Some architectures like x86_64 and aarch64 support efficient unaligned
64-bit reads. On such architectures, we already know that each string
passed to field_start() has some margin at the end because it's parsed
using fgets2() which looks for the trailing LF using the same method.
Thus let's skip spaces by packs of 8. This increases the parsing speed
by 35%.
Modern compilers were producing producing less efficient code in the
field_start() loop, by not emitting two conditional jumps for a single
test. However by reordering the test we can merge the optimal case and
the default one and get back to good performance so let's simplify the
test. This improves the parsing speed by 5%.
The usage message was starting to have long lines, it's preferable that
it still fits well into a default 80-col display so that options are
easy to find. Also cut that into the 3 parts (input filter, modifier,
output format) for improved legibility.
This patch adds support for extracting captured header fields to halog. A field
can be extracted by passing the `-hdr <block>:<field>` output filter.
Both `<block>` and `<field>` are 1-indexed.
`<block>` refers to the index of the brace-delimited list of headers. If both
request and response headers are captured, then request headers are referenced
by `<block> = 1`, response headers are `2`. If only one direction is captured,
there will only be a single block `1`.
`<field>` refers to a single field within the selected block.
The output will contain one line, possibly empty, per log line processed.
Passing a non-existent `<block>` or `<field>` will result in an empty line.
Example:
capture request header a len 50
capture request header b len 50
capture request header c len 50
capture response header d len 50
capture response header e len 50
capture response header f len 50
`-srv 1:1` will extract request header `a`
`-srv 1:2` will extract request header `b`
`-srv 1:3` will extract request header `c`
`-srv 2:3` will extract response header `f`
This resolves GitHub issue #1146.
Our use-case for this is a dynamic application that performs routing based on
the query string. Without this option all URLs will just point to the central
entrypoint of this location, making the output completely useless.
Dmitry reported this warning on FreeBSD since the introduction of -Wundef:
admin/halog/fgets2.c:38:30: warning: '__GLIBC__' is not defined, evaluates to 0 [-Wundef]
#if defined(__x86_64__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 15))
^
A defined() was missing.
The ExecStartPre line was introduced a long time ago in the systemd unit
file, at the time of systemd wrapper. With the haproxy master worker
mode, this line is now useless, since starting haproxy itself will check
the configuration.
However this does not concern the check in the ExecReload which is still
needed to return a reload status to HAProxy.
It probably shouldn't be backported.
This line should disappear in a future version but we should still fix
ExecStartPre with -Ws like we've done in 9def142.
It's a complementary fix that must be backported with 9def142
("BUG/MINOR: systemd: must check the configuration using -Ws").
When doing a reload with a configuration which requires the
master-worker mode, the configuration check will fail because the check
is not done with -W/-Ws.
Example:
wla@kikyo:~/haproxy$ ./haproxy -Ws -c -f haproxy.cfg
Configuration file is valid
wla@kikyo:~/haproxy$ ./haproxy -c -f haproxy.cfg
[NOTICE] (13153) : haproxy version is 2.5-dev2-4567b3-16
[NOTICE] (13153) : path to executable is ./haproxy
[ALERT] (13153) : config : Can't use a 'program' section without master worker mode.
[ALERT] (13153) : config : Fatal errors found in configuration.
This patch fixes the issue by adding -Ws on the check command line.
Must be backported in all stable branches. (The file was previously in
contrib/systemd/haproxy.service.in).
This one was deprecated in 2.3 and marked for removal in 2.5. It suffers
too many limitations compared to threads, and prevents some improvements
from being engaged. Instead of a bypassable startup error, there is now
a hard error.
The parsing code was removed, and very few obvious cases were as well.
The code is deeply rooted at certain places (e.g. "for" loops iterating
from 0 to nbproc) so it will not be that trivial to remove everywhere.
The "bind" and "bind-process" parsers will have to be adjusted, though
maybe not completely changed if we later want to support thread groups
for large NUMA machines. Some stats socket restrictions were removed,
and the doc was updated according to what was done. A few places in the
doc still refer to nbproc and will have to be revisited. The master-worker
code also refers to the process number to distinguish between master and
workers and will have to be carefully adjusted. The MAX_PROCS macro was
reset to 1, this will at least reduce the size of some remaining arrays.
Two regtests were dependieng on this directive, one with an explicit
"nbproc 1" and another one testing the master's CLI using nbproc 4.
Both were adapted.
Commit ab0a5192a ("MEDIUM: config: mark "grace" as deprecated") marked
the "grace" keyword as deprecated in 2.3, tentative removal for 2.4
with a hard deadline in 2.5, so let's remove it and return an error now.
This old and outdated feature was incompatible with soft-stop, reload
and socket transfers, and keeping it forced ugly hacks in the lower
layers of the protocol stack.
Some of the Lua doc and a few places still used "Haproxy" or "HAproxy".
There was even one "HA proxy". A few of them were in an example of VTest
output, indicating that VTest ought to be fixed as well. No big deal but
better address all the remaining ones so that these inconsistencies stop
spreading around.
Both structures are identical except the name of the field starting
the period and its description. Let's call them all freq_ctr and the
period's start "curr_tick" which is generic.
This is only a temporary change and fields are expected to remain
the same with no code change (verified).
halog currently emits lots of warnings because it does not benefit from
the default flags. Let's update the main makefile to build it by itself
and remove the other one. The sub-project's makefile was replaced with
A readme indicating how to build it.
There has been a USE_MEMCHR option for ages that was mostly never enabled
because it was unclear when glibc became faster. A quick look at the code
indicates that this arrived with the SSE implementation of memchr() which
arrived at commit 093ecf92998de2 between 2.14 and 2.15, so let's automatically
turn this on on x86_64 with glibc >= 2.15.
This results in ~6GB of logs read per second (20 million lines) and ~2.5GB/s
(8 million lines) parsed for errors or status codes classification, or 1 GB/s
(3 million lines) for time percentiles.
The following components were moved to admin/ because they're generally
used in field by admins:
iprange/ netsnmp-perl/ selinux/ systemd/ wireshark-dissectors/
syntax-highlight/ release-estimator/
