Subsequent to the recent stream-int updates, we started to consider that
SI_FL_WANT_PUT needs to be set when receipt is enabled, but this is wrong
and results in 100% CPU when an HTTP client stays idle after a keep-alive
request because the stream-int has nothing to provide and nothing to send.
In fact just like for applets this flag should reflect the continuation
of an attempt. So it's si_cs_recv() which should set the flag, and clear
it if it has nothing more to provide. This function is called the first
time in process_stream()), and called again during transfers, so it will
always be up to date during stream_int_update() and stream_int_notify().
As a special case, it should also be set when a connection switches to
the established state. And we should absolutely refrain from calling
si_cs_recv() to re-enable reading, normally just setting this flag
(from within the stream-int's handler or prior to calling si_chk_rcv())
is expected to be OK.
A corner case remains where it was observed that in stream_int_notify() we
can sometimes be called with an empty output channel with SI_FL_WAIT_ROOM
and no CF_WRITE_PARTIAL, so there's no way to detect that we should
re-enable receiving. It's easy to also take care of this condition
there for the time it takes to figure if this situation is expected
or not.
Now it becomes more obvious that relying on a single flag to request
room (or on two flags to arbiter activity) is not workable given the
autonomy of both sides. The mux_h2 has taught us that blocking flags
are much more reliable, require much less condition and are much easier
to deal with. That's probably something to consider quickly in this
area.
No backport is needed.
Before calling the mux to get incoming data, we get the amount of space
available at the input of the buffer. If there is no space, we don't try to read
more data. This is good enough when raw data are stored in the buffer. But this
info has no meaning when structured data are stored. Because with the HTTP
refactoring, such kind of data will be stored in buffers, it is a bit annoying.
So, to avoid any problems, we always call the mux. It is the mux's responsiblity
to notify the stream interface it needs more space to store more data. This must
be done by setting the flag CS_FL_RCV_MORE on the conn_stream.
This is exactly what we do in the pass-through mux when <count> is null.
This flag is set on the stream interface when we should wait for more space in
the channel's buffer to store more incoming data. This means we should wait some
outgoing data are sent before retrying to receive more data.
But in stream interface functions, at many places, instead of checking this
flag, we use the function channel_may_recv to know if we can (re)start
reading. This currently works but it is not really consistent. And, it works
because only raw data are stored in buffers. But it will be a problem when we
start to store structured data in buffers.
So to avoid any problems with futur implementations, we now rely only on
SI_FL_WAIT_ROOM. The function channel_may_recv can still be called, but only
when we are sure to handle raw data (for instance in functions ci_put*). To do
so, among other things, we must be sure to unset SI_FL_WAIT_ROOM and offer an
opportunity to call chk_rcv() on a stream interface when some data are sent
on the other end, which is now granted by the previous patch series.
We exclusively use stream_int_update() now, the lower layers are not
called anymore so let's remove them, as well as si_update() which used
to be their wrapper.
It's far from being clean, but at least it allows to resync both CS and
applets from the same place, taking into account the fact that CS are
processed synchronously for the send side while appletx are processed
outside of the process_stream() loop. The arrangement is optimised to
minimize the amount of iteration by handling send first, then updating
the SI_FL_WAIT_ROOM flags and only then dealing with si_chk_rcv() on
both sides. The SI_FL_WANT_PUT flag is set if needed before calling
si_chk_rcv() since this is done prior to calling stream_int_update().
Now there's no risk that stream_int_notify() is called anymore during
such operations, thus we cannot have any spurious wake-up anymore. The
case where a successful send() could complete a pending connect() is
handled by taking any stream-int state changes into account at the
call place, which is normal since process_stream() is designed to
iterate till stabilisation.
Doing this solves most of the remaining inconsistencies between CS and
applets.
The function used to be called in turn for each side of the stream, but
since it's called exclusively from process_stream(), it prevents us from
making use of the knowledge we have of the operations in progress for
each side, resulting in having to go all the way through functions like
stream_int_notify() which are not appropriate there.
That patch creates a new function, si_update_both() which takes two
stream interfaces expected to belong to the same stream, and processes
their flags in a more suitable order, but for now doesn't change the
logic at all.
