document how to deal with LVM cache resizes properly

This was kind of scattered all over the place and not directly linked
from the (most usual) Ganeti procedure. Now we have the procedure all
in one shot, and with the fat warning about the most common error I've
done twice already.

Used with success in team#41258.

howto/ganeti.md

@@ -402,7 +402,12 @@ Finally resize the filesystem:
 
     resize2fs /dev/vg_test-01/var-opt
 
-See also the [LVM howto](howto/lvm).
+See also the [LVM howto](howto/lvm), particularly if the `lvextend`
+step fails with:
+
+```
+  Unable to resize logical volumes of cache type.
+```
 
 #### Resizing without LVM, no partitions
 

howto/lvm.md

@@ -41,7 +41,7 @@ setup caching
 
 ## Disabling / Recovering from a cache failure
 
-If for some reason the cache LV is destroyed or loss (typically by
+If for some reason the cache LV is destroyed or lost (typically by
 naive operator error), it might be possible to restore the original LV
 functionality with:
 
@@ -192,6 +192,80 @@ Note that the resize process can take a while. Growing an active 20TB
 partition to 30TB took about 5 minutes, for example. The `-p` flag
 that could show progress only works in the "offline" procedure (below).
 
+If the above fails because of the following error:
+
+```
+  Unable to resize logical volumes of cache type.
+```
+
+It's because the logical volume has a cache attached. Follow the above
+procedure to "uncache" the logical volume and then re-enable the
+cache. 
+
+WARNING: Make sure you remove the physical volume cache from the
+volume group before you resize, otherwise the logical volume will be
+extended to also cover that and re-enabling the cache won't be
+possible! A typical, incorrect session looks like:
+
+```
+root@materculae:~# lvextend -l '+100%FREE' vg_materculae/srv
+  Unable to resize logical volumes of cache type.
+root@materculae:~# lvconvert --uncache vg_materculae/srv
+  Logical volume "srv_cache" successfully removed
+  Logical volume vg_materculae/srv is not cached.
+root@materculae:~# lvextend -l '+100%FREE' vg_materculae/srv
+  Size of logical volume vg_materculae/srv changed from <150.00 GiB (38399 extents) to 309.99 GiB (79358 extents).
+  Logical volume vg_materculae/srv successfully resized.
+root@materculae:~# lvs
+  LV   VG            Attr       LSize   Pool Origin Data%  Meta%  Move Log Cpy%Sync Convert
+  srv  vg_materculae -wi-ao---- 309.99g
+root@materculae:~# vgs
+  VG            #PV #LV #SN Attr   VSize   VFree
+  vg_materculae   2   1   0 wz--n- 309.99g    0
+root@materculae:~# pvs
+  PV         VG            Fmt  Attr PSize    PFree
+  /dev/sdc   vg_materculae lvm2 a--   <10.00g    0
+  /dev/sdd   vg_materculae lvm2 a--  <300.00g    0
+```
+
+A proper procedure is:
+
+    VG=vg_$(hostname)
+    FAST=/dev/sdc
+    lvconvert --uncache $VG/srv
+    vgreduce $VG/srv $FAST # remove the cache volume
+    lvextend -l '+100%FREE' $VG/srv # resize the volume
+    vgextend $VG $FAST # re-add the cache volume
+    lvcreate -n cache -l '100%FREE' $VG $FAST
+    lvconvert --type cache --cachevol cache $VG
+
+And here's a successful run:
+
+```
+root@materculae:~# VG=vg_$(hostname)
+root@materculae:~# FAST=/dev/sdc
+root@materculae:~# vgreduce $VG $FAST
+  Removed "/dev/sdc" from volume group "vg_materculae"
+root@materculae:~# vgs
+  VG            #PV #LV #SN Attr   VSize    VFree  
+  vg_materculae   1   1   0 wz--n- <300.00g <10.00g
+root@materculae:~# lvextend -l '+100%FREE' $VG
+  Size of logical volume vg_materculae/srv changed from 150.00 GiB (38400 extents) to <300.00 GiB (76799 extents).
+  Logical volume vg_materculae/srv successfully resized.
+root@materculae:~# vgextend $VG $FAST
+  Volume group "vg_materculae" successfully extended
+root@materculae:~# lvcreate -n cache -l '100%FREE' $VG $FAST
+  Logical volume "cache" created.
+root@materculae:~# lvconvert --type cache --cachevol cache vg_materculae
+Erase all existing data on vg_materculae/cache? [y/n]: y
+  Logical volume vg_materculae/srv is now cached.
+  Command on LV vg_materculae/cache_cvol requires LV with properties: lv_is_visible .
+```
+
+Note that the above output was edited for correctness: the actual run
+was much bumpier and involved shrinking the logical volume as the
+"incorrect" run was actually done in [tpo/tpa/team#41258](https://gitlab.torproject.org/tpo/tpa/team/-/issues/41258).
+
 ## offline procedure (ext2)
 
 To resize the partition to take up all available free space, you