By default H2 uses a 65535 bytes window for the connection, and changing
it requires sending a WINDOW_UPDATE message. We only used to update the
window when receiving data, thus never increasing it further.
As reported by user klzgrad on the mailing list, this seriously limits
the upload bitrate, and will have an even higher impact on the backend
H2 connections to origin servers.
There is no technical reason for keeping this window so low, so let's
increase it to the maximum possible value (2G-1). We do this by
pretending we've already received that many data minus the maximum
data the client might already send (65535), so that an early
WINDOW_UPDATE message is sent right after the SETTINGS frame.
This should be backported to 1.8. This patch depends on previous
patch "BUG/MINOR: mux-h2: refrain from muxing during the preface".
The condition to refrain from processing the mux was insufficient as it
would only handle the outgoing connections. In essence it is not that much
of a problem since we don't have streams yet on an incoming connetion. But
it prevents waiting for the end of the preface before sending an early
WINDOW_UPDATE message, thus causing the connections to fail in this case.
This must be backported to 1.8 with a few minor adaptations.
Since HTX casts the buffer to a struct and stores relative pointers at the
end, it is mandatory that its end is properly aligned. This patch enforces
a buffer size rounding up to the next multiple of two void*, thus 8 on
32-bit and 16 on 64-bit, to match what malloc() already does on the beginning
of the buffer. In pratice it will never be really noticeable since default
sizes already are such multiples.
When the mux's buffer is empty and the htx area contains exactly one
data block of the same size as the requested count, then it's possible
to simply swap the caller's buffer with the mux's output buffer and
adjust offsets and length to match the entire DATA HTX block in the
middle. In this case we perform a true zero-copy operation from
end-to-end. This is the situation that happens all the time with large
files. With this change, the HTX bit rate performance catches up again
with the legacy mode (measured at 97%).
These flags haven't been used for a while. SF_TUNNEL was reintroduced
by commit d62b98c6e ("MINOR: stream: don't set backend's nor response
analysers on SF_TUNNEL") to handle the two-level streams needed to
deal with the first model for H2, and was not removed after this model
was abandonned. SF_INITIALIZED was only set. SF_CONN_TAR was never
referenced at all.
Now that h1 and legacy HTTP are two distinct things, there's no need
to keep the legacy HTTP parsers in h1.c since they're only used by
the legacy code in proto_http.c, and h1.h doesn't need to include
hdr_idx anymore. This concerns the following functions :
- http_parse_reqline();
- http_parse_stsline();
- http_msg_analyzer();
- http_forward_trailers();
All of these were moved to http_msg.c.
Lots of HTTP code still uses struct http_msg. Not only this code is
still huge, but it's part of the legacy interface. Let's move most
of these functions to a separate file http_msg.c to make it more
visible which file relies on what. It's mostly symmetrical with
what is present in http_htx.c.
The function http_transform_header_str() which used to rely on two
function pointers to look up a header was simplified to rely on
two variants http_legacy_replace_{,full_}header(), making both
sides of the function much simpler.
No code was changed beyond these moves.
All the HTX definition is self-contained and doesn't really depend on
anything external since it's a mostly protocol. In addition, some
external similar files (like h2) also placed in common used to rely
on it, making it a bit awkward.
This patch moves the two htx.h files into a single self-contained one.
The historical dependency on sample.h could be also removed since it
used to be there only for http_meth_t which is now in http.h.
As for the compression filter, the cache filter must be explicitly declared
(using the filter keyword) if other filters than cache are used. It is mandatory
to explicitly define the filters order.
Documentation has been updated accordingly.
This is only true for HTX proxies. On legacy HTTP proxy, if the compression and
the cache are both enabled, an error during HAProxy startup is triggered.
With the HTX, now you can use both in any order. If the compression is defined
before the cache, then the responses will be stored compressed. If the
compression is defined after the cache, then the responses will be stored
uncompressed. So in the last case, when a response is served from the cache, it
will compressed too like any response.
