Scaling Coder

We scale-test Coder with a built-in utility that can be used in your environment for insights into how Coder scales with your infrastructure. For scale-testing Kubernetes clusters we recommend to install and use the dedicated Coder template, scaletest-runner.

Learn more about Coder’s architecture and our scale-testing methodology.

Recent scale tests

Note: the below information is for reference purposes only, and are not intended to be used as guidelines for infrastructure sizing. Review the Reference Architectures for hardware sizing recommendations.

Note: a simulated connection reads and writes random data at 40KB/s per connection.

Scale testing utility

Since Coder's performance is highly dependent on the templates and workflows you support, you may wish to use our internal scale testing utility against your own environments.

Note: This utility is experimental. It is not subject to any compatibility guarantees, and may cause interruptions for your users. To avoid potential outages and orphaned resources, we recommend running scale tests on a secondary "staging" environment or a dedicated Kubernetes playground cluster. Run it against a production environment at your own risk.

Create workspaces

The following command will provision a number of Coder workspaces using the specified template and extra parameters.

coder exp scaletest create-workspaces \
		--retry 5 \
		--count "${SCALETEST_PARAM_NUM_WORKSPACES}" \
		--template "${SCALETEST_PARAM_TEMPLATE}" \
		--concurrency "${SCALETEST_PARAM_CREATE_CONCURRENCY}" \
		--timeout 5h \
		--job-timeout 5h \
		--no-cleanup \
		--output json:"${SCALETEST_RESULTS_DIR}/create-workspaces.json"

# Run `coder exp scaletest create-workspaces --help` for all usage

The command does the following:

1. Create ${SCALETEST_PARAM_NUM_WORKSPACES} workspaces concurrently (concurrency level: ${SCALETEST_PARAM_CREATE_CONCURRENCY}) using the template ${SCALETEST_PARAM_TEMPLATE}.
2. Leave workspaces running to use in next steps (--no-cleanup option).
3. Store provisioning results in JSON format.
4. If you don't want the creation process to be interrupted by any errors, use the --retry 5 flag.

Traffic Generation

Given an existing set of workspaces created previously with create-workspaces, the following command will generate traffic similar to that of Coder's Web Terminal against those workspaces.

# Produce load at about 1000MB/s (25MB/40ms).
coder exp scaletest workspace-traffic \
	--template "${SCALETEST_PARAM_GREEDY_AGENT_TEMPLATE}" \
	--bytes-per-tick $((1024 * 1024 * 25)) \
	--tick-interval 40ms \
	--timeout "$((delay))s" \
	--job-timeout "$((delay))s" \
	--scaletest-prometheus-address 0.0.0.0:21113 \
	--target-workspaces "0:100" \
	--trace=false \
  --output json:"${SCALETEST_RESULTS_DIR}/traffic-${type}-greedy-agent.json"

Traffic generation can be parametrized:

1. Send bytes-per-tick every tick-interval.
2. Enable tracing for performance debugging.
3. Target a range of workspaces with --target-workspaces 0:100.
4. For dashboard traffic: Target a range of users with --target-users 0:100.
5. Store provisioning results in JSON format.
6. Expose a dedicated Prometheus address (--scaletest-prometheus-address) for scaletest-specific metrics.

The workspace-traffic supports also other modes - SSH traffic, workspace app:

1. For SSH traffic: Use --ssh flag to generate SSH traffic instead of Web Terminal.
2. For workspace app traffic: Use --app [wsdi|wsec|wsra] flag to select app behavior. (modes: WebSocket discard, WebSocket echo, WebSocket read).

Cleanup

The scaletest utility will attempt to clean up all workspaces it creates. If you wish to clean up all workspaces, you can run the following command:

coder exp scaletest cleanup \
	--cleanup-job-timeout 2h \
	--cleanup-timeout 15min

This will delete all workspaces and users with the prefix scaletest-.

Scale testing template

Consider using a dedicated scaletest-runner template alongside the CLI utility for testing large-scale Kubernetes clusters.

The template deploys a main workspace with scripts used to orchestrate Coder, creating workspaces, generating workspace traffic, or load-testing workspace apps.

