Hi,
HAProxy 2.7.2 was released on 2023/01/20. It added 91 new commits
after version 2.7.1.
Lots of small issues were fixed, a great part in QUIC, and a few stability
issues were addressed as well, some specific to 2.7 and others older.
* H3/QUIC:
- a double-delete could happen in a list, causing memory corruption and
crashes.
- too much memory was allocated to listeners bound only to certain
threads (one buffer per thread instead of one buffer per bound
thread). Given that we're encouraging users to use "shards by-thread",
this would cause N^2 buffers allocation for N threads, and taking half
a gigabyte per listener for 48 threads.
- the mapping between connection ID and threads was incorrect when
shards were enabled, possibly resulting in an imbalance of the load.
- the QUICv2 protocol version has been updated in the final version of
the spec, so it was updated.
- there is now a global "no-quic" option availble to globally disable
QUIC listeners (they continue to be bound but do not process traffic),
and a "quic_enabled" sample-fetch function that is convenient to
condition Alt-Svc advertisement. The purpose is to make it easier to
globally disable QUIC in a config without having to comment plenty of
lines, should a serious problem happen in the future.
- among a connection, the traffic distribution between streams was
unfair, and the first streams used to be able to take as much of the
window as they wanted, resulting in some streams progressing much
slower than others. This was addressed.
- H3 did not yet implement trailers support, which could possibly be
annoying in some environments, this was now implemented.
- small improvements such as detection of shutdown streams with empty
response, handling of RESET_STREAM and emission of STOP_SENDING,
ability to only abort streams on H3 errors, small fixes on flow
control, reject forbidden headers, explicitly advertise that we don't
perform active connection migration.
* H2:
- a bug was introduced in 2.7, where responses of unknown lengths will
cause a CANCEL to be emitted instead of an end-of-stream.
- an interim response that carries the end-of-stream flag was not
rejected as invalid. The consequences are uncertain in 2.4 and newer,
but on 2.2 and older it could trigger a BUG_ON() condition and cause a
panic.
- logs were not emitted for invalid requests that are blocked due to
forbidden headers or syntaxes. That made it complicated to debug errors
reported by clients. They will now be emitted, and traces will also
reflect this.
* HTTP rules:
- make sure that the logged status matches the reported status even upon
errors and also after http-after-response
- there was a small leak per request when using the "originalto" option,
and another leak (per config entry) for "http-request redirect" lines.
- there was a parsing error reported for responses carrying a websocket
header when the status was not 101.
* listeners:
- multiple races were found and addressed related to closed FDs (mostly
happening on reload, sometimes on resuming after an aborted reload)
- some assertions could be triggered at boot if using shards with inherited
FDs ("bind fd@xxx").
* other:
- a fix for a buffer realignment bug introduced in 1.9 was fixed. It's
uncertain whether it was possible to trigger it or not, but it could
possibly have been responsible for some rare unexplained corruptions.
- JWT ECDSA signatures were not properly handled, this was now fixed.
However, another issue was just discovered after the release that may
still randomly trigger errors.
- fcgi: the path-info subexpression was not properly handled due to an
inverted condition.
- config: the patch adding environment variables HAPROXY_TCP_LOG_FMT,
HAPROXY_HTTP_LOG_FMT and HAPROXY_HTTPS_LOG_FMT which contain the
default log format strings was backported because it simplifies some
configurations by reducing the amount of copy-pasting.
- debug: thread dumps woudln't show backtraces anymore, this was fixed.
- sometimes there could be races between "show threads" and the delivery
of the reload signal, causing an older process to spin at 100% forever.
- bwlim: some checks were insufficient on the set-bandwidth-limit
arguments.
- pools: severe thread contention could be caused by false sharing of a
same cache line between different CPUs due to usage counters and
limits being very close in the structure. This was emphasized by QUIC
which uses pools more often due to the usage of small packets, where
just adding a cache-line alignment resulted in a 130% performance gain
on a 48 threads machine.
- makefile: minor build updates (sorting of enabled features,
fix build of wolfSSL when SSL_INC is set, ...)