To do so, a dedicated configuration has been added on cache filters. Before the
cache filter configuration pointed directly to the cache it used. Now, it is the
dedicated structure cache_flt_conf. Store and use rules also point to this
structure. It is linked to the cache the filter must used. It also contains a
flags field. This will allow us to define the behavior of a cache filter when a
response is stored in the cache or delivered from it.
And now, Store and use rules uses a common parsing function. So if it does not
already exists, a filter is always created for both kind of rules. The cache
filters configuration is checked using their check callback. In the postparser
function, we only check the caches configuration. This removes the loop on all
proxies in the postparser function.
The cache is now able to store and resend HTX messages. When an HTX message is
stored in the cache, the headers are prefixed with their block's info (an
uint32_t), containing its type and its length. Data, on their side, are stored
without any prefix. Only the value is copied in the cache. 2 fields have been
added in the structure cache_entry, hdrs_len and data_len, to known the size, in
the cache, of the headers part and the data part. If the message is chunked, the
trailers are also copied, the same way as data. When the HTX message is
recreated in the cache applet, the trailers size is known removing the headers
length and the data lenght from the total object length.
Instead of calling register_data_filter() when the stream analyze starts, we now
call it when we are sure the response is cacheable. It is done in the
http_headers callback, just before the body analyzis, and only if the headers
was already been cached. And during the body analyzis, if an error occurred or
if the response is too big, we unregistered the cache immediatly.
This patch may be backported in 1.8. It is not a bug but a significant
improvement.
It is not possible to mix the format of messages stored in a cache. So we reject
the configurations with a cache used by an HTX proxy and a legacy HTTP proxy in
same time.
The CLI proxy was not handling payload. To do that, we needed to keep a
connection active on a server and to transfer each new line over that
connection until we receive a empty line.
The CLI proxy handles the payload in the same way that the CLI do it.
Examples:
$ echo -e "@1;add map #-1 <<\n$(cat data)\n" | socat /tmp/master-socket -
$ socat /tmp/master-socket readline
prompt
master> @1
25130> add map #-1 <<
+ test test
+ test2 test2
+ test3 test3
+
25130>
During a payload transfer, we need to wait for the data even when we are
not in interactive mode. Indeed, the data could be received line per
line progressively instead of in one recv.
Previously the CLI was doing a SHUTW just after the first line if it was
not in interactive mode. We now check if we are in payload mode to do
a SHUTW.
Should be backported in 1.8.
Rework the CLI proxy parser to look more like the CLI parser, corner
case and escaping are handled the same way.
The parser now splits the commands in words instead of just handling
the prefixes.
It's easier to compare words and arguments of a command this way and to
parse internal command that will be consumed directly by the CLI proxy.
There were a number of ugly setsockopt() calls spread all over
proto_http.c, proto_htx.c and hlua.c just to manipulate the front
connection's TOS, mark or TCP quick-ack. These ones entirely relied
on the connection, its existence, its control layer's presence, and
its addresses. Worse, inet_set_tos() was placed in proto_http.c,
exported and used from the two other ones, surrounded in #ifdefs.
This patch moves this code to connection.h and makes the other ones
rely on it without ifdefs.
The new function hpack_encode_path() supports encoding a path into
the ":path" header. It knows about "/" and "/index.html" which use
a single byte, and falls back to literal encoding for other ones,
with a fast path for short paths < 127 bytes.
The new function hpack_encode_scheme() supports encoding a scheme
into the ":scheme" header. It knows about "https" and "http" which use
a single byte, and falls back to literal encoding for other ones.
The new function hpack_encode_method() supports encoding a method.
It knows about GET and POST which use a single byte, and falls back
to literal encoding for other ones.
This way we don't open-code the HPACK status codes anymore in the H2
code. Special care was taken not to cause any slowdown as this code is
very sensitive.
This header exists with 7 different values, it's worth taking them
into account for the encoding, hence these functions. One of them
makes use of an integer only and computes the 3 output bytes in case
of literal. The other one benefits from the knowledge of an existing
string, which for example exists in the case of H1 to H2 encoding.
