7577eec93c
Previously, each run puppet runs a parted and a xfs_admin command, which itself triggers udev rules. In production, and during a rebalance, this may trigger a *very* high load on the storage nodes. This patch introduces 2 a new options: "manage_partition" and "manage_filesystem" which are "true" by default (not changing the past behavior), so administrators can, once the cluster is setup and the new disks are in, set this variable to "false", or manage partitions and formating out of puppet. Change-Id: I2740da550157c94cd80805f60321d3cd91381cff
14 lines
704 B
YAML
14 lines
704 B
YAML
---
|
|
features:
|
|
- |
|
|
The class swift::storage::disk now has 2 new parameters called
|
|
manage_partition and manage_filesystem. The swift::storage::xfs now has
|
|
a new parameter manage_filesystem. Both have a value of true by default.
|
|
These new options may be used to skip running parted and xfs_admin once
|
|
the storage disks are setup. The goal is to avoid the kernel to flush
|
|
the HDD buffer, which can seriously increase the load and IO wait on a
|
|
(busy) production server. These parameters should be left to true when
|
|
setting-up a new server, or when changing a disk, and can be set to false
|
|
otherwise, or always set to false if the formating is done outside of
|
|
puppet.
|