Etcd node migration
This procedure explains how to add a new etcd member for your StorageOS etcd cluster while removing one of the current members. This is useful when the nodes hosting etcd must be recycled.
It is assumed that the StorageOS etcd cluster is installed following the production etcd installation page, where etcd nodes are installed on their own machines.
It is also assumed that etcd members are referenced from Kubernetes using a External Service. Example available in the etcd external Service example. This service should be referred to in the
spec.kvbackend.address
section of your StorageOS CustomResource. If that Service is not used, a full restart of the StorageOS cluster will be required. The StorageOS CustomResource would need to be removed, and amended to reflect the new etcd urls created.
Preparation
Prepare the installation of etcd on a new node, making sure that etcd is not starting on that new node.
The steps for preparing an etcd node can be found in the etcd prerequisites page.
-
Back up etcd
$ export ETCDCTL_API=3 $ # Set all your endpoints $ export endpoints="192.168.174.117:2379,192.168.195.168:2379,192.168.174.117:2379" $ etcdctl --endpoints $endpoints snapshot save /var/tmp/etcd-snapshot.db
-
Verify etcd health
$ export ETCDCTL_API=3 $ # Set all your endpoints $ export endpoints="192.168.174.117:2379,192.168.195.168:2379,192.168.174.117:2379" $ etcdctl member list --endpoints $endpoints -wtable +------------------+---------+-----------------------+------------------------------+-----------------------------+------------+ | ID | STATUS | NAME | PEER ADDRS | CLIENT ADDRS | IS LEARNER | +------------------+---------+-----------------------+------------------------------+-----------------------------+------------+ | 7817aa073b059aab | started | etcd-192.168.195.168 | http://192.168.195.168:2380 | http://192.168.195.168:2379 | false | | e22cdd20a03e5e73 | started | etcd-192.168.202.40 | http://192.168.202.40:2380 | http://192.168.202.40:2379 | false | | e5d0f0e242014d3d | started | etcd-192.168.174.117 | http://192.168.174.117:2380 | http://192.168.174.117:2379 | false | +------------------+---------+-----------------------+------------------------------+-----------------------------+------------+ $ etcdctl endpoint health --endpoints $endpoints -wtable +---------------------+--------+------------+-------+ | ENDPOINT | HEALTH | TOOK | ERROR | +---------------------+--------+------------+-------+ |192.168.174.117:2379 | true | 5.048177ms | | |192.168.195.168:2379 | true | 5.926681ms | | |192.168.202.40:2379 | true | 5.526928ms | | +---------------------+--------+------------+-------+ $ etcdctl endpoint status --endpoints $endpoints -wtable +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ | ENDPOINT | ID | VERSION | DB SIZE | IS LEADER | IS LEARNER | RAFT TERM | RAFT INDEX | RAFT APPLIED INDEX | ERRORS | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ |192.168.174.117:2379 | e5d0f0e242014d3d | 3.4.9 | 311 kB | false | false | 2 | 4281 | 4281 | | |192.168.195.168:2379 | 7817aa073b059aab | 3.4.9 | 315 kB | false | false | 2 | 4281 | 4281 | | |192.168.202.40:2379 | e22cdd20a03e5e73 | 3.4.9 | 352 kB | true | false | 2 | 4281 | 4281 | | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+
Migration
In the following procedure NODE4 is a new member to add to the cluster, while NODE1 ought to be removed.
-
Amend etcd configuration to reference the new node (NODE4)
Make the following change on all running etcd members and NODE4
$ # NODE4_IP is the NEW_NODE_ADDRESS $ echo "NODE4_IP=192.168.152.142" >> /etc/etcd.conf $ # Check the contents of /etc/etcd.conf $ cat /etc/etcd.conf CLIENT_PORT=2379 PEERS_PORT=2380 # NODE_IP is the IP of the node where this file resides. NODE_IP=192.168.202.40 NODE1_IP=192.168.195.168 NODE2_IP=192.168.202.40 NODE3_IP=192.168.174.117 NODE4_IP=192.168.152.142
-
Amend SystemD service file on the new etcd node (NODE4)
The SystemD service file is expected in
/etc/systemd/system/etcd3.service
Change the
--initial-cluster-state
toexisting
and add the reference to NODE4 in the--initial-cluster
variable.vi /etc/systemd/system/etcd3.service
The resulting changes would appear as follows:
... ExecStart=/usr/local/sbin/etcd3 --name etcd-${NODE_IP} \ ... --initial-cluster-state existing \ ... --initial-cluster \ etcd-${NODE1_IP}=http://${NODE1_IP}:${PEERS_PORT},\ etcd-${NODE2_IP}=http://${NODE2_IP}:${PEERS_PORT},\ etcd-${NODE3_IP}=http://${NODE3_IP}:${PEERS_PORT},\ etcd-${NODE4_IP}=http://${NODE4_IP}:${PEERS_PORT} ...