For long header values whose index is known, hpack_encodde_long_idx()
may now be used. This function emits the short index and follows with
the header's value.
Most direct calls to HPACK functions are made to encode short header
fields like methods, schemes or statuses, whose lengths and indexes
are known. Let's have a small function to do this.
We'll need these functions from other inline functions, let's make them
accessible. len_to_bytes() was renamed to hpack_len_to_bytes() since it's
now exposed.
We used to have a series of well-known header fields that were looked
up, but most of them were not. The current model couldn't scale with
the addition of the new headers or pseudo-headers required to process
requests, resulting in their encoding being hard-coded in the caller.
This patch implements a quick lookup which retrieves any header from
the static table. A binary stream is made of header names prefixed by
lengths and indexes. These header names are sorted by length, then by
frequency, then by direction (preference for response), then by name,
the the lowest index of each is stored only in case of multiple
entries. A parallel length index table provides the index of the first
header for a given string. This allows to focus on the first few values
matching the same length.
Everything was made to limit the cache footprint. Interestingly, the
lookup ends up being slightly faster than the previous one, while
covering the 54 distinct headers instead of only 10.
A test with a curl request and a basic response showed that the request
size has dropped from 85 to 56 bytes and that the response size has
dropped from 197 to 170 bytes, thus we can now shave roughly 25-30 bytes
per message.
This generates the tables and indexes which will be used by the HPACK
encoder. The headers are sorted by length, then by statistical frequency,
then by direction (preference for responses), then by name, then by index.
The purpose is to speed up their lookup.
For unknown fields, since we know that most of them are less than 127
characters, we don't need to go through the loop and can instead directly
emit the one-byte length encoding. This increases the request rate by
approximately 0.5%.
memcpy() tends to be overkill to copy short strings, better use ist's
naive functions for this. This shows a consistent 1.2% performance
gain with h2load.
The len-to-bytes conversion can be slightly simplified and optimized
by hardcoding a tree lookup. Just doing this increases by 1% the
request rate on H2. It could be made almost branch-free by using
fls() but it looks overkill for most situations since most headers
are very short.
In hpack_encode_header() there is a length check to verify that a literal
header name fits in the buffer, but there it an off-by-one in this length
check, which forgets the byte required to mark the encoding type (literal
without indexing). It should be harmless though as it cannot be triggered
since response headers passing through haproxy are limited by the reserve,
which is not the case of the output buffer.
This fix should be backported to 1.8.
Otherwise, after such replaces, the HTX message appears to wrap but the head
block address is not necessarily the first one. So adding new blocks will
override data of old ones.
If a server sends part of headers and then close its connection, the mux H1
reamins blocked in an infinite loop trying to read more data to finish the
parsing of the message. The flag CS_FL_REOS is set on the conn_stream. But
because there are some data in the input buffer, CS_FL_EOS is never set.
To fix the bug, in h1_process_input, when CS_FL_REOS is set on the conn_stream,
we also set CS_FL_EOS if the input buffer is empty OR if the channel's buffer is
empty.
When a request is fully processed, no more data are parsed until the response is
totally processed and a new transaction starts. But during this time, the mux is
trying to read more data and subscribes to read. If requests are pipelined, we
start to receive the next requests which will stay in the input buffer, leading
to a loop consuming all the CPU. This loop ends when the transaction ends. To
avoid this loop, the flag H1C_F_IN_BUSY has been added. It is set when the
request is fully parsed and unset when the transaction ends. Once set on H1C, it
blocks the reads. So the mux never tries to receive more data in this state.
Condition to process the connection mode on outgoing messages whithout
'Connection' header was wrong. It relied on the wrong H1M
state. H1_MSG_HDR_L2_LWS is only a possible state for messages with at least one
header. Now, to fix the bug, we just check the H1M state is not
H1_MSG_LAST_LF. So, we have the warranty the EOH was not processed yet.
