Because of CockroachDB's multi-active availability design, you can perform a "rolling upgrade" of your CockroachDB cluster. This means that you can upgrade nodes one at a time without interrupting the cluster's overall health and operations.
Step 1. Verify that you can upgrade
Run cockroach sql
against any node in the cluster to open the SQL shell. Then check your current cluster version:
> SHOW CLUSTER SETTING version;
To upgrade to a new version, you must first be on a production release of the previous version. The release does not need to be the latest production release of the previous version, but it must be a production release rather than a testing release (alpha/beta).
Therefore, if you are upgrading from v19.2 to v20.2, or from a testing release (alpha/beta) of v20.1 to v20.2:
First upgrade to a production release of v20.1. Be sure to complete all the steps.
Then return to this page and perform a second rolling upgrade to v20.2.
If you are upgrading from any production release of v20.1, or from any earlier v20.2 release, you do not have to go through intermediate releases; continue to step 2.
Step 2. Prepare to upgrade
Before starting the upgrade, complete the following steps.
Check load balancing
Make sure your cluster is behind a load balancer, or your clients are configured to talk to multiple nodes. If your application communicates with a single node, stopping that node to upgrade its CockroachDB binary will cause your application to fail.
Check cluster health
Verify the overall health of your cluster using the DB Console. On the Overview:
Under Node Status, make sure all nodes that should be live are listed as such. If any nodes are unexpectedly listed as suspect or dead, identify why the nodes are offline and either restart them or decommission them before beginning your upgrade. If there are dead and non-decommissioned nodes in your cluster, it will not be possible to finalize the upgrade (either automatically or manually).
Under Replication Status, make sure there are 0 under-replicated and unavailable ranges. Otherwise, performing a rolling upgrade increases the risk that ranges will lose a majority of their replicas and cause cluster unavailability. Therefore, it's important to identify and resolve the cause of range under-replication and/or unavailability before beginning your upgrade.
In the Node List:
- Make sure all nodes are on the same version. If any nodes are behind, upgrade them to the cluster's current version first, and then start this process over.
- Make sure capacity and memory usage are reasonable for each node. Nodes must be able to tolerate some increase in case the new version uses more resources for your workload. Also go to Metrics > Dashboard: Hardware and make sure CPU percent is reasonable across the cluster. If there's not enough headroom on any of these metrics, consider adding nodes to your cluster before beginning your upgrade.
Review breaking changes
Review the backward-incompatible changes in v20.2 and deprecated features. If any affect your application, make the necessary changes.
Step 3. Decide how the upgrade will be finalized
This step is relevant only when upgrading from v20.1.x to v20.2. For upgrades within the v20.2.x series, skip this step.
By default, after all nodes are running the new version, the upgrade process will be auto-finalized. This will enable certain features and performance improvements introduced in v20.2. However, it will no longer be possible to perform a downgrade to v20.1. In the event of a catastrophic failure or corruption, the only option will be to start a new cluster using the old binary and then restore from one of the backups created prior to performing the upgrade. For this reason, we recommend disabling auto-finalization so you can monitor the stability and performance of the upgraded cluster before finalizing the upgrade, but note that you will need to follow all of the subsequent directions, including the manual finalization in step 5:
Upgrade to v20.1, if you haven't already.
Start the
cockroach sql
shell against any node in the cluster.Set the
cluster.preserve_downgrade_option
cluster setting:> SET CLUSTER SETTING cluster.preserve_downgrade_option = '20.1';
It is only possible to set this setting to the current cluster version.
Features that require upgrade finalization
When upgrading from v20.1 to v20.2, certain features and performance improvements will be enabled only after finalizing the upgrade, including but not limited to:
Spatial features: After finalization, it will be possible to use spatial indexes, and spatial functions, as well as the ability to migrate spatial data from various formats such as Shapefiles, GeoJSON, GeoPackages, and OpenStreetMap.
ENUM
data types: After finalization, it will be possible to create and manage user-definedENUM
data types consisting of sets of enumerated, static values.Altering column data types: After finalization, it will be possible to alter column data types where column data must be rewritten.
User-defined schemas: After finalization, it will be possible to create user-defined logical schemas, as well alter user-defined schemas, drop user-defined schemas, set the schema for a database object, and convert databases to user-defined schemas. For details on migrating a cluster that does not use user-defined schemas in its naming hierarchy, see Migrating namespaces from previous versions of CockroachDB.
Foreign key index requirement: After finalization, it will no longer be required to have an index on the referencing columns of a
FOREIGN KEY
constraint.Minimum password length: After finalization, it will be possible to use the
server.user_login.min_password_length
cluster setting to set a minimum length for passwords.Materialized views: After finalization, it will be possible to create materialized views, or views that store their selection query results on-disk.
CREATELOGIN
privilege: After finalization, theCREATELOGIN
privilege will be required to define or change authentication principals or their credentials.
Step 4. Perform the rolling upgrade
For each node in your cluster, complete the following steps. Be sure to upgrade only one node at a time, and wait at least one minute after a node rejoins the cluster to upgrade the next node. Simultaneously upgrading more than one node increases the risk that ranges will lose a majority of their replicas and cause cluster unavailability.
We recommend creating scripts to perform these steps instead of performing them manually. Also, if you are running CockroachDB on Kubernetes, see our documentation on single-cluster and/or multi-cluster orchestrated deployments for upgrade guidance instead.
