Deploy CockroachDB on Google Cloud Platform GCE

This page shows you how to manually deploy a secure multi-node CockroachDB cluster on Google Cloud Platform's Compute Engine (GCE), using Google's TCP Proxy Load Balancing service to distribute client traffic.

If you are only testing CockroachDB, or you are not concerned with protecting network communication with TLS encryption, you can use an insecure cluster instead. Select Insecure above for instructions.

Requirements

• You must have CockroachDB installed locally. This is necessary for generating and managing your deployment's certificates.

• You must have SSH access to each machine. This is necessary for distributing and starting CockroachDB binaries.

• Your network configuration must allow TCP communication on the following ports:
  • 26257 for intra-cluster and client-cluster communication
  • 8080 to expose your Admin UI

Recommendations

• If you plan to use CockroachDB in production, carefully review the Production Checklist.

• Decide how you want to access your Admin UI:

  Access Level	Description
  Partially open	Set a firewall rule to allow only specific IP addresses to communicate on port 8080.
  Completely open	Set a firewall rule to allow all IP addresses to communicate on port 8080.
  Completely closed	Set a firewall rule to disallow all communication on port 8080. In this case, a machine with SSH access to a node could use an SSH tunnel to access the Admin UI.

Step 1. Configure your network

CockroachDB requires TCP communication on two ports:

• 26257 (tcp:26257) for inter-node communication (i.e., working as a cluster)
• 8080 (tcp:8080) for exposing your Admin UI

Inter-node communication works by default using your GCE instances' internal IP addresses, which allow communication with other instances on CockroachDB's default port 26257. However, to expose your Admin UI and allow traffic from the TCP proxy load balancer and health checker to your instances, you need to create firewall rules for your project.

Creating firewall rules

When creating firewall rules, we recommend using Google Cloud Platform's tag feature, which lets you specify that you want to apply the rule only to instance that include the same tag.

Admin UI

Application data

Applications will not connect directly to your CockroachDB nodes. Instead, they'll connect to GCE's TCP Proxy Load Balancing service, which automatically routes traffic to the instances that are closest to the user. Because this service is implemented at the edge of the Google Cloud, you'll need to create a firewall rule to allow traffic from the load balancer and health checker to your instances. This is covered in Step 4.

{{site.data.alerts.callout_danger}}When using TCP Proxy Load Balancing, you cannot use firewall rules to control access to the load balancer. If you need such control, consider using Network TCP Load Balancing instead, but note that it cannot be used across regions. You might also consider using the HAProxy load balancer (see the On-Premises tutorial for guidance).{{site.data.alerts.end}}

Step 2. Create instances

Create an instance for each node you plan to have in your cluster. If you plan to run a sample workload against the cluster, create a separate instance for that workload.

• Run at least 3 nodes to ensure survivability.

• Use n1-standard or n1-highcpu predefined VMs, or custom VMs, with Local SSDs or SSD persistent disks. For example, Cockroach Labs has used custom VMs (8 vCPUs and 16 GiB of RAM per VM) for internal testing.

• Do not use f1 or g1 shared-core machines, which limit the load on a single core.

• If you used a tag for your firewall rules, when you create the instance, select Management, disk, networking, SSH keys. Then on the Networking tab, in the Network tags field, enter cockroachdb.

For more details, see Hardware Recommendations and Cluster Topology.

Step 3. Synchronize clocks

CockroachDB requires moderate levels of clock synchronization to preserve data consistency. For this reason, when a node detects that its clock is out of sync with at least half of the other nodes in the cluster by 80% of the maximum offset allowed (500ms by default), it spontaneously shuts down. This avoids the risk of consistency anomalies, but it's best to prevent clocks from drifting too far in the first place by running clock synchronization software on each node.

ntpd should keep offsets in the single-digit milliseconds, so that software is featured here, but other methods of clock synchronization are suitable as well.

