Hi,
HAProxy 2.7-dev9 was released on 2022/11/18. It added 168 new commits
after version 2.7-dev8.
I know, one month has elapsed since -dev8, but some of us were fully
busy preparing the haproxyconf and others dealing alone with the bug
reports, so now we have to catch up slowly. On the other hand, these
efforts were rewarded since many bugs were addressed and the feedback
on the conf was generally excellent.
No less than 41 bugs were addressed since -dev8, that's about 2 per work
day, and most of them are now of low importance so we're converging.
However I find that there's still an annoying number of small ones in the
issue tracker, which are not necessarily regressions but which deserve
being analysed a bit, and this takes time.
I'll try to sum up one month of changes below by mostly focusing on
user-visible ones (including developers):
- USE_SHM_OPEN that is used to preserve startup logs across reloads
operated by the master process was extended by default to the
linux-musl and freebsd targets.
- tree-wide update to the "fallthrough" hint to tell the compiler to
shut up on missing breaks between switch/case statements: those who
were dealing with preprocessing followed by compiling with compilers
starting with gcc 7 were still seeing lots of warnings due to the
preprocessor stripping the comments and the compiler not seeing them.
This typically happened for those using distcc or ccache (could
possibly trigger for some distros using build farms).
- xxhash was updated to version 0.8.1 to address the similar fallthrough
issue as above (very few other differences so that's OK).
- fixed the build issue that was breaking DEBUG_MEM_STATS on old
compilers; it was my fault for putting quotes around symbol names
where it was inappropriate.
- fixed build issues for dev/poll and dev/tcploop which were ignoring
changes to the source file, and mis-reporting certain build steps.
- various QUIC build fixes on m68k, 32-bit platforms in general, and
worked around a wrong warning from gcc-12. Usual number of updates
including more counters and code cleanups/refactoring. By the way,
we'd like to thank Gabriel Tzagkarakis for his amazing help in fixing
the remaining QUIC bugs over the last month.
- the support for completely obsolete and abandonned set-cookie2 was
finally removed (thanks Tim for the reminder). It was planned for 2.5
already but we forgot it and it was not suitable to break the 2.5->2.6
transition.
- mux-h1's connection handling was refined so that the stream code depends
less on the low-level connection status. This is part of the permanent
quest to report the most accurate errors possible and not to report any
when everything went fine. Mux-h2 experienced a similar but smaller
cleanup (it's trickier and we don't want to risk regressions now).
- SSL certiticate error checking was ignoring decoding errors and was
fixed. There is a non-nul risk that errors are now reported at boot
time from broken certs that would already not work anyway but would
have been silently ignored.
- ca-ignore-err and crt-ignore-err can now use an error contant name
and not just a value (since openssl found it fun to change the
documented values between 1.x and 3.0). In addition, such options
were previously limited to the 64 first values, this is no longer
the case.
- a new "trace" statement can be placed in the global section, using
the exact same syntax as on the CLI. It requires that the directive
"expose-experimental-directives" is set first because we're not certain
it will remain like this, but it significantly improves operations for
users willing to share traces with developers during debugging sessions,
so we figured that any format would be better than nothing.
- a "quick-exit" keyword was added to the global section so that those
who absolutely need to skip deinit() on quick reload can do it. The
time savings are only marginal, however it could have been a useful
workaround a few times in the past when triggering a double-free in
deinit().
- the "debug dev memstats" output now also displays where buffers are
allocated and released. This should help track leaks when they happen.
- the "peers" section now supports a principle of sharding. The idea is
that when you have to push lots of updates to some aggregating nodes,
the amount of traffic can quickly become a limiting factor. With this
it becomes possible to use different targets for different keys. The
keys are hashed with the tables' names and only the relevant peers get
them.
- the stderr logs could exhibit the prefix "CLI" after an "add server"
command had been issued on the same thread. Now it will always be
properly cleared. However some users who might have been observing
the random faulty behavior could notice a difference.
- SSL error reporting upon startup was made clearer; previously, some
errors could appear ambigous when loading certs and CAs.
- "nbthread", "thread-group" and "thread-groups" will now refuse to be
changed after any section other than "global". We've indeed identified
a number of nasty chicken-and-egg situations where the risk that the
thread count or the thread-to-group mapping changes after a "bind" line
(or possibly other statements) makes certain things very difficult to
deal with. One of them was the need to support "bind" lines spreading
over multiple groups, as well as automatic shards. There's very little
chance anyone has fun adding an extra "global" section after other
ones just to change "nbthread", but just in case we'd rather fire an
error now than during the 2.7->2.8 transition.
- CI improvements such as 32-bit builds and libreSSL updates.
