Skip to content

OVN DB Backup and Recovery

This document describes how to perform database backups and how to perform cluster recovery from existing database files in different situations.

Database Backup

The database files can be backed up for recovery in case of failure. Use the backup command of the kubectl plugin:

# kubectl ko nb backup
tar: Removing leading `/' from member names
backup ovn-nb db to /root/ovnnb_db.060223191654183154.backup

# kubectl ko sb backup
tar: Removing leading `/' from member names
backup ovn-nb db to /root/ovnsb_db.060223191654183154.backup

Cluster Partial Nodes Failure Recovery

If some nodes in the cluster are working abnormally due to power failure, file system failure or lack of disk space, but the cluster is still working normally, you can recover it by following the steps below.

Check the Logs to Confirm Status

Check the log in /var/log/ovn/ovn-northd.log, if it shows similar error as follows, you can make sue that there is an exception in the database:

 * ovn-northd is not running
ovsdb-server: ovsdb error: error reading record 2739 from OVN_Northbound log: record 2739 advances commit index to 6308 but last log index is 6307
 * Starting ovsdb-nb

Kick Node from Cluster

Select the corresponding database for the operation based on whether the log prompt is OVN_Northbound or OVN_Southbound. The above log prompt is OVN_Northbound then for ovn-nb do the following:

# kubectl ko nb status
9182
Name: OVN_Northbound
Cluster ID: e75f (e75fa340-49ed-45ab-990e-26cb865ebc85)
Server ID: 9182 (9182e8dd-b5b0-4dd8-8518-598cc1e374f3)
Address: tcp:[10.0.128.61]:6643
Status: cluster member
Role: leader
Term: 1454
Leader: self
Vote: self

Last Election started 1732603 ms ago, reason: timeout
Last Election won: 1732587 ms ago
Election timer: 1000
Log: [7332, 12512]
Entries not yet committed: 1
Entries not yet applied: 1
Connections: ->f080 <-f080 <-e631 ->e631
Disconnections: 1
Servers:
    f080 (f080 at tcp:[10.0.129.139]:6643) next_index=12512 match_index=12510 last msg 63 ms ago
    9182 (9182 at tcp:[10.0.128.61]:6643) (self) next_index=10394 match_index=12510
    e631 (e631 at tcp:[10.0.131.173]:6643) next_index=12512 match_index=0

Kick abnormal nodes from the cluster:

kubectl ko nb kick e631

Log in to the abnormal node and delete the database file:

mv /etc/origin/ovn/ovnnb_db.db /tmp

Delete the ovn-central pod of the corresponding node and wait for the cluster to recover:

kubectl delete pod -n kube-system ovn-central-xxxx

Recover when Total Cluster Failed

If the majority of the cluster nodes are broken and the leader cannot be elected, please refer to the following steps to recover.

Stop ovn-central

Record the current replicas of ovn-central and stop ovn-central to avoid new database changes that affect recovery:

kubectl scale deployment -n kube-system ovn-central --replicas=0

Select a Backup

As most of the nodes are damaged, the cluster needs to be rebuilt by recovering from one of the database files. If you have previously backed up the database you can use the previous backup file to restore it. If not you can use the following steps to generate a backup from an existing file.

Since the database file in the default folder is a cluster format database file containing information about the current cluster, you can't rebuild the database directly with this file, you need to use ovsdb-tool cluster-to-standalone to convert the format.

Select the first node in the ovn-central environment variable NODE_IPS to restore the database files. If the database file of the first node is corrupted, copy the file from the other machine /etc/origin/ovn to the first machine. Run the following command to generate a database file backup.

docker run -it -v /etc/origin/ovn:/etc/ovn kubeovn/kube-ovn:v1.12.16 bash
cd /etc/ovn/
ovsdb-tool cluster-to-standalone ovnnb_db_standalone.db ovnnb_db.db
ovsdb-tool cluster-to-standalone ovnsb_db_standalone.db ovnsb_db.db

Delete the Database Files on All ovn-central Nodes

In order to avoid rebuilding the cluster with the wrong data, the existing database files need to be cleaned up:

mv /etc/origin/ovn/ovnnb_db.db /tmp
mv /etc/origin/ovn/ovnsb_db.db /tmp

Recovering Database Cluster

Rename the backup databases to ovnnb_db.db and ovnsb_db.db respectively, and copy them to the /etc/origin/ovn/ directory of the first machine in the ovn-central environment variable NODE_IPS

mv /etc/origin/ovn/ovnnb_db_standalone.db /etc/origin/ovn/ovnnb_db.db
mv /etc/origin/ovn/ovnsb_db_standalone.db /etc/origin/ovn/ovnsb_db.db

Restore the number of replicas of ovn-central

kubectl scale deployment -n kube-system ovn-central --replicas=3
kubectl rollout status deployment/ovn-central -n kube-system

微信群 Slack Twitter Support Meeting

Comments