We get a lot of those, especially with web crawlers :
recv(2, 0x810b610, 7000, 0) = -1 ECONNRESET (Connection reset by peer)
shutdown(2, 1 /* send */) = -1 ENOTCONN (Transport endpoint is not connected)
close(2) = 0
There's no need to perform the shutdown() here, the socket is already
in error so it is down.
By default we automatically wait for enough data to fill large
packets if buf->to_forward is not null. This causes a problem
with POST/Expect requests which have a data size but no data
immediately available. Instead of causing noticeable delays on
such requests, simply add a flag to disable waiting when sending
requests.
Many times we see a lot of short responses in HTTP (typically 304 on a
reload). It is a waste of network bandwidth to send that many small packets
when we know we can merge them. When we know that another HTTP request is
following a response, we set BF_EXPECT_MORE on the response buffer, which
will turn MSG_MORE on exactly once. That way, multiple short responses can
leave pipelined if their corresponding requests were also pipelined.
While it could be dangerous to enable MSG_MORE on infinite data (eg:
interactive sessions), it makes sense to enable it when we know the
chunk to be sent is just a part of a larger one.
This new flag may be set by any user on a stream interface to tell
the underlying protocol that there is no need for lingering on the
socket since we know the other side either received everything or
does not care about what we sent.
This will typically be used with forced server close in HTTP mode,
where we want to quickly close a server connection after receiving
its response. Otherwise the system would prevent us from reusing
the same port for some time.
Since we'll soon be able to close a connection with remaining data in a
buffer, it becomes obvious that we can prepare to close when we're about
to send the last chunk of data and not the whole buffer.
Since the introduction of the automatic sizing of buffers during reads,
a bug appeared where the max size could be negative, causing large
chunks of memory to be overwritten during recv() calls if a read pointer
was already past the buffer's limit.
We used to apply a limit to each buffer's size in order to leave
some room to rewrite headers, then we used to remove this limit
once the session switched to a data state.
Proceeding that way becomes a problem with keepalive because we
have to know when to stop reading too much data into the buffer
so that we can leave some room again to process next requests.
The principle we adopt here consists in only relying on to_forward+send_max.
Indeed, both of those data define how many bytes will leave the buffer.
So as long as their sum is larger than maxrewrite, we can safely
fill the buffers. If they are smaller, then we refrain from filling
the buffer. This means that we won't risk to fill buffers when
reading last data chunk followed by a POST request and its contents.
The only impact identified so far is that we must ensure that the
BF_FULL flag is correctly dropped when starting to forward. Right
now this is OK because nobody inflates to_forward without using
buffer_forward().
Yohan Tordjman at Dstorage found that upgrading haproxy to 1.4-dev4
caused truncated objects to be returned. An strace quickly exhibited
the issue which was 100% reproducible :
4297 epoll_wait(0, {}, 10, 0) = 0
4297 epoll_wait(0, {{EPOLLIN, {u32=7, u64=7}}}, 10, 1000) = 1
4297 splice(0x7, 0, 0x5, 0, 0xffffffffffffffff, 0x3) = -1 EINVAL (Invalid argument)
4297 shutdown(7, 1 /* send */) = 0
4297 close(7) = 0
4297 shutdown(2, 1 /* send */) = 0
4297 close(2) = 0
This is caused by the fact that the forward length is taken from
BUF_INFINITE_FORWARD, which is -1. The problem does not appear
in 32-bit mode because this value is first cast to an unsigned
long, truncating it to 32-bit (4 GB). Setting an upper bound
fixes the issue.
Also, a second error check has been added for splice. If EINVAL
is returned, we fall back to recv().
When processing a GET or HEAD request in close mode, we know we don't
need to read anything anymore on the socket, so we can disable it.
Doing this can save up to 40% of the recv calls, and half of the
epoll_ctl calls.
For this we need a buffer flag indicating that we're not interesting in
reading anymore. Right now, this flag also disables both polled reads.
We might benefit from disabling only speculative reads, but we will need
at least this flag when we want to support keepalive anyway.
Currently we don't disable the flag on completion, but it does not
matter as we close ASAP when performing the shutw().
Some rarely information are stored in fdtab, making it larger for no
reason (source port ranges, remote address, ...). Such information
lie there because the checks can't find them anywhere else. The goal
will be to move these information to the stream interface once the
checks make use of it.