Drain and stop the node using one of the following methods:
- If the node was started with a process manager, gracefully stop the node by sending
SIGTERM
with the process manager. If the node is not shutting down after 1 minute, sendSIGKILL
to terminate the process. When usingsystemd
, for example, setTimeoutStopSec=60
in your configuration template and runsystemctl stop <systemd config filename>
to stop the node withoutsystemd
restarting it. - If the node was started using
cockroach start
and is running in the foreground, pressctrl-c
in the terminal. - If the node was started using
cockroach start
and the--background
and--pid-file
flags, runkill <pid>
, where<pid>
is the process ID of the node.
Note:The amount of time you should wait before sending
SIGKILL
can vary depending on your cluster configuration and workload, which affects how long it takes your nodes to complete a graceful shutdown. In certain edge cases, forcefully terminating the process before the node has completed shutdown can result in temporary data unavailability, latency spikes, uncertainty errors, ambiguous commit errors, or query timeouts. If you need maximum cluster availability, you can runcockroach node drain
prior to node shutdown and actively monitor the draining process instead of automating it.Verify that the process has stopped:
$ ps aux | grep cockroach
Alternately, you can check the node's logs for the message
server drained and shutdown completed
.- If the node was started with a process manager, gracefully stop the node by sending
Download and install the CockroachDB binary you want to use:
$ curl https://binaries.cockroachdb.com/cockroach-v20.2.19.darwin-10.9-amd64.tgz|tar -xzf -
$ curl https://binaries.cockroachdb.com/cockroach-v20.2.19.linux-amd64.tgz|tar -xzf -
If you use
cockroach
in your$PATH
, rename the outdatedcockroach
binary, and then move the new one into its place:i="$(which cockroach)"; mv "$i" "$i"_old
$ cp -i cockroach-v20.2.19.darwin-10.9-amd64/cockroach /usr/local/bin/cockroach
i="$(which cockroach)"; mv "$i" "$i"_old
$ cp -i cockroach-v20.2.19.linux-amd64/cockroach /usr/local/bin/cockroach
Start the node to have it rejoin the cluster.
Warning:For maximum availability, do not wait more than a few minutes before restarting the node with the new binary. See this open issue for context.
Without a process manager like
systemd
, re-run thecockroach start
command that you used to start the node initially, for example:$ cockroach start \ --certs-dir=certs \ --advertise-addr=<node address> \ --join=<node1 address>,<node2 address>,<node3 address>
If you are using
systemd
as the process manager, run this command to start the node:$ systemctl start <systemd config filename>
Verify the node has rejoined the cluster through its output to
stdout
or through the DB Console.If you use
cockroach
in your$PATH
, you can remove the old binary:$ rm /usr/local/bin/cockroach_old
If you leave versioned binaries on your servers, you do not need to do anything.
After the node has rejoined the cluster, ensure that the node is ready to accept a SQL connection.
Unless there are tens of thousands of ranges on the node, it's usually sufficient to wait one minute. To be certain that the node is ready, run the following command:
cockroach sql -e 'select 1'
The command will automatically wait to complete until the node is ready.
Repeat these steps for the next node.
Step 5. Finish the upgrade
This step is relevant only when upgrading from v20.1.x to v20.2. For upgrades within the v20.2.x series, skip this step.
If you disabled auto-finalization in step 3, monitor the stability and performance of your cluster for as long as you require to feel comfortable with the upgrade (generally at least a day). If during this time you decide to roll back the upgrade, repeat the rolling restart procedure with the old binary.
Once you are satisfied with the new version:
Start the
cockroach sql
shell against any node in the cluster.Re-enable auto-finalization:
> RESET CLUSTER SETTING cluster.preserve_downgrade_option;
Note:This statement can take up to a minute to complete, depending on the amount of data in the cluster, as it kicks off various internal maintenance and migration tasks. During this time, the cluster will experience a small amount of additional load.
Check the cluster version to confirm that the finalize step has completed:
> SHOW CLUSTER SETTING version;
If you intend to continue upgrading to v21.1 or later, you must ensure that any previously decommissioned nodes are fully decommissioned. Otherwise, they will block the upgrade. For instructions, see Check decommissioned nodes.
Check decommissioned nodes
Check the membership
field in the output of cockroach node status --decommission
. Nodes with decommissioned
membership are fully decommissioned, while nodes with decommissioning
membership have not completed the process. If there are decommissioning
nodes in your cluster, this will block the upgrade.
Before upgrading from v20.2 to v21.1, you must manually change the status of any decommissioning
nodes to decommissioned
. To do this, run cockroach node decommission
on these nodes and confirm that they update to decommissioned
.
In case a decommissioning process is hung, recommission and then decommission those nodes again, and confirm that they update to decommissioned
.
Troubleshooting
After the upgrade has finalized (whether manually or automatically), it is no longer possible to downgrade to the previous release. If you are experiencing problems, we therefore recommend that you:
Run the
cockroach debug zip
command against any node in the cluster to capture your cluster's state.Reach out for support from Cockroach Labs, sharing your debug zip.
In the event of catastrophic failure or corruption, the only option will be to start a new cluster using the old binary and then restore from one of the backups created prior to performing the upgrade.