1. SSH to the first machine.

2. Disable timesyncd, which tends to be active by default on some Linux distributions:

   $ sudo timedatectl set-ntp no
   

   Verify that timesyncd is off:

   $ timedatectl
   

   Look for Network time on: no or NTP enabled: no in the output.

3. Install the ntp package:

   $ sudo apt-get install ntp
   

4. Stop the NTP daemon:

   $ sudo service ntp stop
   

5. Sync the machine's clock with Google's NTP service:

   $ sudo ntpd -b time.google.com
   

   To make this change permanent, in the /etc/ntp.conf file, remove or comment out any lines starting with server or pool and add the following lines:

   server time1.google.com iburst
   server time2.google.com iburst
   server time3.google.com iburst
   server time4.google.com iburst
   

   Restart the NTP daemon:

   $ sudo service ntp start
   

   {{site.data.alerts.callout_info}}We recommend Google's external NTP service because they handle "smearing" the leap second. If you use a different NTP service that doesn't smear the leap second, you must configure client-side smearing manually and do so in the same way on each machine.{{site.data.alerts.end}}

6. Verify that the machine is using a Google NTP server:

   $ sudo ntpq -p
   

   The active NTP server will be marked with an asterisk.

7. Repeat these steps for each machine where a CockroachDB node will run.

Compute Engine instances are preconfigured to use NTP, which should keep offsets in the single-digit milliseconds. However, Google can’t predict how external NTP services, such as pool.ntp.org, will handle the leap second. Therefore, you should:

• Configure each GCE instances to use Google's internal NTP service.
• If you plan to run a hybrid cluster across GCE and other cloud providers or environments, configure the non-GCE machines to use Google's external NTP service.

Amazon provides the Amazon Time Sync Service, which uses a fleet of satellite-connected and atomic reference clocks in each AWS Region to deliver accurate current time readings. The service also smears the leap second.

• If you plan to run your entire cluster on AWS, configure each AWS instance to use the internal Amazon Time Sync Service.
• However, if you plan to run a hybrid cluster across AWS and other cloud providers or environments, configure all machines to use Google's external NTP service, which is comparably accurate and also handles "smearing" the leap second.

ntpd should keep offsets in the single-digit milliseconds, so that software is featured here. However, to run ntpd properly on Azure VMs, it's necessary to first unbind the Time Synchronization device used by the Hyper-V technology running Azure VMs; this device aims to synchronize time between the VM and its host operating system but has been known to cause problems.

1. SSH to the first machine.

2. Find the ID of the Hyper-V Time Synchronization device:

   $ curl -O https://raw.githubusercontent.com/torvalds/linux/master/tools/hv/lsvmbus
   

   $ python lsvmbus -vv | grep -w "Time Synchronization" -A 3
   

   VMBUS ID 12: Class_ID = {9527e630-d0ae-497b-adce-e80ab0175caf} - [Time Synchronization]
       Device_ID = {2dd1ce17-079e-403c-b352-a1921ee207ee}
       Sysfs path: /sys/bus/vmbus/devices/2dd1ce17-079e-403c-b352-a1921ee207ee
       Rel_ID=12, target_cpu=0
   

3. Unbind the device, using the Device_ID from the previous command's output:

   $ echo <DEVICE_ID> | sudo tee /sys/bus/vmbus/drivers/hv_util/unbind
   

4. Install the ntp package:

   $ sudo apt-get install ntp
   

5. Stop the NTP daemon:

   $ sudo service ntp stop
   

6. Sync the machine's clock with Google's NTP service:

   $ sudo ntpd -b time.google.com
   

   To make this change permanent, in the /etc/ntp.conf file, remove or comment out any lines starting with server or pool and add the following lines:

   server time1.google.com iburst
   server time2.google.com iburst
   server time3.google.com iburst
   server time4.google.com iburst
   

   Restart the NTP daemon:

   $ sudo service ntp start
   

   {{site.data.alerts.callout_info}}We recommend Google's NTP service because they handle "smearing" the leap second. If you use a different NTP service that doesn't smear the leap second, be sure to configure client-side smearing in the same way on each machine.{{site.data.alerts.end}}

7. Verify that the machine is using a Google NTP server:

   $ sudo ntpq -p
   

   The active NTP server will be marked with an asterisk.