Note the reference to NODE4 at the end of the
--initial-cluster
variableMake sure etcd is not started on the new member NODE4
-
Add etcd member as a
learner
# Set environment variable for the the new etcd member (NODE4) $ NODE4_IP=192.168.152.142 $ ETCD_NEW_MEMBER="etcd-${NODE4_IP}" $ ETCD_NEW_MEMBER_PEER="http://$NODE4_IP:2380" # Add the new member to the cluster $ export ETCDCTL_API=3 $ etcdctl member add \ --learner $ETCD_NEW_MEMBER \ --peer-urls="$ETCD_NEW_MEMBER_PEER" Member 52e5c9ac117b3df2 added to cluster b4f4ed717ea44b8d ETCD_NAME="etcd-192.168.152.142" ETCD_INITIAL_CLUSTER="etcd-192.168.152.142=http://192.168.152.142:2380,etcd-192.168.195.168=http://192.168.195.168:2380,etcd-192.168.202.40=http://192.168.202.40:2380,etcd-192.168.174.117=http://192.168.174.117:2380" ETCD_INITIAL_ADVERTISE_PEER_URLS="http://192.168.152.142:2380" ETCD_INITIAL_CLUSTER_STATE="existing"
-
Check the etcd members
$ export endpoints="192.168.174.117:2379,192.168.195.168:2379,192.168.174.117:2379" $ etcdctl member list --endpoints $endpoints -wtable +------------------+-----------+----------------------+-----------------------------+-----------------------------+------------+ | ID | STATUS | NAME | PEER ADDRS | CLIENT ADDRS | IS LEARNER | +------------------+-----------+----------------------+-----------------------------+-----------------------------+------------+ | 52e5c9ac117b3df2 | unstarted | | http://192.168.152.142:2380 | | true | | 7817aa073b059aab | started | etcd-192.168.195.168 | http://192.168.195.168:2380 | http://192.168.195.168:2379 | false | | e22cdd20a03e5e73 | started | etcd-192.168.202.40 | http://192.168.202.40:2380 | http://192.168.202.40:2379 | false | | e5d0f0e242014d3d | started | etcd-192.168.174.117 | http://192.168.174.117:2380 | http://192.168.174.117:2379 | false | +------------------+-----------+----------------------+-----------------------------+-----------------------------+------------+
Note that the learner is not started yet
-
Start etcd on the new node (NODE4)
Make sure that
/etc/systemd/system/etcd.service
only have currently active nodes specified in the--initial-cluster
flag.# On the new node (NODE4) systemctl daemon-reload systemctl enable etcd3.service systemctl start etcd3.service
-
Check the etcd members
$ export endpoints="192.168.174.117:2379,192.168.195.168:2379,192.168.174.117:2379" $ etcdctl member list --endpoints $endpoints -wtable +------------------+---------+----------------------+-----------------------------+-----------------------------+------------+ | ID | STATUS | NAME | PEER ADDRS | CLIENT ADDRS | IS LEARNER | +------------------+---------+----------------------+-----------------------------+-----------------------------+------------+ | 52e5c9ac117b3df2 | started | etcd-192.168.152.142 | http://192.168.152.142:2380 | http://192.168.152.142:2379 | true | | 7817aa073b059aab | started | etcd-192.168.195.168 | http://192.168.195.168:2380 | http://192.168.195.168:2379 | false | | e22cdd20a03e5e73 | started | etcd-192.168.202.40 | http://192.168.202.40:2380 | http://192.168.202.40:2379 | false | | e5d0f0e242014d3d | started | etcd-192.168.174.117 | http://192.168.174.117:2380 | http://192.168.174.117:2379 | false | +------------------+---------+----------------------+-----------------------------+-----------------------------+------------+
Note that the learner is started
-
Check that the new learner has the same revision applied as the current members
$ export ETCDCTL_API=3 # Added NODE4 in the endpoints variable $ export endpoints="192.168.174.117:2379,192.168.195.168:2379,192.168.174.117:2379,192.168.152.142:2379" etcdctl endpoint status --endpoints $endpoints -wtable +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ | ENDPOINT | ID | VERSION | DB SIZE | IS LEADER | IS LEARNER | RAFT TERM | RAFT INDEX | RAFT APPLIED INDEX | ERRORS | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ | 192.168.174.117:2379| e5d0f0e242014d3d | 3.4.9 | 352 kB | false | false | 2 | 24570 | 24570 | | | 192.168.195.168:2379| 7817aa073b059aab | 3.4.9 | 352 kB | false | false | 2 | 24570 | 24570 | | | 192.168.202.40:2379 | e22cdd20a03e5e73 | 3.4.9 | 352 kB | true | false | 2 | 24570 | 24570 | | | 192.168.152.142:2379| 52e5c9ac117b3df2 | 3.4.9 | 467 kB | false | true | 2 | 24570 | 24570 | | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+
Wait until the cluster has the learner ready, by ensuring that the RAFT TERM and RAFT INDEX of the learner node match the rest of the cluster.