- lots of code cleanups and some flag renames
- various small doc updates
As you can see there's nothing really outstanding, lots of small stuff
that had to be finished and that takes time. On the todo list of things
that were not yet added to the issue tracker, I can cite:
- need to make the servers' idle conns counter per-thread group. I
noticed already that when using thread groups, we were killing
connections far too aggressively since they can't be shared between
groups and we tend to end up with slightly more than the estimated
limit. I finally figured how to address this but that's yet another
day of work so it will be in next one.
- I found that "timeout check" fails to extend the check delay when
running with short checks. I'm pretty sure it used to work and don't
know when it broke. This will not hold 2.7 though.
The other issues I'm seeing in the tracker are not 2.7-specific for now
but I still want to think about some of them with 2.7 in mind to be sure
not to face post-release trouble.
Hopefully in the second half of next week we can emit a -dev10 if some
stuff requires it to ease testing, and if everything goes well it looks
reasonable to aim for a release in the middle of the week after to stay
on end of month. There's no strict schedule as usual, and I'm fine with
slipping a little bit but only if we have good reasons (i.e. not "I
found this patch in my tree" nor "I haven't had time to test"). This
means that if you'd like to run some tests on your favorite OS, or if
you need to boot the old dusty Solaris, OpenBSD or AIX machine that
sleeps in the basement and takes one hour to build, now is the best
moment for it.
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.git/
Git Web browsing : https://git.haproxy.org/?p=haproxy.git
Changelog : https://www.haproxy.org/download/2.7/src/CHANGELOG
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 (24):
BUILD: ssl_sock: fix null dereference for QUIC build
BUG/MINOR: quic: fix buffer overflow on retry token generation
MINOR: quic: add version field on quic_rx_packet
MINOR: quic: extend pn_offset field from quic_rx_packet
MINOR: quic: define first packet flag
MINOR: quic: extract connection retrieval
MINOR: quic: split and rename qc_lstnr_pkt_rcv()
MINOR: quic: refactor packet drop on reception
MINOR: quic: extend Retry token check function
BUG/MINOR: mux-quic: complete flow-control for uni streams
MINOR: quic: do not crash on unhandled sendto error
MINOR: quic: display unknown error sendto counter on stat page
MINOR: quic: remove unnecessary quic_session_accept()
BUG/MINOR: quic: fix subscribe operation
MINOR: quic: add counter for interrupted reception
BUG/MINOR: quic: fix race condition on datagram purging
CLEANUP: cli: rename dynamic error printing state
MINOR: cli: define usermsgs print context
MINOR: server: clear prefix on stderr logs after add server
BUILD: quic: fix dubious 0-byte overflow on qc_release_lost_pkts
MINOR: ncbuf: complete doc for ncb_advance()
BUG/MEDIUM: quic: fix unsuccessful handshakes on ncb_advance error
BUG/MEDIUM: quic: fix memleak for out-of-order crypto data
MINOR: quic: complete traces/debug for handshake
Aurelien DARRAGON (6):
MINOR: list: fixing typo in MT_LIST_LOCK_ELT
DOC/MINOR: list: fixing MT_LIST_LOCK_ELT macro documentation
MINOR: list: adding MT_LIST_APPEND_LOCKED macro
BUG/MINOR: log: fixing bug in tcp syslog_io_handler Octet-Counting
BUG/MEDIUM: wdt/clock: properly handle early task hangs
BUG/MINOR: http_ana/txn: don't re-initialize txn and req var lists
Christopher Faulet (39):
BUG/MINOR: log: Preserve message facility when the log target is a ring
buffer
BUG/MINOR: ring: Properly parse connect timeout
BUG/MEDIUM: compression: handle rewrite errors when updating response
headers
BUG/MINOR: sink: Only use backend capability for the sink proxies
BUG/MINOR: sink: Set default connect/server timeout for implicit ring
buffers
BUG/MAJOR: stick-table: don't process store-response rules for applets
BUG/MINOR: http-htx: Fix error handling during parsing http replies
BUG/MINOR: resolvers: Don't wait periodic resolution on healthcheck
failure