The next step will consist in trying to reinsert the rest of the socket
layer-specific update code to ultimately update the flags correctly at
the end of the operation.
Instead of clearing the SI_FL_WAIT_ROOM flag and losing the information
about the need from the producer to be woken up, we now call si_chk_rcv()
immediately. This is cheap to do and it could possibly be further improved
by only doing it when SI_FL_WAIT_ROOM was still set, though this will
require some extra auditing of the code paths.
The only remaining place where the flag was cleared without a call to
si_chk_rcv() is si_alloc_ibuf(), but since this one is called from a
receive path woken up from si_chk_rcv() or not having failed, the
clearing was not necessary anymore either.
And there was one place in stream_int_notify() where si_chk_rcv() was
called with SI_FL_WAIT_ROOM still explicitly set so this place was
adjusted in order to clear the flag prior to calling si_chk_rcv().
Now we don't have any situation where we randomly clear SI_FL_WAIT_ROOM
without trying to wake the other side up, nor where we call si_chk_rcv()
with the flag set, so this flag should accurately represent a failed
attempt at putting data into the buffer.
When CF_DONT_READ is set, till now we used to set SI_FL_WAIT_ROOM, which
is not appropriate since it would lose the subscribe status. Instead let's
clear SI_FL_WANT_PUT (just like applets do), and set the flag only when
CF_DONT_READ is cleared.
We have to do this in stream_int_update(), and in si_cs_io_cb() after
returning from si_cs_recv() since it would be a bit invasive to hack
this one for now. It must not be done in stream_int_notify() otherwise
it would re-enable blocked applets.
Last, when si_chk_rcv() is called, it immediately clears the flag before
calling ->chk_rcv() so that we are not tempted to uselessly loop on the
same call until the receive function is called. This is the same principle
as what is done with the applet scheduler.
This flag should already be cleared before calling the *chk_rcv() functions.
Before adapting all call places, let's first make sure si_chk_rcv() clears
it before calling them so that these functions do not have to check it again
and so that they do not adjust it. This function will only call the lower
layers if the SI_FL_WANT_PUT flag is present so that the endpoint can decide
not to be called (as done with applets).
We now do this on the si_cs_recv() path so that we always have
SI_FL_WANT_PUT properly set when there's a need to receive and
SI_FL_WAIT_ROOM upon failure.
It doesn't make sense to limit this code to applets, as any stream
interface can use it. Let's rename it by simply dropping the "applet_"
part of the name. No other change was made except updating the comments.
Commit eafd8ebcf ("MEDIUM: stream-int: call si_cs_process() in
stream_int_update_conn") uncovered a sleeping bug. By calling
si_cs_process() within si_update(), we end up calling stream_int_notify().
We rely on it to update the stream-int before quitting as a hack, but
it happens to immediately wake the task up while the stream int's
state is still SI_ST_CON (during the connection establishment). The
observable effect is that an unreachable server causes haproxy to
use 100% CPU until the connection timeout strikes.
This patch fixes this by not causing the wake up for the SI_ST_CON
state. It would equally be possible to check for states higher than
SI_ST_EST as is done in other places, but for now better stay on the
safe side by covering the only issue that can be triggered. It's
suspected that this issue slightly affects older versions by causing
one extra call to process_stream() during the connection setup for
each activity change on the other side, but this should not have
any observable effect.
No backport is needed.
It took me 17 minutes this morning to figure where si->wait_event was
set (it's in si_reset() which should now probably be renamed since it
doesn't just perform a reset anymore but also an allocation) and what
its task was assigned to (si_cs_io_cb() even for applets and empty SI).
This is too confusing and not intuitive enough, let's at least add a
few comments for now to help figure how this stuff works next time.
This one causes some events to be lost. It has already been tested in
an experimental branch but was not merged until being certain it was
needed. Fred figured that requesting /?k=1&s=447392 from httpterm through
haproxy-master was enough to stall the transfer.
No backport is needed, this only affects 1.9-dev5.
There's a call there to si_cs_send() while we're supposed to come from
si_cs_io_cb() which has just done it. But in fact we can also come here
as a lower layer callback from ->wake() after a connection is established.