-
Remove the node that needs to be evicted (NODE1)
Before promoting the learner to a full member, it is best to remove the node from the cluster that initially was selected to be decommissioned to avoid breaking quorum while having 4 nodes being part of the cluster. For more details, check the official etcd documentation regarding this topic.
$ export ETCDCTL_API=3 $ # Select member of id of the node to remove (NODE1) $ NODE1_MEMBER_ID=e22cdd20a03e5e73 $ etcdctl member remove $NODE1_MEMBER_ID Member e22cdd20a03e5e73 removed from cluster b4f4ed717ea44b8d
-
Promote the learner to a member
$ export ETCDCTL_API=3 $ # Select member of id of the node to remove (NODE1) $ NODE4_MEMBER_ID=52e5c9ac117b3df2 $ etcdctl member promote $NODE4_MEMBER_ID Member 52e5c9ac117b3df2 promoted in cluster b4f4ed717ea44b8d
The promotion will fail if the learner is not in sync with the leader member.
-
Check the etcd health
$ export endpoints=192.168.174.117:2379,192.168.195.168:2379,192.168.152.142:2379 $ etcdctl member list --endpoints $endpoints -wtable +------------------+---------+---------------------+----------------------------+----------------------------+------------+ | ID | STATUS | NAME | PEER ADDRS | CLIENT ADDRS | IS LEARNER | +------------------+---------+---------------------+----------------------------+----------------------------+------------+ | 52e5c9ac117b3df2 | started |etcd-192.168.152.142 |http://192.168.152.142:2380 |http://192.168.152.142:2379 | false | | 7817aa073b059aab | started |etcd-192.168.195.168 |http://192.168.195.168:2380 |http://192.168.195.168:2379 | false | | e5d0f0e242014d3d | started |etcd-192.168.174.117 |http://192.168.174.117:2380 |http://192.168.174.117:2379 | false | +------------------+---------+---------------------+----------------------------+----------------------------+------------+ $ etcdctl endpoint status --endpoints $endpoints -wtable +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ | ENDPOINT | ID | VERSION | DB SIZE | IS LEADER | IS LEARNER | RAFT TERM | RAFT INDEX | RAFT APPLIED INDEX | ERRORS | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+ |192.168.174.117:2379 | e5d0f0e242014d3d | 3.4.9 | 352 kB | false | false | 3 | 35939 | 35939 | | |192.168.195.168:2379 | 7817aa073b059aab | 3.4.9 | 352 kB | true | false | 3 | 35939 | 35939 | | |192.168.152.142:2379 | 52e5c9ac117b3df2 | 3.4.9 | 467 kB | false | false | 3 | 35939 | 35939 | | +---------------------+------------------+---------+---------+-----------+------------+-----------+------------+--------------------+--------+
Note that NODE4 is now a full quorum member, while NODE1 is no longer part of the cluster
-
Edit Endpoints referencing the Kubernetes Service
Remove the reference to NODE1 and add the IP for NODE4
$ kubectl edit -n storageos-etcd endpoints/storageos-etcd
-
Make amendments in the SystemD configuration files removing any reference to NODE1
It is not required to restart the etcd service, but to keep the service file up to date.