Ngx_openresty mainline version 1.4.3.1 released

Detlef_R · October 29, 2013, 9:50pm

Hello folks!

I am pleased to announce that the new mainline version of
ngx_openresty, 1.4.3.1, is now released:

http://openresty.org/#Download

Special thanks go to all the contributors for making this happen!

Below is the complete change log for this release, as compared to the
last (mainline) release, 1.4.2.9:

upgraded the Nginx core to 1.4.3.
- see the changes here: http://nginx.org/en/CHANGES-1.4
upgraded LuaNginxModule to 0.9.1.
- feature: added the new configuration directive
  “lua_use_default_type” for controlling whether to send out a
  default “Content-Type” response header (as defined by the
  default_type directive). default on. thanks aviramc for the
  patch.
- feature: now the raw request cosocket returned by
  ngx.req.socket(true) no longer requires the request body to
  be read already, which means that one can use this cosocket
  to read the raw request body data as well. thanks aviramc
  for the original patch.
- bugfix: when there were no existing “Cache-Control” response
  headers, “ngx.header.cache_control = nil” would
  (incorrectly) create a new “Cache-Control” header with an
  empty value. thanks jinglong for the patch.
- bugfix: the original letter-case of the header name was lost
  when creating the “Cache-Control” response header via the
  ngx.header.HEADER API.
- bugfix: ngx.req.set_header(“Host”, value) would overwrite
  the value of $host with bad values. thanks aviramc for the
  patch.
- bugfix: use of ngx.exit() to abort pending subrequests in
  other “light threads” might lead to segfault or request hang
  when HTTP 1.0 full buffering is in effect.
- bugfix: removing a request header might lead to memory
  corruptions. thanks BjÃ¸rnar Ness for the report.
- bugfix: reading ngx.status might get different values than
  $status. thanks Kevin Burke for the report.
- bugfix: downstream write events might interfere with
  upstream cosockets that are slow to write to. thanks Aviram
  Cohen for the report.
- bugfix: the bookkeeping state for already-freed user threads
  might be incorrectly used by newly-created threads that were
  completely different, which could lead to bad results.
  thanks Sam Lawrence for the report.
- bugfix: calling ngx.thread.wait() on a user thread object
  that is already waited (i.e., already dead) would hang
  forever. thanks Sam Lawrence for the report.
- bugfix: the alert “zero size buf” could be logged when
  assigning an empty Lua string (“”) to “ngx.arg[1]” in
  body_filter_by_lua*.
- bugfix: subrequests initiated by ngx.location.capture* could
  trigger unnecessary response header sending actions in the
  subrequest because our capturing output header filter did
  not set “r->header_sent”.
- bugfix: the Lua error message for the case that ngx.sleep()
  was used in log_by_lua* was not friendly. thanks Jiale Zhi
  for the report.
- bugfix: now ngx.req.socket(true) returns proper error when
  there is some other “light thread” reading the request body.
- bugfix: header_filter_by_lua*, body_filter_by_lua*, and
  ngx.location.capture* might not work properly with multiple
  “http {}” blocks in “nginx.conf”. thanks flygoast for the
  report.
- optimize: made ngx.re.match and ngx.re.gmatch faster for
  LuaJIT 2.x when there is no submatch captures.
- optimize: pre-allocate space for the Lua tables in various
  places.
- doc: fixed the context for the lua_package_path and
  lua_package_cpath directives. thanks duhoobo for the report.
upgraded HeadersMoreNginxModule to 0.23.
- bugfix: removing request headers via
  more_clear_input_headers might lead to memory corruptions.
- bugfix: more_set_input_headers might overwrite the value of
  the $host variable with bad values.
- bugfix: more_set_headers and more_clear_headers might not
  work when multiple “http {}” blocks were used in
  “nginx.conf”.
- bugfix: eliminated use of C global variables during
  configuration phase because it might lead to issues when HUP
