Configure timeouts and resilience.

Query timeout per individual GraphQL query.
Configure if should attempt to execute as many as possible, or bail out
with the first error.
Rename "cache on error" config variable for better clarity.

config_example.json

@@ -4,7 +4,8 @@
   "graphqlAPIToken": "Token SECRET",
   "cacheExpire": 0,
   "queryTimeout": 60,
-  "retryOnError": false,
+  "failFast": false,
+  "extendCacheOnError": false,
   "queries":[
     {
       "query": "query {device_list {name serial custom_fields}} {{NOW \"-1h\"}}",

internal/config/config.go

@@ -8,13 +8,14 @@ import (
 )
 
 type Cfg struct {
-	GraphqlURL      string
-	GraphqlAPIToken string
-	CacheExpire     int64
-	QueryTimeout    int64
-	RetryOnError    bool
-	MetricsPrefix   string
-	Queries         []Query
+	MetricsPrefix      string
+	GraphqlURL         string
+	GraphqlAPIToken    string
+	CacheExpire        int64
+	QueryTimeout       int64
+	FailFast           bool
+	ExtendCacheOnError bool
+	Queries            []Query
 }
 
 type Query struct {

internal/prometheus/prometheus.go

@@ -122,13 +122,21 @@ func (collector *GraphqlCollector) getMetrics() ([]Metric, error) {
 		result, err := graphql.GraphqlQuery(ctx, q.Query)
 		cancel()
 		if err != nil {
-			slog.Error(fmt.Sprintf("query error: %s", err))
-			continue
+			if config.Config.FailFast {
+				return nil, err
+			} else {
+				slog.Error(fmt.Sprintf("query error: %s", err))
+				continue
+			}
 		}
 		err = json.Unmarshal(result, &gql)
 		if err != nil {
-			slog.Error(fmt.Sprintf("unmarshal error: %s", err))
-			continue
+			if config.Config.FailFast {
+				return nil, err
+			} else {
+				slog.Error(fmt.Sprintf("unmarshal error: %s", err))
+				continue
+			}
 		}
 		data := gql.Data.(map[string]interface{})
 		for _, m := range q.Metrics {
@@ -167,7 +175,7 @@ func (collector *GraphqlCollector) updateMetrics() error {
 		defer collector.accessMu.Unlock()
 		if err != nil {
 			slog.Error(fmt.Sprintf("error collecting metrics: %s", err))
-			if !config.Config.RetryOnError {
+			if config.Config.ExtendCacheOnError {
 				collector.cachedAt = time.Now().Unix()
 			}
 			return err