parallelvirtualcluster
/
pvc
1
0
Fork 0
Code Issues 11 Pull Requests Projects Releases 109 Wiki Activity

Adjust wording about extdb

This commit is contained in:
Joshua Boniface 2023-11-04 02:54:25 -04:00
parent ab156b14b7
commit 7cdedde2fb
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
client-cli/pvc/cli/cli.py
View File

@ -3432,9 +3432,9 @@ def cli_storage_osd_add(node, device, weight, ext_db_ratio, ext_db_size, osd_cou
The weight of an OSD should reflect the ratio of the size of the OSD to the other OSDs in the storage cluster. For example, with a 200GB disk and a 400GB disk in each node, the 400GB disk should have twice the weight as the 200GB disk. For more information about CRUSH weights, please see the Ceph documentation. The weight of an OSD should reflect the ratio of the size of the OSD to the other OSDs in the storage cluster. For example, with a 200GB disk and a 400GB disk in each node, the 400GB disk should have twice the weight as the 200GB disk. For more information about CRUSH weights, please see the Ceph documentation.
The "-r"/"--ext-db-ratio" or "-s"/"--ext-db-size" options, if specified, and if a OSD DB VG exists on the node (see "pvc storage osd create-db-vg"), will instruct the OSD to locate its DB database and WAL on a new logical volume on that OSD DB VG. If "-r"/"--ext-db-ratio" is specified, the sizing of this DB LV will be the given ratio (specified as a decimal percentage e.g. 0.05 for 5%) of the size of the OSD (e.g. 0.05 on a 1TB SSD will create a 50GB LV). If "-s"/"--ext-db-size" is specified, the sizing of this DB LV will be the given human-unit size (e.g. 1024M, 20GB, etc.). The "-r"/"--ext-db-ratio" or "-s"/"--ext-db-size" options, if specified, and if a OSD DB VG exists on the node (see "pvc storage osd create-db-vg"), will instruct the OSD to locate its RocksDB database and WAL on a new logical volume on that OSD DB VG. If "-r"/"--ext-db-ratio" is specified, the sizing of this DB LV will be the given ratio (specified as a decimal percentage e.g. 0.05 for 5%) of the size of the OSD (e.g. 0.05 on a 1TB SSD will create a 50GB LV). If "-s"/"--ext-db-size" is specified, the sizing of this DB LV will be the given human-unit size (e.g. 1024M, 20GB, etc.). An 0.05 ratio is recommended; at least 0.02 is required, and more than 0.05 can potentially increase performance in write-heavy workloads.
An external DB is only recommended for relatively slow OSD devices (i.e. SATA SSDs) when there is also a smaller, faster (i.e. NVMe or 3DXPoint SSD) device in the node. For NVMe OSDs, an external DB is not required nor recommended for optimal performance. An "--ext-db-ratio" of 0.05 (5%) is recommended for most workloads and OSD sizes; the Ceph documentation recommends a minimum of 0.02 (2%), and higher values may improve performance under write-heavy workloads with fewer OSDs per node. The explicit size option is also permitted to allow more fine-grained sizing, allowing the administrator to pre-calculate the desired size rather than relying on a ratio. WARNING: An external DB carries important caveats. An external DB is only suggested for relatively slow OSD devices (e.g. SATA SSDs) when there is also a much faster, more robust, but smaller storage device in the system (e.g. an NVMe or 3DXPoint SSD) which can accelerate the OSD. An external DB is NOT recommended for NVMe OSDs as this will hamper performance and reliability. Additionally, it is important to note that the OSD will depend entirely on this external DB device; they cannot be separated without destroying the OSD, and the OSD cannot function without the external DB device, thus introducting a single point of failure. Use this feature with extreme care.
The "-c"/"--osd-count" option allows the splitting of a single block device into multiple logical OSDs. This is recommended in the Ceph literature for extremely fast OSD block devices (i.e. NVMe or 3DXPoint) which can saturate a single OSD process. Usually, 2 or 4 OSDs is recommended, based on the size and performance of the OSD disk; more than 4 OSDs per volume is not recommended, and this option is not recommended for SATA SSDs. The "-c"/"--osd-count" option allows the splitting of a single block device into multiple logical OSDs. This is recommended in the Ceph literature for extremely fast OSD block devices (i.e. NVMe or 3DXPoint) which can saturate a single OSD process. Usually, 2 or 4 OSDs is recommended, based on the size and performance of the OSD disk; more than 4 OSDs per volume is not recommended, and this option is not recommended for SATA SSDs.