Since most of the time we'll end up here with either no data in the buffer
or a blocked output, let's simply check if we're already susbcribed to send
events before calling si_cs_send().
stream_int_notify() is I/O agnostic and should not wake up the tasklet,
it's up to si_cs_process() to do that, just like si_applet_wake_cb()
does it for the applet.
This one was added by commit 53216e7db ("MEDIUM: connections: Don't
directly mess with the polling from the upper layers.") after the
removal of the conditional cs_want_send() call. But after analysis
it turned out that it's not needed since the si_cs_send() call will
either succeed or subscribe.
Calling si_cs_send() alone is always dangerous because it can result
in the loss of an event if it manages to empty the buffer. Indeed, in
this case it's critical to call si_chk_rcv() on the opposite stream-int.
Given that si_cs_process() takes care of all this, let's call it instead.
All this code could possibly be refined soon to avoid redoing the whole
stream_int_notify() and do it only after a send(), but at the moment it's
not important.
With the new synchronous si_cs_send() at the end of process_stream(),
we're seeing re-appear the I/O layer specific part of the stream interface
which is supposed to deal with I/O event subscription. The only difference
is that now we subscribe to I/Os only after having attempted (and failed)
them.
This patch brings a cleanup in this by reintroducing stream_int_update_conn()
with the send code from process_stream(). However this alone would not be
enough because the flags which are cleared afterwards would result in the
loss of the possible events (write events only at the moment). So the flags
clearing and stream-int state updates are also performed inside si_update()
between the generic code and the I/O specific code. This definitely makes
sense as after this call we can simply check again for channel and SI flag
changes and decide to loop once again or not.
Well that's only 3 places (applet.c, stream_interface.c, hlua.c). This
ensures we always clear SI_FL_WAIT_ROOM before setting it on failure,
so that it is granted that SI_FL_WAIT_ROOM always indicates a lack of
room for doing an operation, including the inability to allocate a
buffer for this.
The behaviour of the flag CF_WRITE_PARTIAL was modified by commit
95fad5ba4 ("BUG/MAJOR: stream-int: don't re-arm recv if send fails") due
to a situation where it could trigger an immediate wake up of the other
side, both acting in loops via the FD cache. This loss has caused the
need to introduce CF_WRITE_EVENT as commit c5a9d5bf, to replace it, but
both flags express more or less the same thing and this distinction
creates a lot of confusion and complexity in the code.
Since the FD cache now acts via tasklets, the issue worked around in the
first patch no longer exists, so it's more than time to kill this hack
and to restore CF_WRITE_PARTIAL's semantics (i.e.: there has been some
write activity since we last left process_stream).
This patch mostly reverts the two commits above. Only the part making
use of CF_WROTE_DATA instead of CF_WRITE_PARTIAL to detect the loss of
data upon connection setup was kept because it's more accurate and
better suited.
With the previous connection model, when we purposely decided to stop
receiving in order to avoid polling after a complete request was received
for example, it was needed to set SI_FL_WAIT_ROOM to prevent receive
polling from being re-armed. Now with the new subscription-based model
there is no such thing anymore and there is noone to remove this flag
either. Thus if a request takes more than one packet to come in or
spans over too many packets, this flag will cause it to wait forever.
Let's simply remove this flag now.
This patch should not be backported since older versions still need
that this flag is set here to stop receiving.
Avoid using conn_xprt_want_send/recv, and totally nuke cs_want_send/recv,
from the upper layers. The polling is now directly handled by the connection
layer, it is activated on subscribe(), and unactivated once we got the event
and we woke the related task.
When subscribing, we don't need to provide a list element, only the h2 mux
needs it. So instead, Add a list element to struct h2s, and use it when a
list is needed.
This forces us to use the unsubscribe method, since we can't just unsubscribe
by using LIST_DEL anymore.
This patch is larger than it should be because it includes some renaming.
Instead of using si_cs_io_cb() in process_stream() use si_cs_send/si_cs_recv
instead, as si_cs_io_cb() may lead to process_stream being woken up when it
shouldn't be, and thus timeout would never get triggered.
Instead of waiting for the connection layer to let us know we can read,
attempt to receive as soon as process_stream() is called, and subscribe
to receive events if we can't receive yet.
