Because shutowns for reads are now considered as aborts, the shudowns for
writes can now be considered as shutdowns. Here it is just a flag
renaming. SC_FL_SHUTW_NOW is renamed SC_FL_SHUT_WANTED.
The purpose of this patch is only a one-to-one replacement, as far as
possible.
CF_SHUTR(_NOW) and CF_SHUTW(_NOW) flags are now carried by the
stream-connecter. CF_ prefix is replaced by SC_FL_ one. Of course, it is not
so simple because at many places, we were testing if a channel was shut for
reads and writes in same time. To do the same, shut for reads must be tested
on one side on the SC and shut for writes on the other side on the opposite
SC. A special care was taken with process_stream(). flags of SCs must be
saved to be able to detect changes, just like for the channels.
It was done by hand by callers when a shutdown for read or write was
performed. It is now always handled by the functions performing the
shutdown. This way the callers don't take care of it. This will avoid some
bugs.
This patch removes CF_READ_ERROR and CF_WRITE_ERROR flags. We now rely on
SE_FL_ERR_PENDING and SE_FL_ERROR flags. SE_FL_ERR_PENDING is used for write
errors and SE_FL_ERROR for read or unrecoverable errors.
When a connection error is reported, SE_FL_ERROR and SE_FL_EOS are now set and a
read event and a write event are reported to be sure the stream will properly
process the error. At the stream-connector level, it is similar. When an error
is reported during a send, a write event is triggered. On the read side, nothing
more is performed because an error at this stage is enough to wake the stream
up.
A major change is brought with this patch. We stop to check flags of the
ooposite channel to report abort or timeout. It also means when an read or
write error is reported on a side, we no longer update the other side. Thus
a read error on the server side does no long lead to a write error on the
client side. This should ease errors report.
CF_READ_PARTIAL flag is now merged with CF_READ_EVENT. It means
CF_READ_EVENT is set when a read0 is received (formely CF_READ_NULL) or when
data are received (formely CF_READ_ACTIVITY).
There is nothing special here, except conditions to wake the stream up in
sc_notify(). Indeed, the test was a bit changed to reflect recent
change. read0 event is now formalized by (CF_READ_EVENT + CF_SHUTR).
As for CF_READ_NULL, it appears CF_WRITE_NULL and other write events on a
channel are mainly used to wake up the stream and may be replace by on write
event.
In this patch, we introduce CF_WRITE_EVENT flag as a replacement to
CF_WRITE_EVENT_NULL. There is no breaking change for now, it is just a
rename. Gradually, other write events will be merged with this one.
In order to evenly pick idle connections from other threads, there is
a "next_takeover" index in the server, that is incremented each time
a connection is picked from another thread, and indicates which one to
start from next time.
With thread groups this doesn't work well because the index is the same
regardless of the group, and if a group has more threads than another,
there's even a risk to reintroduce an imbalance.
This patch introduces a new per-tgroup storage in servers which, for now,
only contains an instance of this next_takeover index. This way each
thread will now only manipulate the index specific to its own group, and
the takeover will become fair again. More entries may come soon.
In github issue #1878, Bart Butler reported observing turn-around states
(1 second pause) after connection retries going to different servers,
while this ought not happen.
In fact it does happen because back_handle_st_cer() enforces the TAR
state for any algo that's not round-robin. This means that even leastconn
has it, as well as hashes after the number of servers changed.
Prior to doing that, the call to stream_choose_redispatch() has already
had a chance to perform the correct choice and to check the algo and
the number of retries left. So instead we should just let that function
deal with the algo when needed (and focus on deterministic ones), and
let the former just obey. Bart confirmed that the fixed version works
as expected (no more delays during retries).
This may be backported to older releases, though it doesn't seem very
important. At least Bart would like to have it in 2.4 so let's go there
for now after it has cooked a few weeks in 2.6.
Idle connections do not work on 32-bit machines due to an alignment issue
causing the connection nodes to be indexed with their lower 32-bits set to
zero and the higher 32 ones containing the 32 lower bitss of the hash. The
cause is the use of ebmb_node with an aligned data, as on this platform
ebmb_node is only 32-bit aligned, leaving a hole before the following hash
which is a uint64_t:
$ pahole -C conn_hash_node ./haproxy
struct conn_hash_node {
struct ebmb_node node; /* 0 20 */
/* XXX 4 bytes hole, try to pack */
int64_t hash; /* 24 8 */
struct connection * conn; /* 32 4 */
/* size: 40, cachelines: 1, members: 3 */
/* sum members: 32, holes: 1, sum holes: 4 */
/* padding: 4 */
/* last cacheline: 40 bytes */
};
Instead, eb64 nodes should be used when it comes to simply storing a
64-bit key, and that is what this patch does.
For backports, a variant consisting in simply marking the "hash" member
with a "packed" attribute on the struct also does the job (tested), and
might be preferable if the fix is difficult to adapt. Only 2.6 and 2.5
are affected by this.
The "first req" rule consists in not delivering a connection's first
request to a connection that's not known for being safe so that we
don't deliver a broken page to a client if the server didn't intend to
keep it alive. That's what's used by "http-reuse safe" particularly.
