Portworx requires a key-value database such as etcd for configuring storage. A highly available clustered etcd with persistent storage is preferred.
For production Portworx clusters Portworx, Inc. recommends the following configuration of an etcd cluster:
- Etcd Version > 3.1.x
- Minimum 3 nodes
- Minimum 8G of memory dedicated to each etcd node.
- Each Etcd node in the etcd cluster backed with storage disks (minimum 100GB)
More detailed set of hardware requirements as recommended by etcd can be found here
You can use one of the following methods to setup an etcd cluster
Setup ETCD cluster with static set of nodes
If you have 3 static nodes where you want to run etcd follow this guide to setup systemd services for an etcd cluster.
Setup ETCD cluster using CoreOS documentation
Follow this detailed step by step process provided by etcd to setup a brand new multi-node cluster.
Setup ETCD cluster using Ansible Playbook
Follow this ansible playbook to install a 3 node etcd cluster.
Etcd provides multiple knobs to fine tune the cluster based on your needs. Portworx, Inc. recommends fine tuning the following three settings.
etcd keeps an exact history of its keyspace, this history should be periodically compacted to avoid performance degradation and eventual storage space exhaustion. Regular compaction ensures that the memory usage of the etcd process is under check.
The keyspace can be compacted automatically with etcd’s time windowed history retention policy, or manually with
Portworx, Inc. recommends keeping history for last 3 hours. While setting up etcd you can specify the retention policy in the following way:
Database Size (Space Quota)
The space quota in etcd ensures the cluster operates in a reliable fashion. Without a space quota, etcd may suffer from poor performance if the keyspace grows excessively large, or it may simply run out of storage space, leading to unpredictable cluster behavior.
Portworx, Inc. recommends setting the space quota to max value of 8Gi. While setting up etcd you can specify the space quota in the following way:
Etcd can take periodic snapshots of its keyspace which can be used to restore the etcd cluster in case of a complete disaster. By default etcd takes a snapshot after every 10,000 changes to its key value space. If you want the snapshot strategy to be more aggressive you can tune the frequency in the following way: