Hiya
Bill Nottingham wrote:
Stuart Children (stuart@xxxxxxxxxxxxxx) said:
Exactly... what needs to be done here is tagging of all the scripts/services
that can be shut down cleanly, then some quick changes to /etc/rc should
cut about half the time out of shutdown (based on some tests I ran ~10
months ago or so.)
How about a "prepare-for-system-shutdown" [1] command to the rc scripts? With the meaning of "save anything you need to 'cos the system's going down shortly" rather than "i want you to exit all your processes right now" (the 'stop' command). For most services this could just be a NOP. Others could trivially call their 'stop' command, or do whatever subset of that is necessary.
[1] a more succient name should obviously be used. :) Actually, how about 'save', and 'stop' could in general call that and then do its killings?
Breaks compatibility with third-party scripts.
How do you mean... you're only adding new functionality. Is the problem that third-party scripts wouldn't have a 'save' command? If so then you check the exit value of the script, and if it's 1 then retry with 'stop'. Again, you start to lose some of the time savings then, but this should not be the common case, and would improve as people updated their scripts.
Obviously if this were implemented it would make sense to get upstream init scripts to adopt the new command. I'm sure other distributions would appreciate it too.
Anyway - whether this is used or another method - I think it's important that the application/package can say "I do [not] need do to anything before system-shutdown", rather than having a list somewhere of what things we think are safe to skip shutdown on that is 1) potentially obscured and 2) requires being kept up to date. Having services called "httpd" doesn't help avoid problems here either - perhaps apache can be safely left running, but my database-based webserver needs to write stuff to disk... but that's another rant. :)
Cheers
-- Stuart Children