BUG/MINOR: resolvers: Set port before IP address when processing SRV
records
BUG/MINOR: mux-fcgi: Be sure to send empty STDING record in case of
zero-copy
BUG/MEDIUM: mux-fcgi: Avoid value length overflow when it doesn't fit at
once
REG-TESTS: cache: Remove T-E header for 304-Not-Modified responses
MINOR: mux-h1: Remove usless code inside shutr callback
CLEANUP: mux-h1; Rename H1S_F_ERROR flag into H1S_F_ERROR_MASK
REORG: mux-h1: Reorg the H1C structure
CLEANUP: mux-h1: Rename H1C_F_ST_ERROR and H1C_F_ST_SILENT_SHUT flags
MINOR: mux-h1: Add a dedicated enum to deal with H1 connection state
MEDIUM: mux-h1: Handle H1C states via its state field instead of
H1C_F_ST_*
MINOR: mux-h1: Don't handle subscribe for reads in h1_process_demux()
CLEANUP: mux-h1: Rename H1C_F_ERR_PENDING into H1C_F_ABRT_PENDING
MINOR: mux-h1: Add flag on H1 stream to deal with internal errors
MEDIUM: mux-h1: Rely on the H1C to deal with shutdown for reads
CLEANUP: mux-h1: Reorder H1 connection flags to avoid holes
MEDIUM: mux-h1: Don't report a final error whe a message is aborted
MEDIUM: mux-pt: Don't always set a final error on SE on the sending path
MEDIUM: mux-h2: Introduce flags to deal with connection read/write errors
CLEANUP: mux-h2: Remove unused fields in h2c structures
MEDIUM: mux-fcgi: Introduce flags to deal with connection read/write
errors
MINOR: sconn: Set SE_FL_ERROR only when there is no more data to read
MINOR: mux-h1: Rely on a H1S flag to know a WS key was found or not
DOC: lua-api: Remove warning about the lua filters
BUG/MEDIUM: listener: Fix race condition when updating the global mngmt
task
CLEANUP: listener: Remove useless task_queue from
manage_global_listener_queue
BUG/MINOR: mux-h1: Fix error handling when H1S allocation failed on
client side
CLEANUP: mux-h1: Don't test h1c in h1_shutw_conn()
BUG/MEDIUM: raw-sock: Don't report connection error if something was
received
MINOR: cfgparse: Always check the section position
MEDIUM: thread: Restric nbthread/thread-group(s) to very first global
sections
BUILD: peers: Remove unused variables
Dridi Boukelmoune (1):
IMPORT: slz: mention the potential header in slz_finish()
Emeric Brun (2):
MINOR: peers: handle multiple resync requests using shards
BUG/MEDIUM: peers: messages about unkown tables not correctly ignored
Frédéric Lécaille (3):
BUILD: quic: QUIC mux build fix for 32-bit build
BUILD: quic: Fix build for m68k cross-compilation
MINOR: peers: Support for peer shards
Ilya Shipitsin (5):
BUILD: scripts: disable tests build on QuicTLS build
CI: add monthly gcc cross compile jobs
CLEANUP: assorted typo fixes in the code and comments
CI: switch to the "latest" LibreSSL
CI: enable QUIC for LibreSSL builds
Mickael Torres (1):
BUG/MINOR: mux-h1: Do not send a last null chunk on body-less answers
Miroslav Zagorac (1):
BUG/MINOR: httpclient: fixed memory allocation for the SSL ca_file
Remi Tricot-Le Breton (6):
BUG/MINOR: ssl: Memory leak of DH BIGNUM fields
BUG/MINOR: ssl: Memory leak of AUTHORITY_KEYID struct when loading issuer
BUG/MINOR: ssl: ocsp structure not freed properly in case of error
BUG/MEDIUM: ssl: Verify error codes can exceed 63
BUG/MINOR: ssl: Fix potential overflow
BUG/MINOR: ssl: SSL_load_error_strings might not be defined
Thierry Fournier (1):
BUG/MEDIUM: httpclient: segfault when the httpclient parser fails
William Lallemand (22):
BUG/MEDIUM: httpclient/lua: crash when the lua task timeout before the
httpclient
BUG/MEDIUM: httpclient: check if the httpclient was released in the IO
handler
REGTESTS: httpclient/lua: test the lua task timeout with the httpclient
CI: github: dump the backtrace of coredumps in the alpine container
BUILD: Makefile: add "USE_SHM_OPEN" on the linux-musl target
DOC: lua: add a note about compression w/ httpclient
CLEANUP: mworker/cli: rename the status function to loadstatus
MINOR: mworker/cli: does no try to dump the startup-logs w/o USE_SHM_OPEN
MINOR: ssl: add the SSL error string when failing to load a certificate
MINOR: ssl: add the SSL error string before the chain
MEDIUM: ssl: be stricter about chain error
MINOR: ssl: dump the SSL string error when SSL_CTX_use_PrivateKey()
failed.
