OpenSSL versions in the README 2017-07-18 07:01:26 +02:00
ROADMAP DOC: update the roadmap file with the latest changes 2016-11-25 16:32:20 +01:00
SUBVERS
…
VERDATE [RELEASE] Released version 1.8-dev2 2017-06-02 15:59:51 +02:00
VERSION [RELEASE] Released version 1.8-dev2 2017-06-02 15:59:51 +02:00

README 

                         ----------------------
                             HAProxy how-to
                         ----------------------
                              version 1.8
                             willy tarreau
                               2017/06/02


1) How to build it
------------------

This is a development version, so it is expected to break from time to time,
to add and remove features without prior notification and it should not be used
in production. If you are not used to build from sources or if you are not used
to follow updates then it is recommended that instead you use the packages provided
by your software vendor or Linux distribution. Most of them are taking this task
seriously and are doing a good job at backporting important fixes. If for any
reason you'd prefer a different version than the one packaged for your system,
you want to be certain to have all the fixes or to get some commercial support,
other choices are available at :

                        http://www.haproxy.com/

To build haproxy, you will need :
  - GNU make. Neither Solaris nor OpenBSD's make work with the GNU Makefile.
    If you get many syntax errors when running "make", you may want to retry
    with "gmake" which is the name commonly used for GNU make on BSD systems.
  - GCC between 2.95 and 4.8. Others may work, but not tested.
  - GNU ld

Also, you might want to build with libpcre support, which will provide a very
efficient regex implementation and will also fix some badness on Solaris' one.

To build haproxy, you have to choose your target OS amongst the following ones
and assign it to the TARGET variable :

  - linux22     for Linux 2.2
  - linux24     for Linux 2.4 and above (default)
  - linux24e    for Linux 2.4 with support for a working epoll (> 0.21)
  - linux26     for Linux 2.6 and above
  - linux2628   for Linux 2.6.28, 3.x, and above (enables splice and tproxy)
  - solaris     for Solaris 8 or 10 (others untested)
  - freebsd     for FreeBSD 5 to 10 (others untested)
  - netbsd      for NetBSD
  - osx         for Mac OS/X
  - openbsd     for OpenBSD 5.7 and above
  - aix51       for AIX 5.1
  - aix52       for AIX 5.2
  - cygwin      for Cygwin
  - haiku       for Haiku
  - generic     for any other OS or version.
  - custom      to manually adjust every setting

You may also choose your CPU to benefit from some optimizations. This is
particularly important on UltraSparc machines. For this, you can assign
one of the following choices to the CPU variable :

  - i686 for intel PentiumPro, Pentium 2 and above, AMD Athlon
  - i586 for intel Pentium, AMD K6, VIA C3.
  - ultrasparc : Sun UltraSparc I/II/III/IV processor
  - native : use the build machine's specific processor optimizations. Use with
    extreme care, and never in virtualized environments (known to break).
  - generic : any other processor or no CPU-specific optimization. (default)

Alternatively, you may just set the CPU_CFLAGS value to the optimal GCC options
for your platform.

You may want to build specific target binaries which do not match your native
compiler's target. This is particularly true on 64-bit systems when you want
to build a 32-bit binary. Use the ARCH variable for this purpose. Right now
it only knows about a few x86 variants (i386,i486,i586,i686,x86_64), two
generic ones (32,64) and sets -m32/-m64 as well as -march=<arch> accordingly.

If your system supports PCRE (Perl Compatible Regular Expressions), then you
really should build with libpcre which is between 2 and 10 times faster than
other libc implementations. Regex are used for header processing (deletion,
rewriting, allow, deny). The only inconvenient of libpcre is that it is not
yet widely spread, so if you build for other systems, you might get into
trouble if they don't have the dynamic library. In this situation, you should
statically link libpcre into haproxy so that it will not be necessary to
install it on target systems. Available build options for PCRE are :

  - USE_PCRE=1 to use libpcre, in whatever form is available on your system
    (shared or static)

  - USE_STATIC_PCRE=1 to use a static version of libpcre even if the dynamic
    one is available. This will enhance portability.

  - with no option, use your OS libc's standard regex implementation (default).
    Warning! group references on Solaris seem broken. Use static-pcre whenever
    possible.

If your system doesn't provide PCRE, you are encouraged to download it from
http://www.pcre.org/ and build it yourself, it's fast and easy.

