From b9e1b09a5a584756a27cb1955b4d69512e7c6a56 Mon Sep 17 00:00:00 2001 From: Sheng Yang Date: Tue, 16 Apr 2019 14:52:14 -0700 Subject: [PATCH] Update multidisk.md --- docs/multidisk.md | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) diff --git a/docs/multidisk.md b/docs/multidisk.md index 97b78d9..535aec0 100644 --- a/docs/multidisk.md +++ b/docs/multidisk.md @@ -1,9 +1,11 @@ # Multiple disks support Longhorn supports to use more than one disk on the nodes to store the volume data. -To add a new disk for a node, heading to `Node` tab, select one of the node, and select `Edit Disks` in the drop down menu. +By default, `/var/lib/rancher/longhorn` on the host will be used for storing the volume data. You can avoid using the default directory by adding a new disk, then disable scheduling for `/var/lib/rancher/longhorn`. -By default, `/var/lib/rancher/longhorn` on the host will be used for storing the volume data. +## Add a disk + +To add a new disk for a node, heading to `Node` tab, select one of the node, and select `Edit Disks` in the drop down menu. To add any additional disks, user needs to: 1. Mount the disk on the host to a certain directory. @@ -15,9 +17,17 @@ User can reserve a certain amount of space of the disk to stop Longhorn from usi The kubelet needs to preserve node stability when available compute resources are low. This is especially important when dealing with incompressible compute resources, such as memory or disk space. If such resources are exhausted, nodes become unstable. To avoid kubelet `Disk pressure` issue after scheduling several volumes, by default, longhorn reserved 30% of root disk space (`/var/lib/rancher/longhorn`) to ensure node stability. +## Remove a disk Nodes and disks can be excluded from future scheduling. Notice any scheduled storage space won't be released automatically if the scheduling was disabled for the node. -There are two global settings affect the scheduling of the volume as well. +In order to remove a disk, two conditions need to be met: +1. The scheduling for the disk must be disabled +2. There is no existing replica using the disk, include the replica in error state. + +Once those two conditions are met, you should be allowed to remove the disk. + +## Configuration +There are two global settings affect the scheduling of the volume. `StorageOverProvisioningPercentage` defines the upper bound of `ScheduledStorage / (MaximumStorage - ReservedStorage)` . The default value is `500` (%). That means we can schedule a total of 750 GiB Longhorn volumes on a 200 GiB disk with 50G reserved for the root file system. Because normally people won't use that large amount of data in the volume, and we store the volumes as sparse files.