8. Repeat these steps for each machine where a CockroachDB node will run.

Step 4. Set up TCP Proxy Load Balancing

Each CockroachDB node is an equally suitable SQL gateway to your cluster, but to ensure client performance and reliability, it's important to use load balancing:

• Performance: Load balancers spread client traffic across nodes. This prevents any one node from being overwhelmed by requests and improves overall cluster performance (queries per second).

• Reliability: Load balancers decouple client health from the health of a single CockroachDB node. In cases where a node fails, the load balancer redirects client traffic to available nodes.

GCE offers fully-managed TCP Proxy Load Balancing. This service lets you use a single IP address for all users around the world, automatically routing traffic to the instances that are closest to the user.

{{site.data.alerts.callout_danger}}When using TCP Proxy Load Balancing, you cannot use firewall rules to control access to the load balancer. If you need such control, consider using Network TCP Load Balancing instead, but note that it cannot be used across regions. You might also consider using the HAProxy load balancer (see the On-Premises tutorial for guidance).{{site.data.alerts.end}}

To use GCE's TCP Proxy Load Balancing service:

1. For each zone in which you're running an instance, create a distinct instance group.
   • To ensure that the load balancer knows where to direct traffic, specify a port name mapping, with tcp26257 as the Port name and 26257 as the Port number.
2. Add the relevant instances to each instance group.
3. Configure TCP Proxy Load Balancing.
   • During backend configuration, create a health check, setting the Protocol to HTTPS, the Port to 8080, and the Request path to path /health?ready=1. This health endpoint ensures that load balancers do not direct traffic to nodes that are live but not ready to receive requests.
     • If you want to maintain long-lived SQL connections that may be idle for more than tens of seconds, increase the backend timeout setting accordingly.
   • During frontend configuration, reserve a static IP address and note the IP address and the port you select. You'll use this address and port for all client connections.
4. Create a firewall rule to allow traffic from the load balancer and health checker to your instances. This is necessary because TCP Proxy Load Balancing is implemented at the edge of the Google Cloud.
   • Be sure to set Source IP ranges to 130.211.0.0/22 and 35.191.0.0/16 and set Target tags to cockroachdb (not to the value specified in the linked instructions).

Step 5. Generate certificates

You can use either cockroach cert commands or openssl commands to generate security certificates. This section features the cockroach cert commands.

Locally, you'll need to create the following certificates and keys:

• A certificate authority (CA) key pair (ca.crt and ca.key).
• A node key pair for each node, issued to its IP addresses and any common names the machine uses, as well as to the IP addresses and common names for machines running load balancers.
• A client key pair for the root user. You'll use this to run a sample workload against the cluster as well as some cockroach client commands from your local machine.

{{site.data.alerts.callout_success}}Before beginning, it's useful to collect each of your machine's internal and external IP addresses, as well as any server names you want to issue certificates for.{{site.data.alerts.end}}

1. Install CockroachDB on your local machine, if you haven't already.

2. Create two directories:

   $ mkdir certs
   

   $ mkdir my-safe-directory
   

   • certs: You'll generate your CA certificate and all node and client certificates and keys in this directory and then upload some of the files to your nodes.
   • my-safe-directory: You'll generate your CA key in this directory and then reference the key when generating node and client certificates. After that, you'll keep the key safe and secret; you will not upload it to your nodes.

