The etcd distributed key-value store.
Etcd is a central component in Kubernetes. it is a distributed reliable key-value store and uses raft for consensus.
Intended to be included when starting or building a Kubernetes image.
Manual test:
./etcd.sh test start > $log
# On vm-001;
etcdctl member list
etcdctl put Hello World
etcdctl get Hello
etcdctl get "" --from-key
etcdctl get "" --prefix=true
etcdctl get --prefix=true '' -w fields | grep '"Key"'
The various etcd API versions are very confusing. Kubernetes seem to require;
ETCDCTL_API=3
which is set in the etcd overlay. The help text and options/parameters are different for different api versions;
ETCDCTL_API=2 etcdctl help
If the data-dir is lost or corrupt on a node it can never recover automatically(!). The node must be removed and re-added. This is cosidered to be perfecly ok and HA(!!).
It will happen on a xc scalein since the disk is re-initiated on
a subsequent scaleout. What will happen is that etcd bails out with
a message like;
2018-01-30 10:13:43.566760 C | etcdmain: member e7db4eebbdb94a11 has already been bootstrapped
Etcd is only started on vm 1-3 which allows one of these to fail but
not two (consensus is lost).