Hi supervision@! I realise this is probably more of a skaware@ question as it makes extensive reference to execline and s6-l-i, but I've been perusing the source code of the latter, hoping that I could learn what "the fifo trick" actually was (I have to admit, I've also been perusing skarnet.org/software) and I think I now get it. In case the fifo doesn't exist, the stage 1 initializer (since I won't be using s6-l-i as I'm not on Linux) has to create the fifo that $LOGGERSERVICE reads (see note [loggerservice-runscript] later in this email), then, let's call that fifo $LOGGERSERVICE/fifo, that initializer has to exec into (execline) redirfd -w -n -b 1 $LOGGERSERVICE/fifo s6-svscan. This is so technically nonsensical that I should probably be barred from ever using a computer again, but please stick with me here. That svscan invocation (or any other service supervisor that doesn't mind SIGPIPE) would then initiate supervision of [loggerservice-runscript] and the knot would be untangled.
[loggerservice-runscript] #!/package/admin/execline/command/execlineb /package/admin/execline/command/redirfd -r 0 $LOGGERSERVICE/fifo s6-log n10 s10485760 T !"gzip -nq9" $LOGDIRECTORY/s6-svscan # obviously one does not simply use $variables in a runscript - these'd be replaced by the absolute paths at compile time. Forgiving privilege separation failures and minor grammatical mistakes, does it look as if I understand the fifo trick's application in practice? Thanks a lot, -- Ellenor Bjornsdottir (she) sysadmin umbrellix.net aspiring OS integrator
