Datasources: health check logging improvements (#96403)

* basic healthcheck logging

* show detailed error for admins

* fix linting errors

* fix config summary keys

* passing context to logger

* tracking health check errors

* remove tracking

pkg/tsdb/grafana-postgresql-datasource/postgres.go

@@ -260,17 +260,9 @@ func (t *postgresQueryResultTransformer) TransformQueryError(_ log.Logger, err e
 func (s *Service) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
 	dsHandler, err := s.getDSInfo(ctx, req.PluginContext)
 	if err != nil {
-		return nil, err
+		return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: err.Error()}, nil
 	}
-
+	return dsHandler.CheckHealth(ctx, req)
-	err = dsHandler.Ping()
-
-	if err != nil {
-		s.logger.Error("Check health failed", "error", err)
-		return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: dsHandler.TransformQueryError(s.logger, err).Error()}, nil
-	}
-
-	return &backend.CheckHealthResult{Status: backend.HealthStatusOk, Message: "Database Connection OK"}, nil
 }
 
 func (t *postgresQueryResultTransformer) GetConverterList() []sqlutil.StringConverter {

pkg/tsdb/grafana-postgresql-datasource/sqleng/handler_checkhealth.go

@@ -0,0 +1,62 @@
+package sqleng
+
+import (
+	"context"
+	"encoding/json"
+
+	"github.com/grafana/grafana-plugin-sdk-go/backend"
+	"github.com/grafana/grafana-plugin-sdk-go/backend/log"
+)
+
+func (e *DataSourceHandler) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
+	err := e.Ping()
+	if err != nil {
+		logCheckHealthError(ctx, e.dsInfo, err, e.log)
+		if req.PluginContext.User.Role == "Admin" {
+			return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: err.Error()}, nil
+		}
+		errResponse := &backend.CheckHealthResult{
+			Status:  backend.HealthStatusError,
+			Message: e.TransformQueryError(e.log, err).Error(),
+		}
+		return errResponse, nil
+	}
+	return &backend.CheckHealthResult{Status: backend.HealthStatusOk, Message: "Database Connection OK"}, nil
+}
+
+func logCheckHealthError(ctx context.Context, dsInfo DataSourceInfo, err error, logger log.Logger) {
+	configSummary := map[string]any{
+		"config_url_length":                 len(dsInfo.URL),
+		"config_user_length":                len(dsInfo.User),
+		"config_database_length":            len(dsInfo.Database),
+		"config_json_data_database_length":  len(dsInfo.JsonData.Database),
+		"config_max_open_conns":             dsInfo.JsonData.MaxOpenConns,
+		"config_max_idle_conns":             dsInfo.JsonData.MaxIdleConns,
+		"config_conn_max_life_time":         dsInfo.JsonData.ConnMaxLifetime,
+		"config_conn_timeout":               dsInfo.JsonData.ConnectionTimeout,
+		"config_timescaledb":                dsInfo.JsonData.Timescaledb,
+		"config_ssl_mode":                   dsInfo.JsonData.Mode,
+		"config_tls_configuration_method":   dsInfo.JsonData.ConfigurationMethod,
+		"config_tls_skip_verify":            dsInfo.JsonData.TlsSkipVerify,
+		"config_timezone":                   dsInfo.JsonData.Timezone,
+		"config_time_interval":              dsInfo.JsonData.TimeInterval,
+		"config_enable_secure_proxy":        dsInfo.JsonData.SecureDSProxy,
+		"config_allow_clear_text_passwords": dsInfo.JsonData.AllowCleartextPasswords,
+		"config_authentication_type":        dsInfo.JsonData.AuthenticationType,
+		"config_ssl_root_cert_file_length":  len(dsInfo.JsonData.RootCertFile),
+		"config_ssl_cert_file_length":       len(dsInfo.JsonData.CertFile),
+		"config_ssl_key_file_length":        len(dsInfo.JsonData.CertKeyFile),
+		"config_encrypt_length":             len(dsInfo.JsonData.Encrypt),
+		"config_server_name_length":         len(dsInfo.JsonData.Servername),
+		"config_password_length":            len(dsInfo.DecryptedSecureJSONData["password"]),
+		"config_tls_ca_cert_length":         len(dsInfo.DecryptedSecureJSONData["tlsCACert"]),
+		"config_tls_client_cert_length":     len(dsInfo.DecryptedSecureJSONData["tlsClientCert"]),
+		"config_tls_client_key_length":      len(dsInfo.DecryptedSecureJSONData["tlsClientKey"]),
+	}
+	configSummaryJson, marshalError := json.Marshal(configSummary)
+	if marshalError != nil {
+		logger.Error("Check health failed", "error", err, "message_type", "ds_config_health_check_error", "plugin_id", "grafana-postgresql-datasource")
+		return
+	}
+	logger.Error("Check health failed", "error", err, "message_type", "ds_config_health_check_error_detailed", "plugin_id", "grafana-postgresql-datasource", "details", string(configSummaryJson))
+}

pkg/tsdb/mysql/mysql_service.go

@@ -37,7 +37,7 @@ func (s *Service) getDataSourceHandler(ctx context.Context, pluginCtx backend.Pl
 func (s *Service) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
 	dsHandler, err := s.getDataSourceHandler(ctx, req.PluginContext)
 	if err != nil {
-		return nil, err
+		return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: err.Error()}, nil
 	}
 
