You can install Portworx with a built-in internal KVDB. Portworx automatically deploys the internal KVDB cluster on a set of three nodes in your cluster and removes the requirement for an external KVDB such as etcd.
To install Portworx with a built-in internal KVDB, follow the instructions in the next sections, depending on your environment.
On the Portworx spec generator page in PX-Central, under the ETCD section select the Built-in option to deploy Portworx with internal KVDB.
While deploying the Portworx framework, provide the
-b argument in the extra arguments section.
For all other schedulers, Portworx requires an external etcd, to bootstrap its internal KVDB. Portworx will use this external etcd, to discover its own internal KVDB nodes. While installing Portworx provide the
-b argument to instruct Portworx to setup internal KVDB. Along with
-b, provide the
-k argument with a list of external etcd endpoints.
Storage for internal KVDB
Internal KVDB metadata needs to be stored on a persistent disk. There are two ways in which internal KVDB can store this data
Metadata Drive (Recommended)
Provide a separate drive to Portworx nodes through the
This metadata drive will be used for storing internal KVDB data as well as Portworx metadata such as journal or timestamp data.
This is the recommended method as the disk IO for internal KVDB is not shared with Portworx volume IO.
Auto (Not recommended)
If a metadata drive is not provided Portworx will reserve some space in the same storage pool which is used for storing your volume data.
This method is not recommended as the disk will be shared between internal KVDB IO and Portworx volume IO causing degraded internal KVDB performance.
-metadataargument, cloud deployments require a 150GB SSD disk for the internal KVDB.
Designate nodes as KVDB nodes
There are two ways in which you can designate a node as a KVDB node:
You can use custom labels to designate nodes as KVDB nodes through the
If you are running Portworx on Kubernetes, you can use the
px/metadata-node=truelabel to designate the set of nodes that will run the internal KVDB.
Use the following command to label nodes in Kubernetes:
kubectl label nodes <list of node-names> px/metadata-node=true
Depending upon the labels and their values a decision will be made
- If a node is labeled px/metadata-node=true then it will be a part of the internal KVDB cluster
- If node is labeled px/metadata-node=false then it will NOT be a part of the internal KVDB cluster
- If no node labels are found then all the nodes can potentially run internal kvdb
- If an incorrect label is present on the node like px/metadata-node=blah then Portworx will not start on that node.
- If no node is labelled as px/metadata-node=true , but one node is labeled as px/metadata-node=false then that node will never be a part of KVDB cluster, but rest of the nodes are potential internal kvdb nodes.
PX_METADATA_NODE_LABELenvironment variable takes precedence over the
KVDB Node Failover
If a Portworx node which was a part of the internal KVDB cluster goes down for more than 3 minutes, then any other available storage nodes which are not part of the KVDB cluster will try to join it.
This is a set of actions that are taken on the node which tries to join the KVDB cluster
- The down member is removed from the internal KVDB cluster. (Note: The node will be still part of the Portworx cluster)
- Internal KVDB is started on the new node
- The new node adds itself to the internal KVDB cluster
The node label logic holds true even during failover, so if a node has a label set to false, then it will not try to join the KVDB cluster.
If the offline node comes back up, it will NOT rejoin the internal KVDB cluster, once another node replaces it.
Portworx takes regular internal backups of its key-value space and dumps them into a file on the host.
Portworx takes these internal KVDB backups every 2 minutes and places them into the
/var/lib/osd/kvdb_backup directory on all nodes in your cluster. It also keeps a rolling count of 10 backups at a time within the internal KVDB storage drive.
A typical backup file will look like this
These backup files can be used for recovering the internal KVDB cluster in case of a disaster.
In an event of a cluster failure such as:
- All the internal KVDB nodes and their corresponding drives where their data resides are lost and are unrecoverable
- Quorum no. of internal KVDB nodes are lost and are unrecoverable.
Portworx can recover its internal KVDB cluster from one of the backup files that are dumped on each of the nodes.
Please reach out to Portworx support to recover your cluster from such a state.