Fix spurious use of indexResolver preventing idling (#589)

We have noticed that Quesma still has problems with properly idling (not
making queries to ClickHouse Cloud to allow it to idle).

The current idling logic relies on checking if table discovery was
recently used. It turns out that idle Kibana very often does inserts to
an internal `.kibana` index and to `/_monitoring/bulk` endpoint.

Even though those requests are routed to Elastic, we still checked if
`.kibana` or `_monitoring` are names of enabled indices using
indexResolver - this use bumped the LastAccessTime, preventing idling.

Fix the issue by avoiding a use of indexResolver if it's an endpoint
(`_endpoint`) or if it's an internal index (starting with `.`).

quesma/feature/not_supported.go

@@ -5,7 +5,9 @@ package feature
 import (
 	"context"
 	"fmt"
+	"quesma/elasticsearch"
 	"regexp"
+	"strings"
 )
 
 // https://www.elastic.co/guide/en/elasticsearch/reference/current/search-template.html
@@ -80,6 +82,16 @@ func checkIfOurIndex(ctx context.Context, method string, path string, indexResol
 	if len(match) > 1 {
 		indexNamePart := match[1]
 		var matched bool
+
+		if strings.HasPrefix(indexNamePart, "_") {
+			// This is not actually an index, this is some endpoint (for example /_monitoring/bulk)
+			return false
+		}
+		// Optimization: avoid using indexResolver for internal indexes
+		if elasticsearch.IsInternalIndex(indexNamePart) {
+			return false
+		}
+
 		indexes, err := indexResolver(ctx, indexNamePart)
 		if err != nil {
 			logMessage("Error resolving index: %v", err)