For now, we move them to an fdinfo array. This simple change might
have improved the cache hit ratio a little bit because a 0.5% of
performance increase has measured.
In old versions, before 1.3.16, we had to refresh the timeouts after
each call to process_session() because the stream socket handler did
not do it. Now that the sockets can exchange data for a long period
without calling process_session(), we can detect an old activity and
refresh a timeout long after the last activity, causing too late a
detection of some timeouts.
The fix simply consists in not checking for activity anymore in
stream_sock_data_finish() but only set a timeout if it was not
previously set.
By default, when data is sent over a socket, both the write timeout and the
read timeout for that socket are refreshed, because we consider that there is
activity on that socket, and we have no other means of guessing if we should
receive data or not.
While this default behaviour is desirable for almost all applications, there
exists a situation where it is desirable to disable it, and only refresh the
read timeout if there are incoming data. This happens on sessions with large
timeouts and low amounts of exchanged data such as telnet session. If the
server suddenly disappears, the output data accumulates in the system's
socket buffers, both timeouts are correctly refreshed, and there is no way
to know the server does not receive them, so we don't timeout. However, when
the underlying protocol always echoes sent data, it would be enough by itself
to detect the issue using the read timeout. Note that this problem does not
happen with more verbose protocols because data won't accumulate long in the
socket buffers.
When this option is set on the frontend, it will disable read timeout updates
on data sent to the client. There probably is little use of this case. When
the option is set on the backend, it will disable read timeout updates on
data sent to the server. Doing so will typically break large HTTP posts from
slow lines, so use it with caution.
We had to add a new stream_interface flag : SI_FL_DONT_WAKE. This flag
is used to indicate that a stream interface is being updated and that
no wake up should be sent to its owner. This will be required for tasks
embedded into stream interfaces. Otherwise, we could have the
owner task send wakeups to itself during status updates, thus
preventing the state from converging. As long as a stream_interface's
status is being monitored and adjusted, there is no reason to wake it
up again, as we know its changes will be seen and considered.
In TCP, we don't want to forward chunks of data, we want to forward
indefinitely. This patch introduces a special value for the amount
of data to be forwarded. When buffer_forward() is called with
BUF_INFINITE_FORWARD, it configures the buffer to never stop
forwarding until the end.
The BF_EMPTY flag was once used to indicate an empty buffer. However,
it was used half the time as meaning the buffer is empty for the reader,
and half the time as meaning there is nothing left to send.
"nothing to send" is only indicated by "->send_max=0 && !pipe". Once
we fix this, we discover that the flag is not used anymore. So the
flags has been renamed BF_OUT_EMPTY and means exactly the condition
above, ie, there is nothing to send.
Doing so has allowed us to remove some unused tests for emptiness,
but also to uncover a certain amount of situations where the flag
was not correctly set or tested.
The BF_WRITE_ENA buffer flag became very complex to deal with, because
it was used to :
- enable automatic connection
- enable close forwarding
- enable data forwarding
The last point was not very true anymore since we introduced ->send_max,
but still the test remained everywhere. This was causing issues such as
impossibility to connect without forwarding data, impossibility to prevent
closing when data was forwarded, etc...
This patch clarifies the situation by getting rid of this multi-purpose
flag and replacing it with :
- data forwarding based only on ->send_max || ->pipe ;
- a new BF_AUTO_CONNECT flag to allow automatic connection and only
that ;
- ability to perform an automatic connection when ->send_max or ->pipe
indicate that data is waiting to leave the buffer ;
- a new BF_AUTO_CLOSE flag to let the producer automatically set the
BF_SHUTW_NOW flag when it gets a BF_SHUTR.
During this cleanup, it was discovered that some tests were performed
twice, or that the BF_HIJACK flag was still tested, which is not needed
anymore since ->send_max replcaed it. These places have been fixed too.
These cleanups have also revealed a few areas where the other flags
such as BF_EMPTY are not cleanly used. This will be an opportunity for
a second patch.
This flag was incorrectly used as meaning "close immediately",
while it needs to say "close ASAP". ASAP here means when unsent
data pending in the buffer are sent. This helps cleaning up some
dirty tricks where the buffer output was checking the BF_SHUTR
flag combined with EMPTY and other such things. Now we have a
clearly defined semantics :
- producer sets SHUTR and *may* set SHUTW_NOW if WRITE_ENA is
set, otherwise leave it to the session processor to set it.