3. Create the CA certificate and key:

   	$ cockroach cert create-ca \
   	--certs-dir=certs \
   	--ca-key=my-safe-directory/ca.key
   

4. Create the certificate and key for the first node, issued to all common names you might use to refer to the node as well as to the load balancer instances:

   	$ cockroach cert create-node \
   	<node1 internal IP address> \
   	<node1 external IP address> \
   	<node1 hostname>  \
   	<other common names for node1> \
   	localhost \
   	127.0.0.1 \
   	<load balancer IP address> \
   	<load balancer hostname>  \
   	<other common names for load balancer instances> \
   	--certs-dir=certs \
   	--ca-key=my-safe-directory/ca.key
   

5. Upload certificates to the first node:

   	# Create the certs directory:
   	$ ssh <username>@<node1 address> "mkdir certs"
   

   	# Upload the CA certificate and node certificate and key:
   	$ scp certs/ca.crt \
   	certs/node.crt \
   	certs/node.key \
   	<username>@<node1 address>:~/certs
   

6. Delete the local copy of the node certificate and key:

   $ rm certs/node.crt certs/node.key
   

   {{site.data.alerts.callout_info}}This is necessary because the certificates and keys for additional nodes will also be named node.crt and node.key As an alternative to deleting these files, you can run the next cockroach cert create-node commands with the --overwrite flag.{{site.data.alerts.end}}

7. Create the certificate and key for the second node, issued to all common names you might use to refer to the node as well as to the load balancer instances:

   	$ cockroach cert create-node \
   	<node2 internal IP address> \
   	<node2 external IP address> \
   	<node2 hostname>  \
   	<other common names for node2> \
   	localhost \
   	127.0.0.1 \
   	<load balancer IP address> \
   	<load balancer hostname>  \
   	<other common names for load balancer instances> \
   	--certs-dir=certs \
   	--ca-key=my-safe-directory/ca.key
   

8. Upload certificates to the second node:

   	# Create the certs directory:
   	$ ssh <username>@<node2 address> "mkdir certs"
   

   	# Upload the CA certificate and node certificate and key:
   	$ scp certs/ca.crt \
   	certs/node.crt \
   	certs/node.key \
   	<username>@<node2 address>:~/certs
   

9. Repeat steps 6 - 8 for each additional node.

10. Create a client certificate and key for the root user:

    $ cockroach cert create-client \
    root \
    --certs-dir=certs \
    --ca-key=my-safe-directory/ca.key
    

11. Upload certificates to the machine where you will run a sample workload:

    # Create the certs directory:
    $ ssh <username>@<workload address> "mkdir certs"
    

    # Upload the CA certificate and client certificate and key:
    $ scp certs/ca.crt \
    certs/client.root.crt \
    certs/client.root.key \
    <username>@<workload address>:~/certs
    

    In later steps, you'll also use the root user's certificate to run cockroach client commands from your local machine. If you might also want to run cockroach client commands directly on a node (e.g., for local debugging), you'll need to copy the root user's certificate and key to that node as well.

{{site.data.alerts.callout_info}} On accessing the Admin UI in a later step, your browser will consider the CockroachDB-created certificate invalid and you’ll need to click through a warning message to get to the UI. You can avoid this issue by using a certificate issued by a public CA. {{site.data.alerts.end}}

Step 6. Start nodes

You can start the nodes manually or automate the process using systemd.

Step 7. Initialize the cluster

On your local machine, run the cockroach init command to complete the node startup process and have them join together as a cluster:

$ cockroach init --certs-dir=certs --host=<address of any node>

After running this command, each node prints helpful details to the standard output, such as the CockroachDB version, the URL for the admin UI, and the SQL URL for clients.

Step 8. Test the cluster

CockroachDB replicates and distributes data for you behind-the-scenes and uses a Gossip protocol to enable each node to locate data across the cluster.

