Many daemons output some lines in their logs (here assumed to have been
redirected to logging pipes in some supervision framework) that can be
used as a kind of markers for readiness. So if we can implement the
following program, we will able to trivially add readiness support to
these daemons without patching their source code, while reducing the
daemon-specific code (the pattern recogniser) to a minimum:
* The program is double-forked from the ./run script of a service, and
also watches for the death of the daemon: in case the daemon dies
before logging the readiness marker (and optionally, if a specified
timeout is reached), this program dies as well, killing its subprocess
(the pattern recogniser, discussed below). To maximise portability
across supervision frameworks, the death notification can be
implemented with a pipe that can be written to by an ancillary
program ran from the ./finish script.
* The program somehow (directly or indirectly) gets a copy of the
log from the reading side of the logging pipe, and feeds it into a
specified pattern recogniser, which should exit 0 upon finding the
readiness marker. A special exit value of the subprocess can be
considered a notification for fatal error erroneously ignored by the
daemon, leading to the program exiting some special value which can
be handled by a `s6-svc -t' invocation. In order to avoid potential
feedback loops, the subprocess should be killed (and waited for)
before the program outputs any warning; the subprocess should be
responsible for keeping itself quiet when running normally.
Any idea on how the log "teeing" may be done cleanly (and portably
if possible; something akin to `tail -f' seems unsuitable because of
potential log rotation), and perhaps any flaw or redundancy in the
design above?
--
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2022.09.20)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C
Received on Sat Oct 23 2021 - 18:03:01 CEST