- consumer only checks SHUTW_NOW to decide whether or not to
call shutw().
This also induced very minor changes at some locations which were
not protected against buffer changes while the SHUTW_NOW flag was
set. Now we prevent send_max from changing when the flag is set.
Several tests have been run without any unexpected behaviour detected.
Some more cleanups are needed, as it clearly appears that some tests
could be removed with stricter semantics.
Please consider the following patches. They are required to
compile haproxy-1.4-dev2 on FreeBSD.
Summary:
1) include <sys/types.h> before <netinet/tcp.h>
2) Use IPPROTO_TCP instead of SOL_TCP
(they are both defined as 6, TCP protocol number)
send() supports the MSG_MORE flag on Linux, which does the same
as TCP_CORK except that we don't have to remove TCP_NODELAY before
and we don't need any syscall to set/remove it. This can save up
to 4 syscalls around a send() (two for setting it, two for removing
it), and it's much cleaner since it is not persistent. So make use
of it instead.
The first step towards dynamic buffer size consists in removing
all static definitions of the buffer size. Instead, we store a
buffer's size in itself. Right now they're all preinitialized
to BUFSIZE, but we will change that.
Andrew Azarov reported that haproxy-1.4-dev1 does not build
under FreeBSD 7.2 because SOL_TCP is not defined. So add a
check for its definition before using it. This only impacts
network optimisations anyway.
During a direct data transfer from the server to the client, if the
system did not have enough buffers anymore, haproxy would not enable
write polling again if it could write at least one data chunk. Under
normal conditions, this would remain undetected because the remaining
data would be pushed by next data chunks.
However, when this happens on the last chunk of a session, or the last
in a series in an interactive bidirectional TCP transfer, haproxy would
only start sending again when the read timeout was reached on the side
it stopped writing, causing long pauses on some protocols such as SQL.
This bug was reported by an Exceliance customer who generously offered
to help us by sending large amounts of traces and running various tests
on production systems.
It is quite hard to trigger it but it becomes easier with a ping-pong
TCP service which transfers random data sizes, with a modified version
of send() able to send packets smaller than the average transfer size.
A cleaner fix would imply only updating the write timeout when data
transfers are *attempted*, not succeeded, but that requires more
sensible code changes without fixing the result. It is a candidate
for a later patch though.
As reported by Jean-Baptiste Quenot and Robbie Aelter, sometimes a
backend server error is converted to a 502 error if the backend stops
before reading all the request. The reason is that the remote system
sends a TCP RST packet because there are still unread data pending in
the socket buffer. This RST is translated as a socket error on the
local system, and this error is reported by the poller.
However, most of the time, it's a write error, but the system is
still able to read the remaining pending data, such as in the trace
below :
send(7, "GET /aaa HTTP/1.0\r\nUser-Agent: Mo"..., 1123, MSG_DONTWAIT|MSG_NOSIGNAL) = 1123
epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLIN, {u32=7, u64=7}}) = 0
epoll_wait(3, {{EPOLLIN|EPOLLERR|EPOLLHUP, {u32=7, u64=7}}}, 8, 1000) = 1
gettimeofday({1247593958, 643572}, NULL) = 0
recv(7, "HTTP/1.0 400 Bad request\r\nCache-C"..., 7000, MSG_NOSIGNAL) = 187
setsockopt(6, SOL_TCP, TCP_NODELAY, [0], 4) = 0
setsockopt(6, SOL_TCP, TCP_CORK, [1], 4) = 0
send(6, "HTTP/1.0 400 Bad request\r\nCache-C"..., 187, MSG_DONTWAIT|MSG_NOSIGNAL) = 187
shutdown(6, 1 /* send */) = 0
The recv succeeded while epoll_wait() reported an error.
Note: This case is very hard to reproduce and requires that the backend
server is reached via the loopback in order to minimise latency and
reduce the risk of sent data being ACKed.
When we close a socket with unread data in the buffer, or when the
nolinger option is set, we regularly lose the last fragment, which
often contains the error message. This typically occurs when sending
too large a request. Only the RST is seen due to the close() (since
not all data were read) and the output message never reaches the
network.
Doing a shutdown() before the close() solves this annoying issue
because the data are really pushed before the system sends the RST.
The splice code did not consider compatibility between both ends
of the connection. Now we set different capabilities on each
stream interface, depending on what the protocol can splice to/from.
