AzureMonitor: Migrate to backend checkHealth API (#50448)

* Add check health functions for each datasource and generic checkHealth function * Log backend errors * Update testDatasource function - Remove unused testDatasource functions from pseudo datasources * Switch datasource to extend DataSourceWithBackend * Improve errors and responses from health endpoint * Fix backend lint issues * Remove unneeded frontend tests * Remove unused/unnecessary datasource methods * Update types * Improve message construction * Stubbing out checkHealth tests * Update tests - Remove comments - Simplify structure * Update log analytics health check to query data rather than retrieve workspace metadata * Fix lint issue * Fix frontend lint issues * Update pkg/tsdb/azuremonitor/azuremonitor.go Co-authored-by: Andres Martinez Gotor <andres.martinez@grafana.com> * Updates based on PR comments - Don't use deprecated default workspace field - Handle situation if no workspace is found by notifying user - Correctly handle health responses * Remove debug line * Make use of defined api versions * Remove field validation functions * Expose errors in frontend * Update errors and tests * Remove instanceSettings * Update error handling * Improve error handling and presentation * Update tests and correctly check error type * Refactor AzureHealthCheckError and update tests * Fix lint errors Co-authored-by: Andres Martinez Gotor <andres.martinez@grafana.com>
2025-02-25 18:55:37 -06:00 · 2022-06-20 11:33:13 +01:00
parent 62c2b1ec78
commit ecaa1dcbfd
10 changed files with 525 additions and 233 deletions
--- a/pkg/tsdb/azuremonitor/azuremonitor.go
+++ b/pkg/tsdb/azuremonitor/azuremonitor.go
@@ -1,10 +1,14 @@
 package azuremonitor

 import (
+	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
+	"io"
 	"net/http"
+	"strconv"