  reload failed.
upgraded SrcacheNginxModule to 0.23.
- bugfix: this module might not work properly with multiple
  “http {}” blocks in “nginx.conf”.
- bugfix: we might (incorrectly) return 500 in our output
  filters.
- bugfix: we did not set “r->header_sent” when we want to
  discard the header in our header filter.
upgraded RdsJsonNginxModule to 0.12.
- bugfix: in case of multiple “http {}” blocks in
  “nginx.conf”, our output filters might be disabled even when
  this module is configured properly.
- bugix: we did not check the “NULL” pointer returned by an
  Nginx array element allocation.
upgraded RdsCsvNginxModule to 0.05.
- optimize: we now only register our output filters when this
  module is indeed used (the only exception is when multiple
  “http {}” blocks are used).
upgraded XssNginxModule to 0.04.
- optimize: we now only register our output filters when this
  module is indeed used (the only exception is when multiple
  “http {}” blocks are used).
upgraded EchoNginxModule to 0.49.
- bugfix: echo_before_body and echo_after_body might now work
  properly when multiple “http {}” blocks were used in
  “nginx.conf”.
upgraded LuaRestyRedisLibrary to 0.17.
- optimize: added an optional argument “n” to init_pipeline()
  as a hint for the number of pipelined commands.
- optimize: use LuaJIT 2.1’s new table.new() primitive to
  pre-allocate space for Lua tables.
upgraded LuaRestyUploadLibrary to 0.09.
- bugfix: removed use of the module() function to prevent bad
  side-effects.
- optimize: Removed use of lua tables and table.concat() for
  simple one-line Lua string concatenations.
upgraded LuaRestyMySQLLibrary to 0.14.
- bugfix: avoided using Lua 5.1’s module() function for
  defining our Lua modules because it has bad side effects.
- optimize: added an optional new argument “nrows” to the
  query() and read_result() methods, which can speed up things
  a bit.
- optimize: use LuaJIT v2.1’s new table.new() API to optimize
  Lua table allocations. when table.new is missing, just fall
  back to the good old “{}” constructor. this gives 12%
  overall speed-up for a typical result set with 500 rows when
  LuaJIT 2.1 is used.
- optimize: eliminated use of table.insert() because it is
  slower than “tb[#tb + 1] = val”.
- optimize: switched over to the multi-argument form of
  string.char().
- optimize: no longer use Lua tables and table.concat() to
  construct simple query strings.
upgraded LuaRestyWebSocketLibrary to 0.02.
- optimize: use LuaJIT 2.1’s table.new() to preallocate space
  for Lua tables, eliminating the overhead of Lua table
  rehash.
feature: applied the proxy_host_port_vars patch to the Nginx
core to make $proxy_host and $proxy_port accessible for dynamic
languages like Lua and Perl.
bugfix: applied the gzip_flush_bug patch to the Nginx core to
fix request hang caused by the ngx_gzip and ngx_gunzip modules
when using ngx.flush(true), for example. Thanks Maxim D. for
the review.
bugfix: applied the cache_lock_hang_in_subreq patch to the Nginx
core to fix the request hang when using proxy_cache_lock in
subrequests and the cache lock timeout happens.
bugfix: backported Maxim D.'s patch to fix an issue in the
ngx_gzip module: it did not clear “r->connection->buffered” when
the pending data was already flushed out. this could hang
LuaNginxModule’s ngx.flush(true) call, for example.

The HTML version of the change log with lots of helpful hyper-links
can be browsed here:

http://openresty.org/#ChangeLog1004003

OpenResty (aka. ngx_openresty) is a full-fledged web application
server by bundling the standard Nginx core, lots of 3rd-party Nginx
modules and Lua libraries, as well as most of their external
dependencies. See OpenResty’s homepage for details:

http://openresty.org/

We have run extensive testing on our Amazon EC2 test cluster and
ensured that all the components (including the Nginx core) play well
together. The latest test report can always be found here:

http://qa.openresty.org

Enjoy!
-agentzh