Right now, only TCP is supported. Thanks to this, we're now able to
automatically detect when splice() is not implemented and automatically
disable it on one end instead of reporting errors to the upper layer.
When the nolinger option is used, we must not close too fast because
some data might be left unsent. Instead we must proceed with a normal
shutdown first, then a close. Also, we want to avoid merging FIN with
the last segment if nolinger is set, because if that one gets lost,
there is no chance for it to be retransmitted.
Setting TCP_CORK on a socket before sending the last segment enables
automatic merging of this segment with the FIN from the shutdown()
call. Playing with TCP_CORK is not easy though as we have to track
the status of the TCP_NODELAY flag since both are mutually exclusive.
Doing so saves one more packet per session and offers about 5% more
performance.
There is no reason not to do it, so there is no associated option.
Under some circumstances, it appears possible to refresh a timeout
just after a side has been shut. For instance, if poll() plans to
call both read and write, and the read side calls chk_snd() which
in turn causes a shutw to occur, then stream_sock_write could update
its write timeout. The same problem happens the other way.
The timeout checks will then not catch these cases because they
ignore timeouts in case of shut{r,w}.
This is very likely to be the major cause of the 100% CPU usages
reported by Bart Bobrowski.
The fix consists in always ensuring that a side is not shut before
updating its timeout.
Upon read or write error, we cannot immediately close the FD because
we want to first report the error to the upper layer which will do it
itself. However, we want to prevent any further I/O from being performed
on the FD. This is especially important in case of speculative I/O where
nothing else could stop the FD from still being polled until the upper
layer takes care of the condition.
The stream_interface timeout was not reset upon a connect success or
error, leading to busy loops when requeuing tasks in the past.
Thanks to Bart Bobrowski for reporting the issue.
When the reader does not expect to read lots of data, it can
set BF_READ_DONTWAIT on the request buffer. When it is set,
the stream_sock_read callback will not try to perform multiple
reads, it will return after only one, and clear the flag.
That way, we can immediately return when waiting for an HTTP
request without trying to read again.
On pure request/responses schemes such as monitor-uri or
redirects, this has completely eliminated the EAGAIN occurrences
and the epoll_ctl() calls, resulting in a performance increase of
about 10%. Similar effects should be observed once we support
HTTP keep-alive since we'll immediately disable reads once we
get a full request.
If we get very large data at once, it's almost certain that it's
worthless trying to read again, because we got everything we could
get.
Doing this has made all -EAGAIN disappear from splice reads. The
threshold has been put in the global tunable structures so that if
we one day want to make it accessible from user config, it will be
easy to do so.
When data are forwarded between socket, we must update the output
socket's write timeout. This was forgotten, causing sessions to
unexpectedly expire during long posts.
As subject when i try to compile haproxy with -DDEBUG_FULL it stop at
stream_sock.c file with:
gcc -Iinclude -Wall -O2 -g -DDEBUG_FULL -DTPROXY -DENABLE_POLL
-DENABLE_EPOLL -DENABLE_SEPOLL -DNETFILTER -DUSE_GETSOCKNAME
-DCONFIG_HAPROXY_VERSION=\"1.3.15\"
-DCONFIG_HAPROXY_DATE=\"2008/04/19\" -c -o src/stream_sock.o
src/stream_sock.c
src/stream_sock.c: In function 'stream_sock_chk_rcv':
src/stream_sock.c:905: error: 'fd' undeclared (first use in this function)
src/stream_sock.c:905: error: (Each undeclared identifier is reported only once
src/stream_sock.c:905: error: for each function it appears in.)
src/stream_sock.c:905: error: 'ob' undeclared (first use in this function)
src/stream_sock.c: In function 'stream_sock_chk_snd':
src/stream_sock.c:940: error: 'fd' undeclared (first use in this function)
src/stream_sock.c:940: error: 'ib' undeclared (first use in this function)
make: *** [src/stream_sock.o] Error 1
With this patch all build fine:
Using pipe pools makes pipe management a lot easier. It also allows to
remove quite a bunch of #ifdefs in areas which depended on the presence
or not of support for kernel splicing.
The buffer now holds a pointer to a pipe structure which is always NULL
except if there are still data in the pipe. When it needs to use that
pipe, it dynamically allocates it from the pipe pool. When the data is
consumed, the pipe is immediately released.
