[SOLVED] LXC: More than 10 mount points possible?

[bogo22]

Hello,
i am using a linux container (samba server). I wanted to mount all zfs datasets from proxmox host with mount points to the linux-container to share them via samba.
I wondered why not all datasets are shared correlty (mp0 to mp9 is shared correctly and mp10 to mp22 isn't shared).
I finally found the answer (which is kind of disappointing): "and you can configure up to 10 additional mount points".
So I can't use more than 10 mount points? Do you know a good alternative way to share more than 10 datasets? Only way I know is to install samba-server directly on proxmox and share them there.

thanks

 

[LnxBil]

I have no other way, so I patched PVE to be able to have more. You can just edit the file /usr/share/perl5/PVE/LXC/Config.pm directly, yet here is a more automatic way

--- /tmp/Config.pm    2016-07-17 18:59:28.884082554 +0200
+++ /usr/share/perl5/PVE/LXC/Config.pm    2016-07-17 18:59:36.811969832 +0200
@@ -16,7 +16,7 @@
 my $lockdir = "/run/lock/lxc";
 mkdir $lockdir;
 mkdir "/etc/pve/nodes/$nodename/lxc";
-my $MAX_MOUNT_POINTS = 10;
+my $MAX_MOUNT_POINTS = 32;
 my $MAX_UNUSED_DISKS = $MAX_MOUNT_POINTS;

 # BEGIN implemented abstract methods from PVE::AbstractConfig

Create a file with the contents in /root/pct_increase_max_mount_points_to_32.diff and apply with:

cd / && patch -p1 < /root/pct_increase_max_mount_points_to_32.diff

I'm working (if I have more time) on an easier way to apply own PVE patches that are update-aware.

 

[fabian]

I'm working (if I have more time) on an easier way to apply own PVE patches that are update-aware.

I know it can sometimes be cumbersome or seem like unnecessary extra work, but in the long-term you are almost always better off trying to upstream custom patches. even if you script a way to auto-apply them after updates, "patch applies" does not equal "patch is working as intended"

 

[LnxBil]

Actually, I discussed it on the devel mailinglist:

https://pve.proxmox.com/pipermail/pve-devel/2016-May/021043.html

 

[bogo22]

Thanks for the patch and pointing to the mailing list.

@fabian It would be really great to be able to setup more than 10 mount points so the patch should be upstreamed.

PS: I read the mailing list and here is my modest opinion it: Even If more than 10 mount points really slows down the system i would accept that because for me it is more important to be able to mount all zfs datasets than the possible decrease in performance. So maybe you just add a note on the WIKI or inside the proxmox webgui (where i can add a mount point) that multiple mount points can slow down the container a bit. So every user can decide it by themself. LnxBil said that he doesn't noticed a performance slowdown though.

 

[fabian]

Actually, I discussed it on the devel mailinglist:

https://pve.proxmox.com/pipermail/pve-devel/2016-May/021043.html

indeed, it sometimes takes some perseverance to convince devs to merge changes for which they themselves don't see an immediate use.

maybe we can re-evaluate offering recursive bind mounts? if I understand you correctly, that would be exactly what you need? bind mounts are already a root only feature, but still the security implications need to be thoroughly evaluated.. (just take a peek at the current bind mount related checks needed to prevent symlink based shenanigans). maybe we should take this to pve-devel again?

 

[johannes]

We just ran into the limit of 10 mount points per container. Would it be sufficient to adjust $MAX_MOUNT_POINTS on all servers in the cluster? Could this become a per-cluster configuration option?

 

[bogo22]

I am still waiting that $MAX_MOUNT_POINTS will be increased by default too. So far i have to manually edit it after update. Currently i have over 20 mount points inside one lxc and it works great (no errors or slow startup/high CPU usage etc). What is the progress of that? If i am correct we just have to increase the default $MAX_MOUNT_POINTS and update the webgui so it can display all / more than 9 mount points under LXC->Resources

 

[johannes]

Which lines do you change for the mount points to be visible in the GUI. Do have a diff to be applied to the Proxmox code? Do you automatically re-apply it after every update?

 

[elurex]

what about using lxc.mount.entry = command in the config

 

[johannes]

As far as I understand, the mp options and lxc.mount.entry are the same except for the /var/lib/lxc/$VMID prefix. Are both variants reflected in the GUI? Is mixing problematic?

 

[elurex]

I only use lxc.mount.entry because it allows migration between nodes

 

[johannes]

I always append ,backup=1,shared=1 to every mount point to be able to migrate it.

 

[bogo22]

Which lines do you change for the mount points to be visible in the GUI. Do have a diff to be applied to the Proxmox code? Do you automatically re-apply it after every update?

i don't have a patch, I only wanted to point out what must be done by the dev team. So, as far as I know there are really some users who would love to be able to set more than 10 mount points so I hope the dev team will change it. I don't know what the blocker of this change is, I only can say that the performance is pretty good and there is not delay in lxc boot etc if this is what the dev's are concerned about.

 

[bogo22]

@fabian I would like to ask the status of it or what the final decision of increasing the default mount points ist. Like i said i set my $MAX_MOUNT_POINTS = 40; without any issues/delay.

 

[bogo22]

Seems like the new default for $MAX_MOUNT_POINTS is increased to 256, so this should not be a problem anymore. Thread can be closed IMO. Thanks