- various CI, regtest, error messages, and doc updates
I hope that I didn't miss anything!
Please find the usual URLs below :
Site index : https://www.haproxy.org/
Documentation : https://docs.haproxy.org/
Wiki : https://github.com/haproxy/wiki/wiki
Discourse : https://discourse.haproxy.org/
Slack channel : https://slack.haproxy.org/
Issue tracker : https://github.com/haproxy/haproxy/issues
Sources : https://www.haproxy.org/download/2.7/src/
Git repository : https://git.haproxy.org/git/haproxy-2.7.git/
Git Web browsing : https://git.haproxy.org/?p=haproxy-2.7.git
Changelog : https://www.haproxy.org/download/2.7/src/CHANGELOG
Dataplane API :
https://github.com/haproxytech/dataplaneapi/releases/latest
Pending bugs : https://www.haproxy.org/l/pending-bugs
Reviewed bugs : https://www.haproxy.org/l/reviewed-bugs
Code reports : https://www.haproxy.org/l/code-reports
Latest builds : https://www.haproxy.org/l/dev-packages
Willy
---
Complete changelog :
Amaury Denoyelle (17):
BUG/MEDIUM: mux-quic: fix double delete from qcc.opening_list
MEDIUM: mux-quic: implement shutw
MINOR: mux-quic: do not count stream flow-control if already closed
MINOR: mux-quic: handle RESET_STREAM reception
MEDIUM: mux-quic: implement STOP_SENDING emission
MINOR: h3: use stream error when needed instead of connection
BUG/MINOR: mux-quic: ignore remote unidirectional stream close
BUG/MINOR: mux-quic: fix transfer of empty HTTP response
MINOR: mux-quic: add traces for flow-control limit reach
MAJOR: mux-quic: rework stream sending priorization
MEDIUM: h3: send SETTINGS before STREAM frames
MINOR: mux-quic: use send-list for STOP_SENDING/RESET_STREAM emission
MINOR: mux-quic: use send-list for immediate sending retry
BUG/MINOR: h3: properly handle connection headers
MINOR: h3: extend function for QUIC varint encoding
MINOR: h3: implement TRAILERS encoding
MINOR: h3: implement TRAILERS decoding
Aurelien DARRAGON (5):
BUG/MINOR: proxy: free orgto_hdr_name in free_proxy()
REGTEST: fix the race conditions in json_query.vtc
REGTEST: fix the race conditions in add_item.vtc
REGTEST: fix the race conditions in digest.vtc
REGTEST: fix the race conditions in hmac.vtc
Christopher Faulet (15):
BUG/MEDIUM: mux-h2: Refuse interim responses with end-stream flag set
BUG/MINOR: pool/stats: Use ullong to report total pool usage in bytes in
stats
BUG/MINOR: http-fetch: Only fill txn status during prefetch if not
already set
BUG/MAJOR: buf: Fix copy of wrapping output data when a buffer is
realigned
DOC: config: fix alphabetical ordering of http-after-response rules
DOC: config: remove duplicated "http-response sc-set-gpt0" directive
BUG/MINOR: h1-htx: Remove flags about protocol upgrade on non-101
responses
BUG/MINOR: hlua: Fix Channel.line and Channel.data behavior regarding the
doc
BUG/MINOR: resolvers: Wait the resolution execution for a do_resolv action
BUG/MINOR: promex: Don't forget to consume the request on error
BUG/MINOR: http-ana: Report SF_FINST_R flag on error waiting the request
body
BUG/MINOR: http-fetch: Don't block HTTP sample fetch eval in
HTTP_MSG_ERROR state
BUG/MEDIUM: mux-h2: Don't send CANCEL on shutw when response length is
unkown
BUG/MINOR: bwlim: Check scope for period expr for set-bandwitdh-limit
actions
BUG/MINOR: bwlim: Fix parameters check for set-bandwidth-limit actions
Daniel Corbett (1):
DOC: config: fix "Address formats" chapter syntax
Frédéric Lécaille (5):
MINOR: quic: Useless test about datagram destination addresses
MINOR: quic: Disable the active connection migrations