Recent systems can resolve IPv6 host names using getaddrinfo(). This primitive
is not present in all libcs and does not work in all of them either. Support in
glibc was broken before 2.3. Some embedded libs may not properly work either,
thus, support is disabled by default, meaning that some host names which only
resolve as IPv6 addresses will not resolve and configs might emit an error
during parsing. If you know that your OS libc has reliable support for
getaddrinfo(), you can add USE_GETADDRINFO=1 on the make command line to enable
it. This is the recommended option for most Linux distro packagers since it's
working fine on all recent mainstream distros. It is automatically enabled on
Solaris 8 and above, as it's known to work.

It is possible to add native support for SSL using the GNU makefile, by passing
"USE_OPENSSL=1" on the make command line. The libssl and libcrypto will
automatically be linked with haproxy. Some systems also require libz, so if the
build fails due to missing symbols such as deflateInit(), then try again with
"ADDLIB=-lz".

Your are strongly encouraged to always use an up-to-date version of OpenSSL, as
found on https://www.openssl.org/ as vulnerabilities are occasionally found and
you don't want them on your systems. HAProxy is known to build correctly on all
currently supported branches (0.9.8, 1.0.0, 1.0.1, 1.0.2 and 1.1.0 at the time
of writing). Branch 1.0.2 is currently recommended for the best combination of
features and stability. Asynchronous engines require OpenSSL 1.1.0 though. It's
worth mentionning that some OpenSSL derivatives are also reported to work but
may occasionally break. Patches to fix them are welcome but please read the
CONTRIBUTING file first.

To link OpenSSL statically against haproxy, build OpenSSL with the no-shared
keyword and install it to a local directory, so your system is not affected :

    $ export STATICLIBSSL=/tmp/staticlibssl
    $ ./config --prefix=$STATICLIBSSL no-shared
    $ make && make install_sw

When building haproxy, pass that path via SSL_INC and SSL_LIB to make and
include additional libs with ADDLIB if needed (in this case for example libdl):

    $ make TARGET=linux26 USE_OPENSSL=1 SSL_INC=$STATICLIBSSL/include SSL_LIB=$STATICLIBSSL/lib ADDLIB=-ldl

It is also possible to include native support for zlib to benefit from HTTP
compression. For this, pass "USE_ZLIB=1" on the "make" command line and ensure
that zlib is present on the system. Alternatively it is possible to use libslz
for a faster, memory less, but slightly less efficient compression, by passing
"USE_SLZ=1".

Zlib is commonly found on most systems, otherwise updates can be retrieved from
http://www.zlib.net/. It is easy and fast to build. Libslz can be downloaded
from http://1wt.eu/projects/libslz/ and is even easier to build.

By default, the DEBUG variable is set to '-g' to enable debug symbols. It is
not wise to disable it on uncommon systems, because it's often the only way to
get a complete core when you need one. Otherwise, you can set DEBUG to '-s' to
strip the binary.

For example, I use this to build for Solaris 8 :

    $ make TARGET=solaris CPU=ultrasparc USE_STATIC_PCRE=1

And I build it this way on OpenBSD or FreeBSD :

    $ gmake TARGET=freebsd USE_PCRE=1 USE_OPENSSL=1 USE_ZLIB=1

And on a classic Linux with SSL and ZLIB support (eg: Red Hat 5.x) :

    $ make TARGET=linux26 USE_PCRE=1 USE_OPENSSL=1 USE_ZLIB=1

And on a recent Linux >= 2.6.28 with SSL and ZLIB support :

    $ make TARGET=linux2628 USE_PCRE=1 USE_OPENSSL=1 USE_ZLIB=1

In order to build a 32-bit binary on an x86_64 Linux system with SSL support
without support for compression but when OpenSSL requires ZLIB anyway :

    $ make TARGET=linux26 ARCH=i386 USE_OPENSSL=1 ADDLIB=-lz

The SSL stack supports session cache synchronization between all running
processes. This involves some atomic operations and synchronization operations
which come in multiple flavors depending on the system and architecture :

  Atomic operations :
    - internal assembler versions for x86/x86_64 architectures

    - gcc builtins for other architectures. Some architectures might not
      be fully supported or might require a more recent version of gcc.
      If your architecture is not supported, you willy have to either use
      pthread if supported, or to disable the shared cache.

    - pthread (posix threads). Pthreads are very common but inter-process
      support is not that common, and some older operating systems did not
      report an error when enabling multi-process mode, so they used to
      silently fail, possibly causing crashes. Linux's implementation is
      fine. OpenBSD doesn't support them and doesn't build. FreeBSD 9 builds
      and reports an error at runtime, while certain older versions might
      silently fail. Pthreads are enabled using USE_PTHREAD_PSHARED=1.

  Synchronization operations :
    - internal spinlock : this mode is OS-independant, light but will not
      scale well to many processes. However, accesses to the session cache
      are rare enough that this mode could certainly always be used. This
      is the default mode.

