Please do not use version 1.0, it's is broken. The so called stable version had severe flaw in how status file treated. Fact that the file was read at start, and overwrote without re-writing it's contents immediately lead to situation in which the file was empty at stop. I hoped signal action would have ensured this cannot happen, but init script was killing the daemon with signal KILL so signal handler never ran. Oh dear, oh dear, the status end up getting lost - assuming one would restart the daemon without changing state.
The status file problems are now fixed. Other than that the daemon now uses epoll for maximum performance, handling of buffers is made simple, there is more error checking, and few code clean ups. With a bit of luck there will not be need for a new release any time soon, but we will see.
It is time to declare the software is mature enough to come out of zero series. With that I also promise to try harder to avoid version compatibility problems, which indicates this update is again breaking stuff, so please read the four warnings below.
- The status file is moved (again), and this time from /var/spool to /var/lib to be compatible with Linux FHS.
- Remote status check response tells the address who responded. This will break check scripts that use --address option.
- PATH for pre and post scripts has various bin directories in different order. See manual page for details.
- The IPC message got internal format update, that now includes information who sent state change message (pid & tty). This makes old f5gs client incompatible with new server software. (But why anyone would have more than one of these binaries?)
Features and fixes.
The release includes new --force option, that is similar with no-scripts --but will run scripts. At stop systemd is notified the service is going down.
The rest of the changes are none functional improvements, such as more error checking, more regression tests, clarification to code, cleaner signal handling, and so on. See git log for details.
The systemd support is now enabled by default when build has needed libraries. Earlier --quiet option implied --no-scripts option, which was a bug and now fixed. The syslog and journald messages are made to have similar content, so that one can compare functionality in between system easier. When post-scripts fail such is no longer considered fatal. Fix to crash caused by missing state directory. Thread safety improvements. Add ChangeLog to distributions, with content of git shortlog.
Notable interface changes include deprecation of receiving signals to change status. Now daemon will exit if it receives same signals earlier changed the state. The state changes can be done by sending IPC message, that is easiest with the command itself. Rationality of this interface change is that after using this software for real it has become clear other software will never send signals to this daemon. Automatic state changes are likely to use the command line interface, which means rather messy signal catching can be replaced with a lot less problematic IPC messaging.
Another interfece change is addition of --reason "message here" and --why options, that persist over daemon restarts and server boot. The --why printout will include time stamp of when the state change happen, and how long it has been since that time. While the message is likely to be omitted by most of the users the automatic time stamp is something I think has value. Again the actual use of the software has made it clear that sometimes one wants to know what happen and when. So who ever adds state changes to scripts should always add --reason. Otherwise debugging is more annoying.
Release v0.4. When ever a developer tells program is ready to be used is is reasonable to doubt statement being somehow wrong by mistake, or due overconfidence. The latest version changes persistent state file path, and format.
Format got go be changed because use of state numbers changed, and it became obvious the state file has to be versioned in order to recover from similar mess in future. This time the version checking is not done quite by the book, because version file did not persist over reboot. Earlier use of /var/run was a mistake, now the state is wrote to /var/spool instead.
If you upgrade be sure to set preferred state after an upgrade action. Lets hope there are no more breaking changes in future, that require user interaction.
This release adds support to systemd and makes messaging to use journald when they are available. Various threading and signaling issues are fixed. The contrib directory has an example to puppet management of the f5gs, and a script demonstrating how to perform pre-state change actions. Some of the changes make the command performance to be better, while other ensure the daemon is very difficult to knock out with (d)dos. The making of abuse hard meant fixing all resource leaks.
IMHO the software has production quality.
This makes the utility to behave like a reasonable daemon. It also includes manual page, autotools setup with gnulib to make the thing portable.
There is also support for pre and post state change scripts, that one can use to do additional tasks at a time of an event. The reason why a script support was added is local installation I am working with. When a java service is put to maintenance or disable state, e.g., F5 is told not to send traffic, broadcast messaging from that java service has to be halted as well. The easiest way to do that is to add a iptables deny rule for broadcasts, but I cannot possibly add such to the f5gs, so the support for 'random' actions has to be external.
My intention is to control contents of pre annd post scripts with puppet. The nice thing about this decision is that the puppet is environment aware, so I can fairly precisely tell what sort of additional actions are needed on a server I plan to control.
The beginning.