On Fri, Feb 10, 2017 at 12:27 PM, Chris Adams <linux@xxxxxxxxxxx> wrote: > Once upon a time, Justin Forbes <jmforbes@xxxxxxxxxxx> said: >> * AGREED: Discussion on anaconda LVM change is delayed until 2-17 >> provided open questions get answered (+1:6,0:0,-1:0) (jforbes, >> 17:24:33) > > So, not sure if this is in the "open questions", but on the mailing list > I brought up the fact that MD RAID has an automatic cron job to do > consistency checks and LVM RAID does not. I don't see anything about > that in the change proposal, but IMHO that's a regression (and a > significant one, given studies that show undetected RAID failures are a > real thing). The cron job exists but as far as I can tell it is not doing anything out of the box. I've never seen md scrubs happen on a schedule automatically. LVM does have a way to do scrubs, and it'd be fairly easy to make a systemd timer that'd do it on a schedule. The lack of it existing or enabled by default I'd say is not good. But as far as who's responsibility it is? I'd say it's the GUI program that creates the array. If a GUI program is going to get into the business of making it easier to create arrays, then it needs to be in the business of configuring preventative maintenance and warnings. But we don't even have configuration for email warnings for mdadm arrays. And further there's a rather significant kernel deficiency with SCSI command timers being set by default to 30 seconds, which is totally contrary to getting proper bad sector recoveries with the vast majority of drives - whether they are in an array or not. And while the upstream file system and md kernel developers know about this, there is an incredible amount of resistance changing it. It's bad enough that this often thwarts scrubs (whether mdadm, LVM, or Btrfs based). So the net here is that there are all sorts of ways this can suck for the uninitiated user out of the box. Is the lack of scrub happening out of the box a problem? Yes but enabling it without fixing the bullshit 30 second SCSI command timer default is a bigger problem, and just enabling scrubs without fixing the timer means you end up with a greater chance md is going to kick a drive out of the array for the very simple problem of one bad sector, instead of fixing the bad sector with a remap. And if we aren't getting device faulty notifications in GNOME shell, now we're in a much net worse situation. > > I also don't know if LVM RAID can email (or otherwise notify admins) > about failures like MD RAID already does. Nope. The monitoring works with dmeventd, but I'm not sure how sophisticated the monitoring is or what messaging method it uses (dbus?) or what monitors it (udisksd or storaged?). But again, the loss of email notification is not a show stopper. I think email notification is archaic anyway, I'd rather it support some kind of ticketing system, reporting it all to storaged or other daemon of my choice, where I can choose how I want to be notified: pop-up, email, text message, tweet, light a match...etc. -- Chris Murphy _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
