<div dir="auto">The trick for us on AIX in the inittab I have a script fswait.ksh and monitors for the cluster mount point to be available before allowing the cluster dependent startup item (lower in the inittab) I'm pretty sure Linux has a way to define a dependent service.. define a cluster ready service and mark everything else as dependent on that or one of it's descendents. You could simply put the wait on FS in your dependent services start script as an option as well.<div dir="auto"><br></div><div dir="auto">Lookup systemd and then After= or Part of= if memory serves me right on Linux.</div><div dir="auto"><br></div><div dir="auto">For the mmfsup script it goes into /var/mmfs/etc/mmfsup</div><div dir="auto"><br></div><div dir="auto">The cluster will call it if present when the node is ready.</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Feb 22, 2022, 12:13 PM Skylar Thompson <<a href="mailto:skylar2@uw.edu">skylar2@uw.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">The problem might be that the service indicates success when mmstartup<br>
returns rather than when the mount is actually active (requires quorum<br>
checking, arbitration, etc.). A couple tricks I can think of would be using<br>
ConditionPathIsMountPoint from systemd.unit[1], or maybe adding a<br>
callback[2] that triggers on the mount condition for your filesystem that<br>
makes the bind mount rather than systemd.<br>
<br>
[1] <a href="https://www.freedesktop.org/software/systemd/man/systemd.unit.html#ConditionPathIsMountPoint=" rel="noreferrer noreferrer" target="_blank">https://www.freedesktop.org/software/systemd/man/systemd.unit.html#ConditionPathIsMountPoint=</a><br>
[2] <a href="https://www.ibm.com/docs/en/spectrum-scale/5.1.2?topic=reference-mmaddcallback-command" rel="noreferrer noreferrer" target="_blank">https://www.ibm.com/docs/en/spectrum-scale/5.1.2?topic=reference-mmaddcallback-command</a><br>
<br>
These are both on our todo list for improving our own GPFS mounting as we<br>
have problems with our job scheduler not starting reliably on reboot, but<br>
for us we can have Puppet start it on the next run so it just means nodes<br>
might not return to service for 30 minutes or so.<br>
<br>
On Tue, Feb 22, 2022 at 03:05:58PM -0500, Justin Cantrell wrote:<br>
> This is how we're currently solving this problem, with systemd timer and<br>
> mount. None of the requires seem to work with gpfs since it starts so late.<br>
> I would like a better solution.<br>
> <br>
> Is it normal for gpfs to start so late?? I think it doesn't mount until<br>
> after the gpfs.service starts, and even then it's 20-30 seconds.<br>
> <br>
> <br>
> On 2/22/22 14:42, Skylar Thompson wrote:<br>
> > Like Tina, we're doing bind mounts in autofs. I forgot that there might be<br>
> > a race condition if you're doing it in fstab. If you're on system with systemd,<br>
> > another option might be to do this directly with systemd.mount rather than<br>
> > let the fstab generator make the systemd.mount units:<br>
> > <br>
> > <a href="https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.freedesktop.org%2Fsoftware%2Fsystemd%2Fman%2Fsystemd.mount.html&amp;data=04%7C01%7Cjcantrell1%40gsu.edu%7C2a65cd0ddefd48cb81a308d9f63bb840%7C515ad73d8d5e4169895c9789dc742a70%7C0%7C0%7C637811559082622923%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=%2BWWD7cCNSMeJEYwELldYT3pLdXVX3AxJj7gqZQCqUv4%3D&amp;reserved=0" rel="noreferrer noreferrer" target="_blank">https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.freedesktop.org%2Fsoftware%2Fsystemd%2Fman%2Fsystemd.mount.html&amp;data=04%7C01%7Cjcantrell1%40gsu.edu%7C2a65cd0ddefd48cb81a308d9f63bb840%7C515ad73d8d5e4169895c9789dc742a70%7C0%7C0%7C637811559082622923%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=%2BWWD7cCNSMeJEYwELldYT3pLdXVX3AxJj7gqZQCqUv4%3D&amp;reserved=0</a><br>
> > <br>
