Release Notes: A real fix for an incorrectly fixed timeout bug in 1.1.25. New control of per-process FD limits in the configuration, a new "monitor-net" keyword to bypass processing for external health-checkers, and even finer grained error reporting in the logs, along with the number of active sessions at any time. It now correctly builds on PPC64 and Debian 3.1, and has been running for about 1 month on OpenBSD. The "logging" section of the documentation has been greatly improved with detailed examples.
Release Notes: A real fix for an incorrectly fixed timeout bug in 1.1.25. New control of per-process FD limits in the configuration, a new "monitor-net" keyword to bypass processing for external health-checkers, and even finer grained error reporting in the logs, along with the number of active sessions at any time. It now correctly builds on PPC64, and has been tested in production for about 1 month on Solaris, Linux, and OpenBSD. The "logging" section of the documentation has been greatly improved with detailed examples.
Release Notes: This release fixes a nasty bug related to the epoll() optimization where some events would not be re-armed if an fd was closed and reassigned during the same iteration. A cleaner fix will be done in a later version. As select() is not an option above 10000 hits/s, epoll() users must upgrade. Non-epoll users are not affected.
Release Notes: This version introduces poll() and epoll() support. Epoll syscall definitions are even provided in case libc does not provide them. Initial tests of epoll have shown consistent performance of 13000 hits/s from 1 to 40,000 simultaneous sessions. This version seems stable enough that non-mission critical users are invited try it. Solaris' default FD_SETSIZE was raised from 1024 to 65536. Everything else kept in sync with 1.1.31. Solaris and Linux binaries are also provided.
Release Notes: HTTP/303 redirection was added. Regex syntax checking in the configuration parser was improved to avoid trouble at runtime. FD_SETSIZE was configured to 65536 by default on Solaris because several users got hit by the default limit of 1024 fds. Solaris users are invited to either upgrade or recompile their current version to avoid being hit by this low limit.
Release Notes: This version is strictly equivalent to 1.1.30, to which it adds IPv6 support on the client side. It is very stable and can be used in production if IPv6 is required.
Release Notes: The max header size has increased to 8 KB, a new option "source" allows per-server source binding, several servers can share the same id (this eases soft-stop and off-site persistence), a new cookie "prefix" method enables persistence on thin clients limited to one cookie (e.g., PDAs or mobile phones), the server call order now matches the documentation, and a completely new architecture guide shows how to address specific needs such as integration with Layer4 load-balancers, VRRP, and site fail-over with multi-level persistence.
Release Notes: This release adds an option to log arbitrary request and response headers. Now unprintable characters are logged encoded to prevent exploits or log-escape attacks. A bug where a TCP connection could be logged twice if the 'logasap' option was set without 'tcplog'. Now the Solaris build does not need -DSOLARIS anymore. RPM sources and binaries are now provided for RHEL3/i386. Several contribution patches have been published on the home page.
Release Notes: This version added a new command-line parameter for checking the config file without starting the daemon, the ability to block responses from servers based on header matching, a single line option to disable HTTP keep-alive without having to resort to four regexes, an option to log as soon as the session establishes, and an option to block a response if it contained a 'set-cookie' field and was cacheable. A Red Hat spec and init scripts was included in this release, although no RPM is available yet.
Release Notes: A problem has been fixed where the HTTP health check code assumed that the request was still of fixed length, which is obviously wrong since it became configurable. This could make HTTP health checks fail on non-default configurations.