pstorage-config − global Pstorage settings
pstorage −c cluster get−config
pstorage −c cluster set−config var=value
Pstorage stores some of its settings in the replicated MDS journal which makes it possible to change easily these settings globally for the whole cluster and all its participants. These settings can be obtained with pstorage get−config and modified with pstorage set−config. Please refer to pstorage−config−files(7) if you are looking for description of local config files.
NOTE: Global configuration parameters are subject to change in future and should not be modified from their default values unless you are absolutely sure what you are doing. Changing non−documented parameters may lead to storage unavailablity and inability to recover from disasters, so please do not change without consulting support first.
Here are some important settings with their default values:
mds.alloc.fill_margin=5
The minimum percent of free space relative to the total CS space required to allocate from this CS. If there is less free space, allocation will not happen.
mds.alloc.free_space_margin=512
The minimum space left in megabytes required to allocate from this CS. If there is less free space, allocation will not happen.
mds.wd.max_offline_cs_hosts=2
The maximum number of offline hosts. If exceeded the cluster is considered as temporarily disconnected. Automatic replication will be suppressed to avoid burst of unneeded and uncontrolled replication all over the cluster. Such situations are extraordinary and may require human intervention anyway. Note, hosts are distinguished by unique host identifier stored at /etc/pstorage/host_id.
mds.wd.offline_tout=300000
Watchdog offline timeout in milliseconds. If an inactive CS has not registered during this time, it is marked as offline and recovery replication is initiated.
mds.wd.overcommit_margin=1
Number of extra replicas which are not dropped immediately, but are given some grace period (mds.wd.overcommit_tout) before being deleted. This prevents chunks from replicating, deleting and replicating back massively on frequent temporary disconnects.
mds.wd.overcommit_tout=300000
If the number of replicas of a chunk exceeds mds.wd.overcommit_margin + replicas.norm, extra replicas are dropped immediately. Otherwise the chunk is allowed to live in the overcommitted state for some time. See mds.wd.overcommit_margin description.
mds.wd.verify_chunks=2
The number of chunk verification requests sent to every CS per minute. These chunks are read to make sure they are still correct and readable. When SSD journal is used together with CSD, chunks checksums are also verified. Such process is called "data scrubbing".
mds.alloc.strict_tier=0
Enables chunk allocation from other storage tiers if the requested one is out of disk space. If strict tiering is disabled (0) and the requested tier runs out of disk space, the cluster will try to allocate disk space on lower tiers. If lower tiers run out of space as well, the cluster will try to allocate disk space on upper tiers. For example, if a user wants to allocate disk space on tier 2 which is out of space, then the cluster will attempt to allocate space on tiers 1, then 0, then, finally, tier 3.
mds.alloc.strict_failure_domain=1
Enables strict replica allocation based on failure domains information. Not recommended to change until absolutely sure. See pstorage−failure−domain(1) for more details.
mds.log.level, cs.log.level
Defines global clusterwide CS/MDS logging verbosity.
gen.license_status
Read−only status of current license. For active license returned status is 6. 2 for expired and 7 for grace period.
mds.alloc.separation.home
Allocation score bonus for replicas stored on the same host where client is running. This parameter affects how much allocator tries to push replicas on the client host. It’s not recommended to change this parameter default value except for the 0 value, which makes sense when it’s not important to allocate replicas locally and instead higher randomness with potentially better performance of a single client is more preferred.
−c, −−cluster=CLUSTER_NAME
Name of the cluster to connect to (mandatory). If the −c option is omitted, the PSTORAGE_NAME environment variable is used.
0
Success
Non−zero
Failure (syntax or usage error; configuration error; cluster failure; unexpected error).
pstorage
−c test_cluster get−config
pstorage −c test_cluster set−config
mds.alloc.strict_tier=0
pstorage −c test_cluster set−config
mds.log.level=7
Copyright © 2011−2013, Parallels, Inc. All rights reserved.
pstorage(1), pstorage−set−attr(1), pstorage−overview(7), pstorage−config−files(7), pstorage−failure−domain(1)