What causes "an error occurred communicating with k6 Cloud" in grafana/k6?

Retry the command — transient empty responses usually do not repeat

an error occurred communicating with k6 Cloud

Error message

an error occurred communicating with k6 Cloud

What it means

errUnknown is the v6 cloud client's catch-all for the pathological case where the generated OpenAPI client returns neither a response nor an error (res == nil with err == nil after CheckResponse passed). k6 refuses to dereference a nil response and reports this generic communication failure instead of panicking. It can come back from any v6 call: listProjectsPage, listLoadTestsPage, ListLoadZones, ValidateToken, ValidateOptions, createTest, StartTest, FetchTest, etc.

Source

Thrown at internal/cloudapi/v6/errors.go:14

package cloudapi

import (
	"errors"
	"fmt"
	"net/http"
	"strings"

	k6cloud "github.com/grafana/k6-cloud-openapi-client-go/k6"
)

var (
	errTestNotExists = errors.New("load test not found")
	errUnknown       = errors.New("an error occurred communicating with k6 Cloud")
)

// ResponseError represents an error cause by talking to the API
type ResponseError struct {
	Response *http.Response        `json:"-"`
	APIError k6cloud.ErrorApiModel `json:"error"`
}

func (e ResponseError) Error() string {
	err := e.APIError
	msg := err.Message

	if err.Target.IsSet() {
		msg += " (target: '" + *err.Target.Get() + "')"
	}

	details := make([]string, len(err.Details))
	for i, v := range err.Details {

View on GitHub (pinned to 01ffac6f24)

Solutions

Retry the command — transient empty responses usually do not repeat
Update k6 so the bundled cloud client matches the current API contract
Verify K6_CLOUD_HOST is unset or points at the official https://api.grafana.com/cloudloadtesting (do not override unless instructed)
Test the same call from another network to rule out proxy tampering
If persistent, capture verbose logs (--log-output or K6_LOG_LEVEL=debug) and report to Grafana support

Defensive patterns

Strategy: retry

Prevention

Leave K6_CLOUD_HOST at its default unless Grafana support instructs otherwise
Retry cloud commands once on this generic error — nil-response conditions are almost always transient
If it recurs, capture K6_LOG_LEVEL=debug output and test from a different network to isolate proxy tampering

When it happens

Trigger: The cloud service (or an intermediary) returns a 2xx with an empty or undecodable body, so the SDK yields a nil result without an error; keep-alive connection races or proxies returning truncated responses can produce the same nil-response outcome.

Common situations: Corporate proxies or service meshes rewriting/truncating responses; transient Grafana Cloud incidents where a 200 carries an empty body; pointing K6_CLOUD_HOST at an endpoint that is not the real cloud API and returns bare 200s.

Understand the failure class

Background: "empty response", "returned no data", "empty embeddings": what HTTP 200-with-empty-body errors mean across libraries — this error's family across 36 libraries.

Related errors

AI-assisted analysis of grafana/k6@01ffac6f24 (2026-08-18). Data as JSON: /api/errors/c0a8c711a39ccbd6. Report an issue: GitHub.