Now, except for idle connections, the recv(), send() and wake() methods are
no more, all the lower layers do is waking tasklet for anybody waiting
for I/O events.
Remove the recv() method from mux and conn_stream.
The goal is to always receive from the upper layers, instead of waiting
for the connection later. For now, recv() is still called from the wake()
method, but that should change soon.
For struct connection, struct conn_stream, and for the h2 mux, add 2 new
lists, one that handles waiters for recv, and one that handles waiters for
recv and send. That way we can ask to subscribe for either recv or send.
Call si_cs_send() at the beginning of si_cs_wake_cb(), instead of from
stream_int_notify-), so that if we get a connection error while trying to
send, the stream_interface will get SI_FL_ERR, the associated task will
be woken up, and the connection will be properly destroyed.
No backport needed.
If we subscribed to send, and the callback is called, call the wake callback
after, so that process_stream() may be woken up if needed.
This is 1.9-specific, no backport is needed.
It is possible that the conn_stream gets detached from the stream_interface,
and as it subscribed to the wait list, si_cs_io_cb() gets called anyway,
so make sure we have a conn_stream before attempting to send more data.
This is 1.9-specific, no backport is needed.
Instead of just using the conn_stream wait_list, give the stream_interface
its own. When the conn_stream will have its own buffers, the stream_interface
may have to wait on it.
Instead of using si_cs_send() as a task handler, define a new function,
si_cs_io_cb(), and give si_cs_send() its original prototype. Right now
si_cs_io_cb() just handles send, but later it'll handle recv() too.
It is mandatory to be sure to process data blocked in the RX buffer of the
conn_stream while the shutr/read0 was already processed. The stream interface
doesn't need to rely on this flags because it already tests CS_FL_EOS.
This function is generic and is able to automatically transfer data from a
buffer to the conn_stream's tx buffer. It does this automatically if the mux
doesn't define another snd_buf() function.
It cannot yet be used as-is with the conn_stream's txbuf without risking to
lose data on close since conn_streams need to be orphaned for this.
There is a long-time issue which affects some applets, at least the stats
applet. If a large stats page is read over a slow link, regularly the
channel's buffer contains too many response data to allow another round
of ci_putblk() to copy a new message. In this case the applet calls
si_applet_cant_put() to mention that it failed to emit data into the
channel's buffer, and wants to be called only once some room is made.
The problem is that stream_int_update(), which is called from
process_stream(), will clear this flag whenever it sees there's some
spare room in the channel's buffer. It causes the applet to be woken
again immediately. This is very visible when reading a large stats
page over a slow link, because in this case haproxy will run at 100%
CPU and strace shows mostly epoll_wait(0). It is very likely that some
other applets like CLI, Lua, peers or SPOE have also been affected but
that the effect were less noticeable because it was mixed with traffic.
Ideally stream_int_update() should not touch these flags, but changing
this would require a very careful auditing of all users. Instead here
what we do is that we respect the flag if the channel still has output
data. This way the flag will automatically disappear once the buffer is
empty, and the applet function will be called only when input data
remains, if at all.
This patch alone is not enough to observe the behaviour change on the
stats page because another bug takes over, addressed by next patch
(BUG/MEDIUM: stats: don't ask for more data as long as we're responding).
When both are applied, dumping stats for 5k backends over a 10 Mbps link
take 1% CPU instead of 100%, with 1.5k epoll_wait() calls instead of 80k.
This fix should be backported at least as far as 1.5.
If the cs has data pending or shutdown and the input channel is still
waiting for reads, let's simply call the recv() function from the wake()
callback. This will allow the lower layers to simply wake the upper one
up without having to consider the recv() nor anything else.
This function is generic and is able to automatically transfer data
from a conn_stream's rx buffer to the destination buffer. It does this
automatically if the mux doesn't define another rcv_buf() function.
This new function wl_set_waitcb() prepopulates a wait_list with a tasklet
and a context and returns it so that it can be passed to ->subscribe() to
be added to a connection or conn_stream's wait_list. The caller doesn't
need to know all the insiders details anymore this way.
Totally nuke the "send" method, instead, the upper layer decides when it's
time to send data, and if it's not possible, uses the new subscribe() method
to be called when it can send data again.