Parameters

The scaletest-runner offers the following configuration options:

• Workspace size selection: minimal/small/medium/large (default: minimal, which contains just enough resources for a Coder agent to run without additional workloads)
• Number of workspaces
• Wait duration between scenarios or staggered approach

The template exposes parameters to control the traffic dimensions for SSH connections, workspace apps, and dashboard tests:

• Traffic duration of the load test scenario
• Traffic percentage of targeted workspaces
• Bytes per tick and tick interval
• For workspace apps: modes (echo, read random data, or write and discard)

Scale testing concurrency can be controlled with the following parameters:

• Enable parallel scenarios - interleave different traffic patterns (SSH, workspace apps, dashboard traffic, etc.)
• Workspace creation concurrency level (default: 10)
• Job concurrency level - generate workspace traffic using multiple jobs (default: 0)
• Cleanup concurrency level

Kubernetes cluster

It is recommended to learn how to operate the scaletest-runner before running it against the staging cluster (or production at your own risk). Coder provides different workspace configurations that operators can deploy depending on the traffic projections.

There are a few cluster options available:

Note: Review the selected cluster template and edit the node affinity to match your setup.

Greedy agent

The greedy agent variant is a template modification that makes the Coder agent transmit large metadata (size: 4K) while reporting stats. The transmission of large chunks puts extra overhead on coderd instances and agents when handling and storing the data.

Use this template variant to verify limits of the cluster performance.

Observability

During scale tests, operators can monitor progress using a Grafana dashboard. Coder offers a comprehensive overview dashboard that can seamlessly integrate into the internal Grafana deployment.

This dashboard provides insights into various aspects, including:

• Utilization of resources within the Coder control plane (CPU, memory, pods)
• Database performance metrics (CPU, memory, I/O, connections, queries)
• Coderd API performance (requests, latency, error rate)
• Resource consumption within Coder workspaces (CPU, memory, network usage)
• Internal metrics related to provisioner jobs

Note: Database metrics are disabled by default and can be enabled by setting the environment variable CODER_PROMETHEUS_COLLECT_DB_METRICS to true.

It is highly recommended to deploy a solution for centralized log collection and aggregation. The presence of error logs may indicate an underscaled deployment of Coder, necessitating action from operators.

Autoscaling

We generally do not recommend using an autoscaler that modifies the number of coderd replicas. In particular, scale down events can cause interruptions for a large number of users.

Coderd is different from a simple request-response HTTP service in that it services long-lived connections whenever it proxies HTTP applications like IDEs or terminals that rely on websockets, or when it relays tunneled connections to workspaces. Loss of a coderd replica will drop these long-lived connections and interrupt users. For example, if you have 4 coderd replicas behind a load balancer, and an autoscaler decides to reduce it to 3, roughly 25% of the connections will drop. An even larger proportion of users could be affected if they use applications that use more than one websocket.

The severity of the interruption varies by application. Coder's web terminal, for example, will reconnect to the same session and continue. So, this should not be interpreted as saying coderd replicas should never be taken down for any reason.

We recommend you plan to run enough coderd replicas to comfortably meet your weekly high-water-mark load, and monitor coderd peak CPU & memory utilization over the long term, reevaluating periodically. When scaling down (or performing upgrades), schedule these outside normal working hours to minimize user interruptions.

A note for Kubernetes users

When running on Kubernetes on cloud infrastructure (i.e. not bare metal), many operators choose to employ a cluster autoscaler that adds and removes Kubernetes nodes according to load. Coder can coexist with such cluster autoscalers, but we recommend you take steps to prevent the autoscaler from evicting coderd pods, as an eviction will cause the same interruptions as described above. For example, if you are using the Kubernetes cluster autoscaler, you may wish to set cluster-autoscaler.kubernetes.io/safe-to-evict: "false" as an annotation on the coderd deployment.

Troubleshooting

If a load test fails or if you are experiencing performance issues during day-to-day use, you can leverage Coder's Prometheus metrics to identify bottlenecks during scale tests. Additionally, you can use your existing cloud monitoring stack to measure load, view server logs, etc.