 	return dsHandler.CheckHealth(ctx, req)

pkg/tsdb/mysql/sqleng/handler_checkhealth.go

@@ -0,0 +1,64 @@
+package sqleng
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+
+	"github.com/go-sql-driver/mysql"
+	"github.com/grafana/grafana-plugin-sdk-go/backend"
+	"github.com/grafana/grafana-plugin-sdk-go/backend/log"
+)
+
+func (e *DataSourceHandler) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
+	err := e.db.Ping()
+	if err != nil {
+		logCheckHealthError(ctx, e.dsInfo, err, e.log)
+		if req.PluginContext.User.Role == "Admin" {
+			return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: err.Error()}, nil
+		}
+		var driverErr *mysql.MySQLError
+		if errors.As(err, &driverErr) {
+			return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: e.TransformQueryError(e.log, driverErr).Error()}, nil
+		}
+		return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: e.TransformQueryError(e.log, err).Error()}, nil
+	}
+	return &backend.CheckHealthResult{Status: backend.HealthStatusOk, Message: "Database Connection OK"}, nil
+}
+
+func logCheckHealthError(ctx context.Context, dsInfo DataSourceInfo, err error, logger log.Logger) {
+	configSummary := map[string]any{
+		"config_url_length":                 len(dsInfo.URL),
+		"config_user_length":                len(dsInfo.User),
+		"config_database_length":            len(dsInfo.Database),
+		"config_json_data_database_length":  len(dsInfo.JsonData.Database),
+		"config_max_open_conns":             dsInfo.JsonData.MaxOpenConns,
+		"config_max_idle_conns":             dsInfo.JsonData.MaxIdleConns,
+		"config_conn_max_life_time":         dsInfo.JsonData.ConnMaxLifetime,
+		"config_conn_timeout":               dsInfo.JsonData.ConnectionTimeout,
+		"config_timescaledb":                dsInfo.JsonData.Timescaledb,
+		"config_ssl_mode":                   dsInfo.JsonData.Mode,
+		"config_tls_configuration_method":   dsInfo.JsonData.ConfigurationMethod,
+		"config_tls_skip_verify":            dsInfo.JsonData.TlsSkipVerify,
+		"config_timezone":                   dsInfo.JsonData.Timezone,
+		"config_time_interval":              dsInfo.JsonData.TimeInterval,
+		"config_enable_secure_proxy":        dsInfo.JsonData.SecureDSProxy,
+		"config_allow_clear_text_passwords": dsInfo.JsonData.AllowCleartextPasswords,
+		"config_authentication_type":        dsInfo.JsonData.AuthenticationType,
+		"config_ssl_root_cert_file_length":  len(dsInfo.JsonData.RootCertFile),
+		"config_ssl_cert_file_length":       len(dsInfo.JsonData.CertFile),
+		"config_ssl_key_file_length":        len(dsInfo.JsonData.CertKeyFile),
+		"config_encrypt_length":             len(dsInfo.JsonData.Encrypt),
+		"config_server_name_length":         len(dsInfo.JsonData.Servername),
+		"config_password_length":            len(dsInfo.DecryptedSecureJSONData["password"]),
+		"config_tls_ca_cert_length":         len(dsInfo.DecryptedSecureJSONData["tlsCACert"]),
+		"config_tls_client_cert_length":     len(dsInfo.DecryptedSecureJSONData["tlsClientCert"]),
+		"config_tls_client_key_length":      len(dsInfo.DecryptedSecureJSONData["tlsClientKey"]),
+	}
+	configSummaryJson, marshalError := json.Marshal(configSummary)
+	if marshalError != nil {
+		logger.Error("Check health failed", "error", err, "message_type", "ds_config_health_check_error", "plugin_id", "mysql")
+		return
+	}
+	logger.Error("Check health failed", "error", err, "message_type", "ds_config_health_check_error_detailed", "plugin_id", "mysql", "details", string(configSummaryJson))
+}

pkg/tsdb/mysql/sqleng/sql_engine.go

@@ -14,7 +14,6 @@ import (
 	"sync"
 	"time"
 
-	"github.com/go-sql-driver/mysql"
 	"github.com/grafana/grafana-plugin-sdk-go/backend"
 	"github.com/grafana/grafana-plugin-sdk-go/backend/gtime"
 	"github.com/grafana/grafana-plugin-sdk-go/backend/log"
@@ -153,19 +152,6 @@ func (e *DataSourceHandler) Dispose() {
 	e.log.Debug("DB disposed")
 }
 
-func (e *DataSourceHandler) CheckHealth(ctx context.Context, req *backend.CheckHealthRequest) (*backend.CheckHealthResult, error) {
-	err := e.db.Ping()
-
-	if err != nil {
-		var driverErr *mysql.MySQLError
-		if errors.As(err, &driverErr) {
-			return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: e.TransformQueryError(e.log, driverErr).Error()}, nil
-		}
-		return &backend.CheckHealthResult{Status: backend.HealthStatusError, Message: e.TransformQueryError(e.log, err).Error()}, nil
-	}
-	return &backend.CheckHealthResult{Status: backend.HealthStatusOk, Message: "Database Connection OK"}, nil
-}
-
 func (e *DataSourceHandler) QueryData(ctx context.Context, req *backend.QueryDataRequest) (*backend.QueryDataResponse, error) {
 	result := backend.NewQueryDataResponse()
 	ch := make(chan DBDataResponse, len(req.Queries))