X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/e14975a4a482dd4a6c1579fbeee9038d7227b385..89be4b30feccc3680ca77339711b29367754dc05:/sdk/go/health/aggregator.go

diff --git a/sdk/go/health/aggregator.go b/sdk/go/health/aggregator.go
index 334584b622..564331327a 100644
--- a/sdk/go/health/aggregator.go
+++ b/sdk/go/health/aggregator.go
@@ -1,8 +1,13 @@
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: Apache-2.0
+
 package health
 
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net"
 	"net/http"
@@ -82,7 +87,7 @@ type ClusterHealthResponse struct {
 	// exposes problems that can't be expressed in Checks, like
 	// "service S is needed, but isn't configured to run
 	// anywhere."
-	Services map[string]ServiceHealth `json:"services"`
+	Services map[arvados.ServiceName]ServiceHealth `json:"services"`
 }
 
 type CheckResult struct {
@@ -103,13 +108,13 @@ func (agg *Aggregator) ClusterHealth(cluster *arvados.Cluster) ClusterHealthResp
 	resp := ClusterHealthResponse{
 		Health:   "OK",
 		Checks:   make(map[string]CheckResult),
-		Services: make(map[string]ServiceHealth),
+		Services: make(map[arvados.ServiceName]ServiceHealth),
 	}
 
 	mtx := sync.Mutex{}
 	wg := sync.WaitGroup{}
-	for node, nodeConfig := range cluster.SystemNodes {
-		for svc, addr := range nodeConfig.ServicePorts() {
+	for profileName, profile := range cluster.NodeProfiles {
+		for svc, addr := range profile.ServicePorts() {
 			// Ensure svc is listed in resp.Services.
 			mtx.Lock()
 			if _, ok := resp.Services[svc]; !ok {
@@ -123,10 +128,10 @@ func (agg *Aggregator) ClusterHealth(cluster *arvados.Cluster) ClusterHealthResp
 			}
 
 			wg.Add(1)
-			go func(node, svc, addr string) {
+			go func(profileName string, svc arvados.ServiceName, addr string) {
 				defer wg.Done()
 				var result CheckResult
-				url, err := agg.pingURL(node, addr)
+				url, err := agg.pingURL(profileName, addr)
 				if err != nil {
 					result = CheckResult{
 						Health: "ERROR",
@@ -138,7 +143,7 @@ func (agg *Aggregator) ClusterHealth(cluster *arvados.Cluster) ClusterHealthResp
 
 				mtx.Lock()
 				defer mtx.Unlock()
-				resp.Checks[svc+"+"+url] = result
+				resp.Checks[fmt.Sprintf("%s+%s", svc, url)] = result
 				if result.Health == "OK" {
 					h := resp.Services[svc]
 					h.N++
@@ -147,7 +152,7 @@ func (agg *Aggregator) ClusterHealth(cluster *arvados.Cluster) ClusterHealthResp
 				} else {
 					resp.Health = "ERROR"
 				}
-			}(node, svc, addr)
+			}(profileName, svc, addr)
 		}
 	}
 	wg.Wait()
@@ -199,16 +204,20 @@ func (agg *Aggregator) ping(url string, cluster *arvados.Cluster) (result CheckR
 	err = json.NewDecoder(resp.Body).Decode(&result.Response)
 	if err != nil {
 		err = fmt.Errorf("cannot decode response: %s", err)
-		return
 	} else if resp.StatusCode != http.StatusOK {
 		err = fmt.Errorf("HTTP %d %s", resp.StatusCode, resp.Status)
-		return
+	} else if h, _ := result.Response["health"].(string); h != "OK" {
+		if e, ok := result.Response["error"].(string); ok && e != "" {
+			err = errors.New(e)
+		} else {
+			err = fmt.Errorf("health=%q in ping response", h)
+		}
 	}
 	return
 }
 
 func (agg *Aggregator) checkAuth(req *http.Request, cluster *arvados.Cluster) bool {
-	creds := auth.NewCredentialsFromHTTPRequest(req)
+	creds := auth.CredentialsFromRequest(req)
 	for _, token := range creds.Tokens {
 		if token != "" && token == cluster.ManagementToken {
 			return true