To test this, use the built-in SQL client locally as follows:

1. On your local machine, launch the built-in SQL client:

   $ cockroach sql --certs-dir=certs --host=<address of any node>
   

2. Create a securenodetest database:

   > CREATE DATABASE securenodetest;
   

3. Use \q to exit the SQL shell.

4. Launch the built-in SQL client against a different node:

   $ cockroach sql --certs-dir=certs --host=<address of different node>
   

5. View the cluster's databases, which will include securenodetest:

   > SHOW DATABASES;
   

   +--------------------+
   |      Database      |
   +--------------------+
   | crdb_internal      |
   | information_schema |
   | securenodetest     |
   | pg_catalog         |
   | system             |
   +--------------------+
   (5 rows)
   

6. Use \q to exit the SQL shell.

Step 9. Run a sample workload

CockroachDB offers a pre-built workload binary for Linux that includes several load generators for simulating client traffic against your cluster. This step features CockroachDB's version of the TPC-C workload.

{{site.data.alerts.callout_success}}For comprehensive guidance on benchmarking CockroachDB with TPC-C, see our Performance Benchmarking white paper.{{site.data.alerts.end}}

1. SSH to the machine where you want to run the sample TPC-C workload.

   This should be a machine that is not running a CockroachDB node, and it should already have a certs directory containing ca.crt, client.root.crt, and client.root.key files.

2. Download workload and make it executable:

   $ wget https://edge-binaries.cockroachdb.com/cockroach/workload.LATEST ; chmod 755 workload.LATEST
   

3. Rename and copy workload into the PATH:

   $ cp -i workload.LATEST /usr/local/bin/workload
   

4. Start the TPC-C workload, pointing it at the IP address of the load balancer and the location of the ca.crt, client.root.crt, and client.root.key files:

   $ workload run tpcc \
   --drop \
   --init \
   --duration=20m \
   --tolerate-errors \
   "postgresql://root@<IP ADDRESS OF LOAD BALANCER:26257/tpcc?sslmode=verify-full&sslrootcert=certs/ca.crt&sslcert=certs/client.root.crt&sslkey=certs/client.root.key"
   

   This command runs the TPC-C workload against the cluster for 20 minutes, loading 1 "warehouse" of data initially and then issuing about 12 queries per minute via 10 "worker" threads. These workers share SQL connections since individual workers are idle for long periods of time between queries.

   {{site.data.alerts.callout_success}}For more tpcc options, use workload run tpcc --help. For details about other load generators included in workload, use workload run --help.

5. To monitor the load generator's progress, open the Admin UI by pointing a browser to the address in the admin field in the standard output of any node on startup.

   For each user who should have access to the Admin UI for a secure cluster, create a user with a password. On accessing the Admin UI, the users will see a Login screen, where they will need to enter their usernames and passwords.

   Since the load generator is pointed at the load balancer, the connections will be evenly distributed across nodes. To verify this, click Metrics on the left, select the SQL dashboard, and then check the SQL Connections graph. You can use the Graph menu to filter the graph for specific nodes.

Step 10. Set up monitoring and alerting

Despite CockroachDB's various built-in safeguards against failure, it is critical to actively monitor the overall health and performance of a cluster running in production and to create alerting rules that promptly send notifications when there are events that require investigation or intervention.

For details about available monitoring options and the most important events and metrics to alert on, see Monitoring and Alerting.

Step 11. Scale the cluster

You can start the nodes manually or automate the process using systemd.

Step 12. Use the database

Now that your deployment is working, you can:

1. Implement your data model.
2. Create users and grant them privileges.
3. Connect your application. Be sure to connect your application to the load balancer, not to a CockroachDB node.

You may also want to adjust the way the cluster replicates data. For example, by default, a multi-node cluster replicates all data 3 times; you can change this replication factor or create additional rules for replicating individual databases and tables differently. For more information, see Configure Replication Zones.

{{site.data.alerts.callout_danger}} When running a cluster of 5 nodes or more, it's safest to increase the replication factor for important internal data to 5, even if you do not do so for user data. For the cluster as a whole to remain available, the ranges for this internal data must always retain a majority of their replicas. {{site.data.alerts.end}}

See also

• Production Checklist
• Manual Deployment
• Orchestrated Deployment
• Monitoring and Alerting
• Performance Benchmarking
• Performance Tuning
• Test Deployment
• Local Deployment