> > You could then set RequiresMountFor=gpfs1.mount in the bind mount unit.<br>
> > <br>
> > On Tue, Feb 22, 2022 at 02:23:53PM -0500, Justin Cantrell wrote:<br>
> > > I tried a bind mount, but perhaps I'm doing it wrong. The system fails<br>
> > > to boot because gpfs doesn't start until too late in the boot process.<br>
> > > In fact, the system boots and the gpfs1 partition isn't available for a<br>
> > > good 20-30 seconds.<br>
> > > <br>
> > > /gfs1/home??? /home??? none???? bind<br>
> > > I've tried adding mount options of x-systemd-requires=gpfs1, noauto.<br>
> > > The noauto lets it boot, but the mount is never mounted properly. Doing<br>
> > > a manual mount -a mounts it.<br>
> > > <br>
> > > On 2/22/22 12:37, Skylar Thompson wrote:<br>
> > > > Assuming this is on Linux, you ought to be able to use bind mounts for<br>
> > > > that, something like this in fstab or equivalent:<br>
> > > > <br>
> > > > /home /gpfs1/home bind defaults 0 0<br>
> > > > <br>
> > > > On Tue, Feb 22, 2022 at 12:24:09PM -0500, Justin Cantrell wrote:<br>
> > > > > We're trying to mount multiple mounts at boot up via gpfs.<br>
> > > > > We can mount the main gpfs mount /gpfs1, but would like to mount things<br>
> > > > > like:<br>
> > > > > /home /gpfs1/home<br>
> > > > > /other /gpfs1/other<br>
> > > > > /stuff /gpfs1/stuff<br>
> > > > > <br>
> > > > > But adding that to fstab doesn't work, because from what I understand,<br>
> > > > > that's not how gpfs works with mounts.<br>
> > > > > What's the standard way to accomplish something like this?<br>
> > > > > We've used systemd timers/mounts to accomplish it, but that's not ideal.<br>
> > > > > Is there a way to do this natively with gpfs or does this have to be done<br>
> > > > > through symlinks or gpfs over nfs?<br>
> > > _______________________________________________<br>
> > > gpfsug-discuss mailing list<br>
> > > gpfsug-discuss at <a href="http://spectrumscale.org" rel="noreferrer noreferrer" target="_blank">spectrumscale.org</a><br>
> > > <a href="https://nam11.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgpfsug.org%2Fmailman%2Flistinfo%2Fgpfsug-discuss&amp;data=04%7C01%7Cjcantrell1%40gsu.edu%7C2a65cd0ddefd48cb81a308d9f63bb840%7C515ad73d8d5e4169895c9789dc742a70%7C0%7C0%7C637811559082622923%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=F4oXAT0zdY%2BS1mR784ZGghUt0G%2F6Ofu36MfJ9WnPsPM%3D&amp;reserved=0" rel="noreferrer noreferrer" target="_blank">https://nam11.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgpfsug.org%2Fmailman%2Flistinfo%2Fgpfsug-discuss&amp;data=04%7C01%7Cjcantrell1%40gsu.edu%7C2a65cd0ddefd48cb81a308d9f63bb840%7C515ad73d8d5e4169895c9789dc742a70%7C0%7C0%7C637811559082622923%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=F4oXAT0zdY%2BS1mR784ZGghUt0G%2F6Ofu36MfJ9WnPsPM%3D&amp;reserved=0</a><br>
> _______________________________________________<br>
> gpfsug-discuss mailing list<br>
> gpfsug-discuss at <a href="http://spectrumscale.org" rel="noreferrer noreferrer" target="_blank">spectrumscale.org</a><br>
> <a href="http://gpfsug.org/mailman/listinfo/gpfsug-discuss" rel="noreferrer noreferrer" target="_blank">http://gpfsug.org/mailman/listinfo/gpfsug-discuss</a><br>
<br>
-- <br>
-- Skylar Thompson (<a href="mailto:skylar2@u.washington.edu" target="_blank" rel="noreferrer">skylar2@u.washington.edu</a>)<br>
-- Genome Sciences Department (UW Medicine), System Administrator<br>
-- Foege Building S046, (206)-685-7354<br>
-- Pronouns: He/Him/His<br>
_______________________________________________<br>
gpfsug-discuss mailing list<br>
gpfsug-discuss at <a href="http://spectrumscale.org" rel="noreferrer noreferrer" target="_blank">spectrumscale.org</a><br>
<a href="http://gpfsug.org/mailman/listinfo/gpfsug-discuss" rel="noreferrer noreferrer" target="_blank">http://gpfsug.org/mailman/listinfo/gpfsug-discuss</a><br>
</blockquote></div>