But the reason this rule was created was precisely because haproxy was
not able to re-emit the request to the server in case of connection
breakage, which is precisely what l7 retries later brought. As such,
there's no reason for enforcing this rule when l7 retries are properly
enabled because such a blank page will trigger a retry and will not be
delivered to the client.
This patch simply checks that the l7 retries are enabled for the 3 cases
that can be triggered on a dead or dying connection (failure, empty, and
timeout), and if all 3 are enabled, then regular idle connections can be
reused.
This could almost be marked as a bug fix because a lot of users relying
on l7 retries do not necessarily think about using http-reuse always due
to the recommendation against it in the doc, while the protection that
the safe mode offers is never used in that mode, and it forces the http
client not to reuse existing persistent connections since it never sets
the "not first" flag.
It could also be decided that the protection is not used either when
the origin is an applet, as in this case this is internal code that
we can decide to let handle the retry by itself (all info are still
present). But at least the httpclient will be happy with this alone.
It would make sense to backport this at least to 2.6 in order to let
the httpclient reuse connections, maybe to older releases if some
users report low reuse counts.
The connection retry counter is incremented too early when a connection
fails. In SC_ST_CER state, errors handling must be performed before
incrementing the counter. Otherwise, we may consider the max connection
attempt is reached while a last one is in fact possible.
This patch must be backported to 2.6.
If the loadbalancing is performed on the source IP address, an internal
error was returned on error. So for an applet on the client side (for
instance an SPOE applet) or for a client connected to a unix socket, an
internal error is returned.
However, when other LB algos fail, a fallback on round-robin is
performed. There is no reson to not do the same here.
This patch should fix the issue #1797. It must be backported to all
supported versions.
We don't want to pick idle connections from another thread group,
this would be very slow by forcing to share undesirable data.
This patch makes sure that we start seeking from the current thread
group's threads only and loops over that range exclusively.
It's worth noting that the next_takeover pointer remains per-server
and will bounce when multiple groups use it at the same time. But we
preserve the perturbation by applying a modulo when retrieving it,
so that when groups are of the same size (most common case), the
index will not even change. At this time it doesn't seem worth
storing one index per group in servers, but that might be an option
if any contention is detected later.
Function arguments and local variables called "cs" were renamed to "sc"
to avoid future confusion. The HTTP analyser and the backend functions
were all updated after being reviewed. Function stream_update_both_cs()
was renamed to stream_update_both_sc()
There's no more reason for keepin the code and definitions in conn_stream,
let's move all that to stconn. The alphabetical ordering of include files
was adjusted.
This file contains all the stream-connector functions that are specific
to application layers of type stream. So let's name it accordingly so
that it's easier to figure what's located there.
The alphabetical ordering of include files was preserved.
We're starting to propagate the stream connector's new name through the
API. Most call places of these functions that retrieve the channel or its
buffer are in applets. The local variable names are not changed in order
to keep the changes small and reviewable. There were ~92 uses of cs_ic(),
~96 of cs_oc() (due to co_get*() being less factorizable than ci_put*),
and ~5 accesses to the buffer itself.
This also follows the natural naming. There are roughly 238 changes, all
totally trivial. conn_stream-t.h has become completely void of any
"conn_stream" related stuff now (except its name).
This renames the "struct conn_stream" to "struct stconn" and updates
the descriptions in all comments (and the rare help descriptions) to
"stream connector" or "connector". This touches a lot of files but
the change is minimal. The local variables were not even renamed, so
there's still a lot of "cs" everywhere.
Just like for the appctx, this is a pointer to a stream endpoint descriptor,
so let's make this explicit and not confuse it with the full endpoint. There
are very few changes thanks to the preliminary refactoring of the flags
manipulation.
This changes all main uses of cs->endp->flags to the sc_ep_*() equivalent
by applying coccinelle script cs_endp_flags.cocci.
Note: 143 locations were touched, manually reviewed and found to be OK,
except a single one that was adjusted in cs_reset_endp() where the flags
are read and filtered to be used as-is and not as a boolean, hence was
replaced with sc_ep_get() & $FLAGS.
The script was applied with all includes:
spatch --in-place --recursive-includes -I include --sp-file $script $files
The mux ->detach() function currently takes a conn_stream. This causes
an awkward situation where the caller cs_detach_endp() has to partially
mark it as released but not completely so that ->detach() finds its
endpoint and context, and it cannot be done later since it's possible
that ->detach() deletes the endpoint. As such the endpoint link between
the conn_stream and the mux's stream is in a transient situation while
we'd like it to be clean so that the mux's ->detach() code can call any
regular function it wants that knows the regular semantics of the
relation between the CS and the endpoint.
A better approach consists in slightly modifying the detach() API to
better match the reality, which is that the endpoint is detached but
still alive and that it's the only part the function is interested in.
As such, this patch modifies the function to take an endpoint there,
and by analogy (or simplicity) does the same for ->attach(), even
though it looks less important there since we're always attaching an
endpoint to a conn_stream anyway. It is possible that in the future
the API could evolve to use more endpoints that provide a bit more
flexibility in the API, but at this point we don't need to go further.