CLEANUP: ssl: remove dead code in ssl_sock_load_pem_into_ckch()
MEDIUM: ssl: {ca,crt}-ignore-err can now use error constant name
MINOR: ssl: x509_v_err_str converter transforms an integer to a
X509_V_ERR name
BUG/MINOR: ssl: bind_conf is uncorrectly accessed when using QUIC
CLEANUP: ssl: remove printf in bind_parse_ignore_err
BUG/MINOR: ssl: crt-ignore-err memory leak with 'all' parameter
MINOR: ssl: ssl_sock_load_cert_chain() display error strings
MINOR: ssl: reintroduce ERR_GET_LIB(ret) == ERR_LIB_PEM in
ssl_sock_load_pem_into_ckch()
BUG/MINOR: ssl: don't initialize the keylog callback when not required
BUILD: Makefile: enable USE_SHM_OPEN by default on freebsd
Willy Tarreau (57):
BUILD: ssl_utils: fix build on gcc versions before 8
BUILD: debug: remove unnecessary quotes in HA_WEAK() calls
CI: emit the compiler's version in the build reports
IMPORT: xxhash: update xxHash to version 0.8.1
IMPORT: slz: declare len to fix debug build when optimal match is enabled
IMPORT: slz: define and use a __fallthrough statement for switch/case
BUILD: compiler: add a macro to detect if another one is set and equals 1
BUILD: compiler: add a default definition for __has_attribute()
BUILD: compiler: define a __fallthrough statement for switch/case
BUILD: sample: use __fallthrough in smp_is_rw() and smp_dup()
BUILD: quic: use __fallthrough in quic_connect_server()
BUILD: ssl/crt-list: use __fallthrough in cli_io_handler_add_crtlist()
BUILD: ssl: use __fallthrough in
cli_io_handler_commit_{cert,cafile_crlfile}()
BUILD: ssl: use __fallthrough in cli_io_handler_tlskeys_files()
BUILD: hlua: use __fallthrough in hlua_post_init_state()
BUILD: stream: use __fallthrough in stats_dump_full_strm_to_buffer()
BUILD: tcpcheck: use __fallthrough in check_proxy_tcpcheck()
BUILD: stats: use __fallthrough in stats_dump_proxy_to_buffer()
BUILD: peers: use __fallthrough in peer_io_handler()
BUILD: hash: use __fallthrough in hash_djb2()
BUILD: tools: use __fallthrough in url_decode()
BUILD: args: use __fallthrough in make_arg_list()
BUILD: acl: use __fallthrough in parse_acl_expr()
BUILD: spoe: use __fallthrough in spoe_handle_appctx()
BUILD: logs: use __fallthrough in build_log_header()
BUILD: check: use __fallthrough in __health_adjust()
BUILD: http_act: use __fallthrough in parse_http_del_header()
BUILD: h1_htx: use __fallthrough in h1_parse_chunk()
BUILD: vars: use __fallthrough in var_accounting_{diff,add}()
BUILD: map: use __fallthrough in cli_io_handler_*()
BUILD: compression: use __fallthrough in comp_http_payload()
BUILD: stconn: use __fallthrough in various shutw() functions
BUILD: prometheus: use __fallthrough in promex_dump_metrics() and IO
handler()
CLEANUP: stick-table: remove the unused table->exp_next
OPTIM: stick-table: avoid atomic ops in stktable_requeue_exp() when
possible
BUG/MEDIUM: stick-table: fix a race condition when updating the
expiration task
MEDIUM: http-ana: remove set-cookie2 support
MINOR: deinit: add a "quick-exit" option to bypass the deinit step
OPTIM: ebtree: make ebmb_insert_prefix() keep a copy the new node's pfx
OPTIM: ebtree: make ebmb_insert_prefix() keep a copy the new node's key
MINOR: pool/debug: create a new pool_alloc_flag() macro
MINOR: dynbuf: switch allocation and release to macros to better track
users
DOC: config: fix alphabetical ordering of global section
MINOR: trace: split the CLI "trace" parser in CLI vs statement
MEDIUM: trace: create a new "trace" statement in the "global" section
BUG/MEDIUM: ring: fix creation of server in uninitialized ring
BUILD: makefile: mark poll and tcploop targets as phony
BUILD: makefile: properly pass CC to sub-projects
BUILD: makefile: move default verbosity settings to
include/make/verbose.mk
BUILD: makefile: use $(cmd_MAKE) in quiet mode
BUILD: makefile: move the compiler option detection stuff to compiler.mk
DEV: poll: make the connect() step an action as well
DEV: poll: strip the "do_" prefix from reported function names
DEV: poll: indicate the FD's side in front of its value
BUG/MINOR: pool/cli: use ullong to report total pool usage in bytes
DOC: internal: commit notes about polling states and flags
DOC: internal: commit notes about polling states and flags on connect()
---