That way, there is no need anymore to care about pipe closure upon
session termination, nor about pipe creation when trying to use
splice().
Another immediate advantage of this method is that it considerably
reduces the number of pipes needed to use splice(). Tests have shown
that even with 0.2 pipe per connection, almost all sessions can use
splice(), because the same pipe may be used by several consecutive
calls to splice().
Kernels before 2.6.27.13 would have splice() return EAGAIN on shutdown.
By adding a few tricks, we can deal with the situation. If splice()
returns EAGAIN and the pipe is empty, then fallback to recv() which
will be able to check if it's an end of connection or not.
The advantage of this method is that it remains transparent for good
kernels since there is no reason that epoll() will return EPOLLIN
without anything to read, and even if it would happen, the recv()
overhead on this check is minimal.
This code provides support for linux 2.6 kernel splicing. This feature
appeared in kernel 2.6.25, but initial implementations were awkward and
buggy. A kernel >= 2.6.29-rc1 is recommended, as well as some optimization
patches.
Using pipes, this code is able to pass network data directly between
sockets. The pipes are a bit annoying to manage (fd creation, release,
...) but finally work quite well.
Preliminary tests show that on high bandwidths, there's a substantial
gain (approx +50%, only +20% with kernel workarounds for corruption
bugs). With 2000 concurrent connections, with Myricom NICs, haproxy
now more easily achieves 4.5 Gbps for 1 process and 6 Gbps for two
processes buffers. 8-9 Gbps are easily reached with smaller numbers
of connections.
We also try to splice out immediately after a splice in by making
profit from the new ability for a data producer to notify the
consumer that data are available. Doing this ensures that the
data are immediately transferred between sockets without latency,
and without having to re-poll. Performance on small packets has
considerably increased due to this method.
Earlier kernels return only one TCP segment at a time in non-blocking
splice-in mode, while newer return as many segments as may fit in the
pipe. To work around this limitation without hurting more recent kernels,
we try to collect as much data as possible, but we stop when we believe
we have read 16 segments, then we forward everything at once. It also
ensures that even upon shutdown or EAGAIN the data will be forwarded.
Some tricks were necessary because the splice() syscall does not make
a difference between missing data and a pipe full, it always returns
EAGAIN. The trick consists in stop polling in case of EAGAIN and a non
empty pipe.
The receiver waits for the buffer to be empty before using the pipe.
This is in order to avoid confusion between buffer data and pipe data.
The BF_EMPTY flag now covers the pipe too.
Right now the code is disabled by default. It needs to be built with
CONFIG_HAP_LINUX_SPLICE, and the instances intented to use splice()
must have "option splice-response" (or option splice-request) enabled.
It is probably desirable to keep a pool of pre-allocated pipes to
avoid having to create them for every session. This will be worked
on later.
Preliminary tests show very good results, even with the kernel
workaround causing one memcpy(). At 3000 connections, performance
has moved from 3.2 Gbps to 4.7 Gbps.
Some older libc don't define the splice() syscall, and some even
define a wrong one. For this reason, we try our best to declare
it correctly. These definitions still work with recent glibc.
When the producer calls stream_sock_chk_snd(), we now try to send
all pending data asynchronously. If it succeeds, we don't have to
enable polling on the FD which saves about half of the calls to
epoll_wait().
In stream_sock_read(), we finally set the WAIT_ROOM flag as soon as
possible, in preparation of the splice code. We reset it when we
detect that some room has been released either in the buffer or in
the splice.
The condition to cakk ->chk_snd() in stream_sock_read() was suboptimal
because we did not call it when the socket was shut down nor when there
was an error after data were added.
Now we ensure to call is whenever there are data pending.
Also, the "full" condition was handled before calling chk_snd(), which
could cause deadlock issues if chk_snd() did consume some data.
stream_sock_write() has been split in two parts :
- the poll callback, intented to be called when an I/O event has
been detected
- the write() core function, which ought to be usable from various
other places, possibly not meant to wake the task up.
The code has also been slightly cleaned up in the process. It's more
readable now.
Some tricks to handle situations where we write nothing were in the
middle of the main loop in stream_sock_write(). This cleanup provides
better source and object code, and slightly shrinks the output code.
This construct collapses into ((flags & (X|Y)) == X) when X is a
single-bit flag. This provides a noticeable code shrink and the
output code results in less conditional jumps.