This flag is no longer needed now that it must always match the presence
of a destination address on the backend conn_stream. Worse, before previous
patch, if it were to be accidently removed while the address is present, it
could result in a leak of that address since alloc_dst_address() would first
be called to flush it.
Its usage has a long history where addresses were stored in an area shared
with the connection, but as this is no longer the case, there's no reason
for putting this burden onto application-level code that should not focus
on setting obscure flags.
The only place where that made a small difference is in the dequeuing code
in case of queue redistribution, because previously the code would first
clear the flag, and only later when trying to deal with the queue, would
release the address. It's not even certain whether there would exist a
code path going to connect_server() without calling pendconn_dequeue()
first (e.g. retries on queue timeout maybe?).
Now the pendconn_dequeue() code will rely on SF_ASSIGNED to decide to
clear and release the address, since that flag is always set while in
a server's queue, and its clearance implies that we don't want to keep
the address. At least it remains consistent and there's no more risk of
leaking it.
These functions dynamically allocate a source or destination address but
start by clearing the previous one. There's a non-null risk of leaking
addresses there in case of misuse. Better have them do nothing if the
address was already allocated.
Only CS_EP_ERROR flag is now removed from the endpoint when a reset is
performed. When a new the endpoint is allocated, flags are preserved. It is
the caller responsibility to remove other flags, depending on its need.
Concretly, during a connection retry or a L7 retry, we must preserve
flags. In tcpcheck and the CLI, we reset flags.
This patch is 2.6-specific. No backport needed.
There were plenty of leftovers from old code that were never removed
and that are not needed at all since these files do not use any
definition depending on fcntl.h, let's drop them.
Almost all of our hash-based LB algorithms are implemented as special
cases of something that can now be achieved using sample expressions,
and some of them have adopted some options to adapt their behavior in
ways that could also be achieved using converters.
There are users who want to hash other parameters that are combined
into variables, and who set headers from these values and use
"balance hdr(name)" for this.
Instead of constantly implementing specific options and having users
hack around when they want a real hash, let's implement a native hash
mode that applies to a standard sample expression. This way, any
fetchable element (including variables) may be used to construct the
hash, even modified by any converter if desired.
Since the idle connections management changed to use eb-trees instead of MT
lists, a lock must be acquired to manipulate servers idle/safe/available
connection lists. However, it remains an unprotected use in
connect_server(), when a connection is removed from an idle list if the mux
has no more streams available. Thus it is possible to remove a connection
from an idle list on a thread, while another one is looking for a idle
connection. Of couse, this may lead to a crash.
To fix the bug, we must take care to acquire the idle connections lock
first. The bug was introduced by the commit f232cb3e9 ("MEDIUM: connection:
replace idle conn lists by eb trees").
The patch must be backported as far as 2.4.
When an appctx is created on the server side, we now set the corresponding
conn-stream to ready state (CS_ST_RDY). When it happens, the backend
conn-stream is in CS_ST_INI state. It is not consistant to let the
conn-stream in this state because it means it is possible to have a target
installed in CS_ST_INI state, while with a connection, the conn-stream is
switch to CS_ST_RDY or CS_ST_EST state.
It is especially anbiguous because we may be tempted to think there is no
endpoint attached to the conn-stream before the CS_ST_CON state. And it is
indeed the reason for a bug leading to a crash because a cs_detach_endp() is
performed if an abort is detected on the backend conn-stream in CS_ST_INI
state. With a mux or a appctx attached to the conn-stream, "->endp" field is
set to NULL. It is unexpected. The API will be changed to be sure it is not
possible. But it exposes a consistency issue with applets.
So, the conn-stream must not stay in CS_ST_INI state when an appctx is
attached. But there is no reason to set it in CS_ST_REQ. The conn-stream
must be set to CS_ST_RDY to handle applets and connections in the same
way. Note that if only the target is set but no appctx is created, the
backend conn-stream is switched from CS_ST_INI to CS_ST_REQ state to be able
to create the corresponding appctx. This part is unchanged.
This patch depends on the commit "MINOR: backend: Don't allow to change
backend applet".
The ambiguity exists on previous versions. But the issue is
2.6-specific. Thus, no backport is needed.
This part was inherited from haproxy-1.5. But since a while (at least 1.8),
the backend applet, once created, is no longer changed. Thus there is no
reason to still check if the target has changed. And in fact, if it was
still possible, there would be a memory leak because the old applet would be
lost and never released.
There is no reason to backport this fix because the leak only exists on a
dead code path.
si_register_applet() and si_applet_release() are renamed
cs_register_applet() and cs_applet_release() and now manipulate a
conn-stream instead of a stream-inteface.
si_shutr(), si_shutw(), si_chk_rcv() and si_chk_snd() are moved in the
conn-stream scope and renamed, respectively, cs_shutr(), cs_shutw(),
cs_chk_rcv(), cs_chk_snd() and manipulate a conn-stream instead of a
stream-interface.
