roachprod: permit disabling of Prometheus registration

This patch makes an empty value for the COCKROACH_PROM_HOST_URL env var
disable the roachprod's cluster registration with the Prometheus
registration service. This is useful for people who don't work at
CockroachLabs and don't have access to this service.

Release note: None
Epic: None

pkg/roachprod/promhelperclient/BUILD.bazel

@@ -9,6 +9,7 @@ go_library(
     importpath = "github.com/cockroachdb/cockroach/pkg/roachprod/promhelperclient",
     visibility = ["//visibility:public"],
     deps = [
+        "//pkg/roachprod/config",
         "//pkg/roachprod/logger",
         "//pkg/roachprod/vm/gce",
         "//pkg/util/httputil",

pkg/roachprod/promhelperclient/client.go

@@ -24,6 +24,7 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/cockroachdb/cockroach/pkg/roachprod/config"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/logger"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/vm/gce"
 	"github.com/cockroachdb/cockroach/pkg/util/httputil"
@@ -34,24 +35,27 @@ import (
 )
 
 const (
-	resourceName    = "instance-configs"
-	resourceVersion = "v1"
-
-	// defaultPrometheusHostUrl for prometheus cluster config
-	defaultPrometheusHostUrl = "https://grafana.testeng.crdb.io/promhelpers"
-	// prometheusHostUrlEnv is the environment variable to override defaultPrometheusHostUrl
-	prometheusHostUrlEnv = "ROACHPROD_PROM_HOST_URL"
-
+	resourceName           = "instance-configs"
+	resourceVersion        = "v1"
 	serviceAccountJson     = "PROM_HELPER_SERVICE_ACCOUNT_JSON"
 	serviceAccountAudience = "PROM_HELPER_SERVICE_ACCOUNT_AUDIENCE"
 )
 
 // SupportedPromProjects are the projects supported for prometheus target
 var SupportedPromProjects = map[string]struct{}{gce.DefaultProject(): {}}
 
+// The URL for the Prometheus registration service. An empty string means that the
+// Prometheus integration is disabled. Should be accessed through
+// getPrometheusRegistrationUrl().
+var promRegistrationUrl = config.EnvOrDefaultString("ROACHPROD_PROM_HOST_URL",
+	"https://grafana.testeng.crdb.io/promhelpers")
+
 // PromClient is used to communicate with the prometheus helper service
 // keeping the functions as a variable enables us to override the value for unit testing
 type PromClient struct {
+	promUrl  string
+	disabled bool
+
 	// httpPut is used for http PUT operation.
 	httpPut func(
 		ctx context.Context, url string, h *http.Header, body io.Reader,
@@ -64,15 +68,26 @@ type PromClient struct {
 		oauth2.TokenSource, error)
 }
 
+// DefaultPromClient is the default instance of PromClient. This instance should
+// be used unless custom configuration is needed.
+var DefaultPromClient = NewPromClient()
+
 // NewPromClient returns a new instance of PromClient
 func NewPromClient() *PromClient {
 	return &PromClient{
+		promUrl:        promRegistrationUrl,
+		disabled:       promRegistrationUrl == "",
 		httpPut:        httputil.Put,
 		httpDelete:     httputil.Delete,
 		newTokenSource: idtoken.NewTokenSource,
 	}
 }
 
+func (c *PromClient) setUrl(url string) {
+	c.promUrl = url
+	c.disabled = false
+}
+
 // instanceConfigRequest is the HTTP request received for generating instance config
 type instanceConfigRequest struct {
 	//Config is the content of the yaml file
@@ -89,19 +104,19 @@ func (c *PromClient) UpdatePrometheusTargets(
 	insecure bool,
 	l *logger.Logger,
 ) error {
-	promUrl := defaultPrometheusHostUrl
-	if v, ok := os.LookupEnv(prometheusHostUrlEnv); ok {
-		promUrl = v
+	if c.disabled {
+		l.Printf("Prometheus registration is disabled")
+		return nil
 	}
 	req, err := buildCreateRequest(nodes, insecure)
 	if err != nil {
 		return err
 	}
-	token, err := c.getToken(ctx, promUrl, forceFetchCreds, l)
+	token, err := c.getToken(ctx, forceFetchCreds, l)
 	if err != nil {
 		return err
 	}
-	url := getUrl(promUrl, clusterName)
+	url := getUrl(c.promUrl, clusterName)
 	l.Printf("invoking PUT for URL: %s", url)
 	h := &http.Header{}
 	h.Set("ContentType", "application/json")
@@ -132,15 +147,15 @@ func (c *PromClient) UpdatePrometheusTargets(
 func (c *PromClient) DeleteClusterConfig(
 	ctx context.Context, clusterName string, forceFetchCreds bool, l *logger.Logger,
 ) error {
-	promUrl := defaultPrometheusHostUrl
-	if v, ok := os.LookupEnv(prometheusHostUrlEnv); ok {
-		promUrl = v
+
+	if c.disabled {
+		return nil
 	}
-	token, err := c.getToken(ctx, promUrl, forceFetchCreds, l)
+	token, err := c.getToken(ctx, forceFetchCreds, l)
 	if err != nil {
 		return err
 	}
-	url := getUrl(promUrl, clusterName)
+	url := getUrl(c.promUrl, clusterName)
 	l.Printf("invoking DELETE for URL: %s", url)
 	h := &http.Header{}
 	h.Set("Authorization", token)
@@ -206,9 +221,9 @@ func buildCreateRequest(nodes map[int]*NodeInfo, insecure bool) (io.Reader, erro
 