    - Futexes, which are Linux-specific highly scalable light weight mutexes
      implemented in user-space with some limited assistance from the kernel.
      This is the default on Linux 2.6 and above and is enabled by passing
      USE_FUTEX=1

    - pthread (posix threads). See above.

If none of these mechanisms is supported by your platform, you may need to
build with USE_PRIVATE_CACHE=1 to totally disable SSL cache sharing. Then
it is better not to run SSL on multiple processes.

If you need to pass other defines, includes, libraries, etc... then please
check the Makefile to see which ones will be available in your case, and
use the USE_* variables in the Makefile.

AIX 5.3 is known to work with the generic target. However, for the binary to
also run on 5.2 or earlier, you need to build with DEFINE="-D_MSGQSUPPORT",
otherwise __fd_select() will be used while not being present in the libc, but
this is easily addressed using the "aix52" target. If you get build errors
because of strange symbols or section mismatches, simply remove -g from
DEBUG_CFLAGS.

You can easily define your own target with the GNU Makefile. Unknown targets
are processed with no default option except USE_POLL=default. So you can very
well use that property to define your own set of options. USE_POLL can even be
disabled by setting USE_POLL="". For example :

    $ gmake TARGET=tiny USE_POLL="" TARGET_CFLAGS=-fomit-frame-pointer


1.1) Device Detection
---------------------

HAProxy supports several device detection modules relying on third party
products. Some of them may provide free code, others free libs, others free
evaluation licenses. Please read about their respective details in the
following files :

    doc/DeviceAtlas-device-detection.txt   for DeviceAtlas
    doc/51Degrees-device-detection.txt     for 51Degrees
    doc/WURFL-device-detection.txt         for Scientiamobile WURFL


2) How to install it
--------------------

To install haproxy, you can either copy the single resulting binary to the
place you want, or run :

    $ sudo make install

If you're packaging it for another system, you can specify its root directory
in the usual DESTDIR variable.


3) How to set it up
-------------------

There is some documentation in the doc/ directory :

    - intro.txt : this is an introduction to haproxy, it explains what it is
      what it is not. Useful for beginners or to re-discover it when planning
      for an upgrade.

    - architecture.txt : this is the architecture manual. It is quite old and
      does not tell about the nice new features, but it's still a good starting
      point when you know what you want but don't know how to do it.

    - configuration.txt : this is the configuration manual. It recalls a few
      essential HTTP basic concepts, and details all the configuration file
      syntax (keywords, units). It also describes the log and stats format. It
      is normally always up to date. If you see that something is missing from
      it, please report it as this is a bug. Please note that this file is
      huge and that it's generally more convenient to review Cyril Bont<6E>'s
      HTML translation online here :

           http://cbonte.github.io/haproxy-dconv/configuration-1.6.html

    - management.txt : it explains how to start haproxy, how to manage it at
      runtime, how to manage it on multiple nodes, how to proceed with seamless
      upgrades.

    - gpl.txt / lgpl.txt : the copy of the licenses covering the software. See
      the 'LICENSE' file at the top for more information.

    - the rest is mainly for developers.

There are also a number of nice configuration examples in the "examples"
directory as well as on several sites and articles on the net which are linked
to from the haproxy web site.


4) How to report a bug
----------------------

It is possible that from time to time you'll find a bug. A bug is a case where
what you see is not what is documented. Otherwise it can be a misdesign. If you
find that something is stupidly design, please discuss it on the list (see the
"how to contribute" section below). If you feel like you're proceeding right
and haproxy doesn't obey, then first ask yourself if it is possible that nobody
before you has even encountered this issue. If it's unlikely, the you probably
have an issue in your setup. Just in case of doubt, please consult the mailing
list archives :

                        http://marc.info/?l=haproxy

Otherwise, please try to gather the maximum amount of information to help
reproduce the issue and send that to the mailing list :

                        haproxy@formilux.org

Please include your configuration and logs. You can mask your IP addresses and
passwords, we don't need them. But it's essential that you post your config if
you want people to guess what is happening.

Also, keep in mind that haproxy is designed to NEVER CRASH. If you see it die
without any reason, then it definitely is a critical bug that must be reported
and urgently fixed. It has happened a couple of times in the past, essentially
on development versions running on new architectures. If you think your setup
is fairly common, then it is possible that the issue is totally unrelated.
Anyway, if that happens, feel free to contact me directly, as I will give you
instructions on how to collect a usable core file, and will probably ask for
other captures that you'll not want to share with the list.


5) How to contribute
--------------------

Please carefully read the CONTRIBUTING file that comes with the sources. It is
mandatory.

-- end