 	"github.com/grafana/grafana-plugin-sdk-go/backend"
 	"github.com/grafana/grafana-plugin-sdk-go/backend/datasource"
@@ -161,3 +165,202 @@ func (s *Service) newQueryMux() *datasource.QueryTypeMux {
 	}
 	return mux
 }
+
+func (s *Service) getDSInfo(pluginCtx backend.PluginContext) (types.DatasourceInfo, error) {
+	i, err := s.im.Get(pluginCtx)
+	if err != nil {
+		return types.DatasourceInfo{}, err
+	}
+
+	instance, ok := i.(types.DatasourceInfo)
+	if !ok {
+		return types.DatasourceInfo{}, fmt.Errorf("failed to cast datsource info")
+	}
+
+	return instance, nil
+}
+
+func checkAzureMonitorMetricsHealth(dsInfo types.DatasourceInfo) (*http.Response, error) {
+	url := fmt.Sprintf("%v/subscriptions?api-version=%v", dsInfo.Routes["Azure Monitor"].URL, metrics.AzureMonitorAPIVersion)
+	request, err := http.NewRequest(http.MethodGet, url, nil)
+	if err != nil {
+		return nil, err
+	}
+
+	res, err := dsInfo.Services["Azure Monitor"].HTTPClient.Do(request)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", types.ErrorAzureHealthCheck, err)
+	}
+
+	return res, nil
+}
+
+func checkAzureLogAnalyticsHealth(dsInfo types.DatasourceInfo) (*http.Response, error) {
+	workspacesUrl := fmt.Sprintf("%v/subscriptions/%v/providers/Microsoft.OperationalInsights/workspaces?api-version=2017-04-26-preview", dsInfo.Routes["Azure Monitor"].URL, dsInfo.Settings.SubscriptionId)
+	workspacesReq, err := http.NewRequest(http.MethodGet, workspacesUrl, nil)
+	if err != nil {
+		return nil, err
+	}
+	res, err := dsInfo.Services["Azure Monitor"].HTTPClient.Do(workspacesReq)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", types.ErrorAzureHealthCheck, err)
+	}
+	var target struct {
+		Value []types.LogAnalyticsWorkspaceResponse
+	}
+	err = json.NewDecoder(res.Body).Decode(&target)
+	if err != nil {
+		return nil, err
+	}
+
+	if len(target.Value) == 0 {
+		return nil, errors.New("no default workspace found")
+	}
+	defaultWorkspaceId := target.Value[0].Properties.CustomerId
+
+	body, err := json.Marshal(map[string]interface{}{
+		"query": "AzureActivity | limit 1",
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	workspaceUrl := fmt.Sprintf("%v/v1/workspaces/%v/query", dsInfo.Routes["Azure Log Analytics"].URL, defaultWorkspaceId)
+	workspaceReq, err := http.NewRequest(http.MethodPost, workspaceUrl, bytes.NewBuffer(body))
+	workspaceReq.Header.Set("Content-Type", "application/json")
+	if err != nil {
+		return nil, err
+	}
+
+	res, err = dsInfo.Services["Azure Log Analytics"].HTTPClient.Do(workspaceReq)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", types.ErrorAzureHealthCheck, err)
+	}
+
+	return res, nil
+}
+
+func checkAzureMonitorResourceGraphHealth(dsInfo types.DatasourceInfo) (*http.Response, error) {
+	body, err := json.Marshal(map[string]interface{}{
+		"query":         "Resources | project id | limit 1",
+		"subscriptions": []string{dsInfo.Settings.SubscriptionId},
+	})
+	if err != nil {
+		return nil, err
+	}
+	url := fmt.Sprintf("%v/providers/Microsoft.ResourceGraph/resources?api-version=%v", dsInfo.Routes["Azure Resource Graph"].URL, resourcegraph.ArgAPIVersion)
+	request, err := http.NewRequest(http.MethodPost, url, bytes.NewBuffer(body))
+	request.Header.Set("Content-Type", "application/json")
+	if err != nil {
+		return nil, err
+	}
+
+	res, err := dsInfo.Services["Azure Resource Graph"].HTTPClient.Do(request)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", types.ErrorAzureHealthCheck, err)
+	}
+
+	return res, nil
+}
+
+func (s *Service) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
+	dsInfo, err := s.getDSInfo(req.PluginContext)
+	if err != nil {
+		return nil, err
+	}
+
+	status := backend.HealthStatusOk
+	metricsLog := "Successfully connected to Azure Monitor endpoint."
+	logAnalyticsLog := "Successfully connected to Azure Log Analytics endpoint."
+	graphLog := "Successfully connected to Azure Resource Graph endpoint."
+
+	metricsRes, err := checkAzureMonitorMetricsHealth(dsInfo)
+	if err != nil || metricsRes.StatusCode != 200 {
+		status = backend.HealthStatusError
+		if err != nil {
+			if ok := errors.Is(err, types.ErrorAzureHealthCheck); ok {
+				metricsLog = fmt.Sprintf("Error connecting to Azure Monitor endpoint: %s", err.Error())
+			} else {
+				return nil, err
+			}
+		} else {
+			body, err := io.ReadAll(metricsRes.Body)
+			if err != nil {
+				return nil, err
+			}
+			metricsLog = fmt.Sprintf("Error connecting to Azure Monitor endpoint: %s", string(body))
+		}
+	}
+
+	logsRes, err := checkAzureLogAnalyticsHealth(dsInfo)
+	if err != nil || logsRes.StatusCode != 200 {
+		status = backend.HealthStatusError
+		if err != nil {
+			if err.Error() == "no default workspace found" {
+				status = backend.HealthStatusUnknown
+				logAnalyticsLog = "No Log Analytics workspaces found."
+			} else if ok := errors.Is(err, types.ErrorAzureHealthCheck); ok {
+				logAnalyticsLog = fmt.Sprintf("Error connecting to Azure Log Analytics endpoint: %s", err.Error())
+			} else {
+				return nil, err
+			}
+		} else {
+			body, err := io.ReadAll(logsRes.Body)
+			if err != nil {
+				return nil, err
+			}
+			logAnalyticsLog = fmt.Sprintf("Error connecting to Azure Log Analytics endpoint: %s", string(body))
+		}
+	}
+
+	resourceGraphRes, err := checkAzureMonitorResourceGraphHealth(dsInfo)
+	if err != nil || resourceGraphRes.StatusCode != 200 {
+		status = backend.HealthStatusError
+		if err != nil {
+			if ok := errors.Is(err, types.ErrorAzureHealthCheck); ok {
+				graphLog = fmt.Sprintf("Error connecting to Azure Resource Graph endpoint: %s", err.Error())
+			} else {
+				return nil, err
+			}
+		} else {
+			body, err := io.ReadAll(resourceGraphRes.Body)
+			if err != nil {
+				return nil, err
+			}
+			graphLog = fmt.Sprintf("Error connecting to Azure Resource Graph endpoint: %s", string(body))
+		}
+	}
+
+	defer func() {
+		if metricsRes != nil {
+			if err := metricsRes.Body.Close(); err != nil {
+				backend.Logger.Error("Failed to close response body", "err", err)
+			}
+		}
+		if logsRes != nil {
+			if err := logsRes.Body.Close(); logsRes != nil && err != nil {
+				backend.Logger.Error("Failed to close response body", "err", err)
+			}
+		}
+		if resourceGraphRes != nil {
+			if err := resourceGraphRes.Body.Close(); resourceGraphRes != nil && err != nil {
+				backend.Logger.Error("Failed to close response body", "err", err)
+			}
+		}
+	}()
+
+	if status == backend.HealthStatusOk {
+		return &backend.CheckHealthResult{
+			Status:  status,
+			Message: "Successfully connected to all Azure Monitor endpoints.",
+		}, nil
+	}
+
+	return &backend.CheckHealthResult{
+		Status:  status,
+		Message: "One or more health checks failed. See details below.",
+		JSONDetails: []byte(
+			fmt.Sprintf(`{"verboseMessage": %s }`, strconv.Quote(fmt.Sprintf("1. %s\n2. %s\n3. %s", metricsLog, logAnalyticsLog, graphLog))),
+		),
+	}, nil
+}