 // getToken gets the Authorization token for grafana
 func (c *PromClient) getToken(
-	ctx context.Context, promUrl string, forceFetchCreds bool, l *logger.Logger,
+	ctx context.Context, forceFetchCreds bool, l *logger.Logger,
 ) (string, error) {
-	if strings.HasPrefix(promUrl, "http:/") {
+	if strings.HasPrefix(c.promUrl, "http:/") {
 		// no token needed for insecure URL
 		return "", nil
 	}

pkg/roachprod/promhelperclient/client_test.go

@@ -39,8 +39,8 @@ func TestUpdatePrometheusTargets(t *testing.T) {
 	}()
 	ctx := context.Background()
 	promUrl := "http://prom_url.com"
-	_ = os.Setenv(prometheusHostUrlEnv, promUrl)
 	c := NewPromClient()
+	c.setUrl(promUrl)
 	t.Run("UpdatePrometheusTargets fails with 400", func(t *testing.T) {
 		c.httpPut = func(ctx context.Context, reqUrl string, h *http.Header, body io.Reader) (
 			resp *http.Response, err error) {
@@ -96,8 +96,8 @@ func TestDeleteClusterConfig(t *testing.T) {
 	}()
 	ctx := context.Background()
 	promUrl := "http://prom_url.com"
-	_ = os.Setenv(prometheusHostUrlEnv, promUrl)
 	c := NewPromClient()
+	c.setUrl(promUrl)
 	t.Run("DeleteClusterConfig fails with 400", func(t *testing.T) {
 		c.httpDelete = func(ctx context.Context, url string, h *http.Header) (
 			resp *http.Response, err error) {
@@ -144,7 +144,8 @@ func Test_getToken(t *testing.T) {
 	}()
 	c := NewPromClient()
 	t.Run("insecure url", func(t *testing.T) {
-		token, err := c.getToken(ctx, "http://test.com", false, l)
+		c.setUrl("http://test.com")
+		token, err := c.getToken(ctx, false, l)
 		require.Nil(t, err)
 		require.Empty(t, token)
 	})
@@ -156,7 +157,8 @@ func Test_getToken(t *testing.T) {
 		c.newTokenSource = func(ctx context.Context, audience string, opts ...idtoken.ClientOption) (oauth2.TokenSource, error) {
 			return nil, fmt.Errorf("invalid")
 		}
-		token, err := c.getToken(ctx, "https://test.com", false, l)
+		c.setUrl("https://test.com")
+		token, err := c.getToken(ctx, false, l)
 		require.NotNil(t, err)
 		require.Empty(t, token)
 		require.Equal(t, "error creating GCS oauth token source from specified credential: invalid", err.Error())
@@ -169,7 +171,8 @@ func Test_getToken(t *testing.T) {
 		c.newTokenSource = func(ctx context.Context, audience string, opts ...idtoken.ClientOption) (oauth2.TokenSource, error) {
 			return &mockToken{token: "", err: fmt.Errorf("failed")}, nil
 		}
-		token, err := c.getToken(ctx, "https://test.com", false, l)
+		c.setUrl("https://test.com")
+		token, err := c.getToken(ctx, false, l)
 		require.NotNil(t, err)
 		require.Empty(t, token)
 		require.Equal(t, "error getting identity token: failed", err.Error())
@@ -182,7 +185,8 @@ func Test_getToken(t *testing.T) {
 		c.newTokenSource = func(ctx context.Context, audience string, opts ...idtoken.ClientOption) (oauth2.TokenSource, error) {
 			return &mockToken{token: "token"}, nil
 		}
-		token, err := c.getToken(ctx, "https://test.com", false, l)
+		c.setUrl("https://test.com")
+		token, err := c.getToken(ctx, false, l)
 		require.Nil(t, err)
 		require.Equal(t, "Bearer token", token)
 	})

pkg/roachprod/roachprod.go

@@ -59,6 +59,9 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+// supportedPromProjects are the projects supported for prometheus target
+var supportedPromProjects = map[string]struct{}{gce.DefaultProject(): {}}
+
 // MalformedClusterNameError is returned when the cluster name passed to Create is invalid.
 type MalformedClusterNameError struct {
 	name        string
@@ -821,7 +824,7 @@ func updatePrometheusTargets(ctx context.Context, l *logger.Logger, c *install.S
 	}
 	wg.Wait()
 	if len(nodeIPPorts) > 0 {
-		if err := promhelperclient.NewPromClient().UpdatePrometheusTargets(ctx,
+		if err := promhelperclient.DefaultPromClient.UpdatePrometheusTargets(ctx,
 			c.Name, false, nodeIPPorts, !c.Secure, l); err != nil {
 			l.Errorf("creating cluster config failed for the ip:ports %v: %v", nodeIPPorts, err)
 		}
@@ -1453,15 +1456,17 @@ func Destroy(
 				// and let the caller retry.
 				cld, _ = cloud.ListCloud(l, vm.ListOptions{IncludeEmptyClusters: true})
 			}
-			return destroyCluster(cld, l, name)
+			return destroyCluster(ctx, cld, l, name)
 		}); err != nil {
 		return err
 	}
 	l.Printf("OK")
 	return nil
 }
 
-func destroyCluster(cld *cloud.Cloud, l *logger.Logger, clusterName string) error {
+func destroyCluster(
+	ctx context.Context, cld *cloud.Cloud, l *logger.Logger, clusterName string,
+) error {
 	c, ok := cld.Clusters[clusterName]
 	if !ok {
 		return fmt.Errorf("cluster %s does not exist", clusterName)