Backup, load and restore CouchDB clusters
Project description
Create an archive of a running CouchDB node, saving CouchDB files data/.shards, data/_dbs.couch and data/shards in this order. To allow backup of a running CouchDB, files are copied before archive creation.
Restore an archive of a CouchDB node to a new CouchDB. The new CouchDB can be a cluster of multiple nodes. The new CouchDB configuration should already be done before using Couchcopy, however, all existing data will be deleted. During restoration, CouchDB will be stopped and restarted on each cluster nodes.
Limitations
Tested at least with CouchDB 3.1.1 and 3.3.3.
To restore an archive, Couchcopy needs to stop and start CouchDB. It assumes that CouchDB is controlled by systemd. If you don’t use systemd you can change parameters --couchdb-start and --couchdb-stop.
Your CouchDB n value should be higher or equal to the number of nodes in your CouchDB cluster. Otherwise saving shards from one node would not be enough to save and restore all databases. See CouchDB documentation for more details on replicas and nodes.
The number of shards per database, i.e. the value of q, should be the same for the origin CouchDB and the destination CouchDB. Otherwise, tree /data/shards is not the same.
Couchcopy assumes you have read and write permissions on CouchDB data directories. If you don’t have them, you can try to use the --use-sudo option.
Get started
Install Couchcopy:
pip install --user couchcopy
Make a backup to backup.tar.gz, from machine old-server with CouchDB data at /var/lib/couchdb:
couchcopy backup old-server,/var/lib/couchdb backup.tar.gz
Restore a backup backup.tar.gz to a 3-node CouchDB cluster where machines are accessible via SSH at cluster_vm1, cluster_vm2, cluster_vm3:
couchcopy restore backup.tar.gz admin:password@cluster_vm1,/var/lib/couchdb \
admin:password@cluster_vm2,/var/lib/couchdb \
admin:password@cluster_vm3,/var/lib/couchdb
Quickly access data from a backup, by spawning a CouchDB instance:
couchcopy load backup.tar.gz
Improve couchcopy load loading time by preconfiguring CouchDB metadata, so that the Updating CouchDB metadata... step is not needed:
couchcopy unbrand slow_backup.tar.gz quick_backup.tar.gz
For more options:
couchcopy -h
couchcopy backup -h
couchcopy unbrand -h
couchcopy load -h
couchcopy restore -h
On Fedora, CouchDB can be installed and configured with the following :
sudo dnf copr enable -y adrienverge/couchdb
sudo dnf install couchdb
sudo sh -c 'echo "admin = password" >> /etc/couchdb/local.ini'
sudo systemctl restart couchdb
If you work with remote machines, CouchDB needs to listen to remote IPs on each machine. You can enable it with the following (for security, revert it afterwards):
sudo sed -i 's/;bind_address = 127.0.0.1/bind_address = 0.0.0.0/g' /etc/couchdb/local.ini
Implementation details
During restoration, if the new CouchDB nodes names are not the same as the old CouchDB, nodes names are updated using CouchDB /_node/_local/_dbs endpoint. See CouchDB /_node/_local/_dbs endpoint documentation.
During restoration, Couchcopy first updates one CouchDB node metadata (i.e. the list of nodes names) then it lets CouchDB itself synchronize metadata to the other nodes. Couchcopy exits when the synchronization is finished for all nodes, using undocumented CouchDB /_dbs endpoint to monitor CouchDB nodes synchronization. You can skip that part if you want, i.e. you can exit Couchcopy safely when the following log trace is displayed [Waiting for CouchDB cluster synchronization...]. For a CouchDB of 10^5 databases, updating the first node metadata takes 35 minutes then metadata synchronization to the other nodes takes 6 minutes. For a CouchDB of 100 databases only, both operations are nearly instantaneous.
Developer notes
To speed up CouchDB nodes synchronization it is possible to:
Disable compaction daemon during synchronization (for 10^5 databases, nodes synchronization goes from 6 minutes down to 4 minutes).
Copy the saved _dbs.couch on every machine, but it sounds dangerous, it sounds better to let CouchDB rebuild these files itself (for 10^5 databases, nodes synchronization goes from 6 minutes down to 0 seconds).
Machines disk IOPS consumption is around 1200 IOPS during restoration.
Sometimes, the nodes synchronization, instead of taking 6 minutes for 10^5 databases, takes more than 3 hours. I wasn’t able to find the cause or eliminate this bad performance reliably. I advise using fast machines on the same local network, and disable compaction.
Interesting discussions on CouchDB:
On nodes renaming after data/* copy for a backup restoration:
On CouchDB cluster internal backfill for a backup restoration:
Unanswered questions about nodes renaming speed, and backup feasibility:
Build and publish
python setup.py sdist
twine upload dist/*
License
This program is licensed under the GNU General Public License version 3.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
File details
Details for the file couchcopy-0.2.4.tar.gz
.
File metadata
- Download URL: couchcopy-0.2.4.tar.gz
- Upload date:
- Size: 24.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.12.5
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | cd6d7ddd47b79133f4cca9bdc610afcbd489785f77a2500d3e72bde0b8779d58 |
|
MD5 | 5042e08f7df7fdcb08313157fe31bf6a |
|
BLAKE2b-256 | e01c4b79c8c9f0d7281af06afe303349471563ca1aad8f65854ce7b33cdbb892 |