MINOR: quic: Add "no-quic" global option
MINOR: sample: Add "quic_enabled" sample fetch
MINOR: quic: Replace v2 draft definitions by those of the final 2 version
Ilya Shipitsin (1):
CI: github: enable github api authentication for OpenSSL tags read
Manu Nicolas (1):
CLEANUP: htx: fix a typo in an error message of http_str_to_htx
Mathias Weiersmueller (1):
DOC: config: added optional rst-ttl argument to silent-drop in action
lists
Paul Barnetta (1):
BUG/MINOR: mux-fcgi: Correctly set pathinfo
Remi Tricot-Le Breton (3):
BUG/MINOR: http: Memory leak of http redirect rules' format string
BUG/MEDIUM: jwt: Properly process ecdsa signatures (concatenated R and S
params)
BUG/MINOR: ssl: Fix compilation with OpenSSL 1.0.2 (missing
ECDSA_SIG_set0)
Sébastien Gross (1):
MINOR: config: add environment variables for default log format
Tim Duesterhus (6):
CI: Improve headline in matrix.py
CI: Add in-memory cache for the latest OpenSSL/LibreSSL
CI: Use proper `if` blocks instead of conditional expressions in matrix.py
CI: Unify the `GITHUB_TOKEN` name across matrix.py and vtest.yml
CI: Explicitly check environment variable against `None` in matrix.py
CI: Reformat `matrix.py` using `black`
William Lallemand (5):
REGTESTS: ssl: enable the ssl_reuse.vtc test for WolfSSL
MINOR: httpclient: don't add body when istlen is empty
CI: github: use the GITHUB_TOKEN instead of a manually generated token
DOC: management: add details on "Used" status
DOC: management: add details about @system-ca in "show ssl ca-file"
Willy Tarreau (29):
OPTIM: pool: split the read_mostly from read_write parts in pool_head
BUG/MEDIUM: quic: properly take shards into account on bind lines
BUG/MINOR: quic: do not allocate more rxbufs than necessary
BUILD: makefile: build the features list dynamically
BUILD: makefile: sort the features list
BUILD: makefile: clean the wolfssl include and lib generation rules
BUILD: makefile: make sure to also ignore SSL_INC when using wolfssl
BUG/MINOR: debug: don't mask the TH_FL_STUCK flag before dumping threads
MINOR: cfgparse-ssl: avoid a possible crash on OOM in ssl_bind_parse_npn()
BUG/MINOR: stick-table: report the correct action name in error message
BUG/MINOR: fd: avoid bad tgid assertion in fd_delete() from deinit()
DEV: tcploop: add minimal support for unix sockets
BUG/MEDIUM: peers: make "show peers" more careful about partial
initialization
BUG/MINOR: http-ana: make set-status also update txn->status
BUG/MINOR: listeners: fix suspend/resume of inherited FDs
DOC: config: fix wrong section number for "protocol prefixes"
DOC: config: fix aliases for protocol prefixes "udp4@" and "udp6@"
DOC: config: mention the missing "quic4@" and "quic6@" in protocol
prefixes
MINOR: listener: also support "quic+" as an address prefix
CLEANUP: stconn: always use se_fl_set_error() to set the pending error
BUG/MEDIUM: stconn: also consider SE_FL_EOI to switch to SE_FL_ERROR
BUG/MINOR: listener: close tiny race between resume_listener() and
stopping
BUG/MEDIUM: fd/threads: fix again incorrect thread selection in wakeup
broadcast
BUG/MINOR: thread: always reload threads_enabled in loops
MINOR: threads: add a thread_harmless_end() version that doesn't wait
BUG/MEDIUM: debug/thread: make the debug handler not wait for
!rdv_requests
BUG/MINOR: mux-h2: make sure to produce a log on invalid requests
BUG/MINOR: mux-h2: add missing traces on failed headers decoding
BUILD: hpack: include global.h for the trash that is needed in debug mode
---
