feat(metadata): introduce a separate split interval for recent query …

…window (#11897)

(cherry picked from commit 9e7725b)

CHANGELOG.md

@@ -54,6 +54,7 @@
 * [11143](https://github.com/grafana/loki/pull/11143) **sandeepsukhani** otel: Add support for per tenant configuration for mapping otlp data to loki format
 * [11499](https://github.com/grafana/loki/pull/11284) **jmichalek132** Config: Adds `frontend.log-query-request-headers` to enable logging of request headers in query logs.
 * [11817](https://github.com/grafana/loki/pull/11817) **ashwanthgoli** Ruler: Add support for filtering results of `/prometheus/api/v1/rules` endpoint by rule_name, rule_group, file and type.
+* [11897](https://github.com/grafana/loki/pull/11897) **ashwanthgoli** Metadata: Introduces a separate split interval of `split_recent_metadata_queries_by_interval` for `recent_metadata_query_window` to help with caching recent metadata query results.
 
 ##### Fixes
 * [11074](https://github.com/grafana/loki/pull/11074) **hainenber** Fix panic in lambda-promtail due to mishandling of empty DROP_LABELS env var.

docs/sources/configure/_index.md

@@ -2906,6 +2906,30 @@ The `limits_config` block configures global and per-tenant limits in Loki.
 # CLI flag: -querier.split-metadata-queries-by-interval
 [split_metadata_queries_by_interval: <duration> | default = 1d]
 
+# Experimental. Split interval to use for the portion of metadata request that
+# falls within `recent_metadata_query_window`. Rest of the request which is
+# outside the window still uses `split_metadata_queries_by_interval`. If set to
+# 0, the entire request defaults to using a split interval of
+# `split_metadata_queries_by_interval.`.
+# CLI flag: -experimental.querier.split-recent-metadata-queries-by-interval
+[split_recent_metadata_queries_by_interval: <duration> | default = 1h]
+
+# Experimental. Metadata query window inside which
+# `split_recent_metadata_queries_by_interval` gets applied, portion of the
+# metadata request that falls in this window is split using
+# `split_recent_metadata_queries_by_interval`. The value 0 disables using a
+# different split interval for recent metadata queries.
+# 
+# This is added to improve cacheability of recent metadata queries. Query split
+# interval also determines the interval used in cache key. The default split
+# interval of 24h is useful for caching long queries, each cache key holding 1
+# day's results. But metadata queries are often shorter than 24h, to cache them
+# effectively we need a smaller split interval. `recent_metadata_query_window`
+# along with `split_recent_metadata_queries_by_interval` help configure a
+# shorter split interval for recent metadata queries.
+# CLI flag: -experimental.querier.recent-metadata-query-window
+[recent_metadata_query_window: <duration> | default = 0s]
+
 # Interval to use for time-based splitting when a request is within the
 # `query_ingesters_within` window; defaults to `split-queries-by-interval` by
 # setting to 0.

pkg/bloomgateway/cache.go

@@ -182,6 +182,7 @@ func NewBloomGatewayClientCacheMiddleware(
 		},
 		cacheGen,
 		retentionEnabled,
+		false,
 	)
 
 	return &ClientCache{

pkg/querier/queryrange/index_stats_cache.go

@@ -123,6 +123,7 @@ func NewIndexStatsCacheMiddleware(
 		},
 		parallelismForReq,
 		retentionEnabled,
+		false,
 		metrics,
 	)
 }

pkg/querier/queryrange/labels_cache.go

@@ -11,21 +11,42 @@ import (
 	"github.com/grafana/loki/pkg/querier/queryrange/queryrangebase"
 	"github.com/grafana/loki/pkg/storage/chunk/cache"
 	"github.com/grafana/loki/pkg/storage/chunk/cache/resultscache"
-	"github.com/grafana/loki/pkg/util"
+	"github.com/grafana/loki/pkg/util/validation"
 )
 
 type cacheKeyLabels struct {
 	Limits
 	transformer UserIDTransformer
-	iqo         util.IngesterQueryOptions
+}
+
+// metadataSplitIntervalForTimeRange returns split interval for series and label requests.
+// If `recent_metadata_query_window` is configured and the query start interval is within this window,
+// it returns `split_recent_metadata_queries_by_interval`.
+// For other cases, the default split interval of `split_metadata_queries_by_interval` will be used.
+func metadataSplitIntervalForTimeRange(limits Limits, tenantIDs []string, ref, start time.Time) time.Duration {
+	split := validation.MaxDurationOrZeroPerTenant(tenantIDs, limits.MetadataQuerySplitDuration)
+
+	recentMetadataQueryWindow := validation.MaxDurationOrZeroPerTenant(tenantIDs, limits.RecentMetadataQueryWindow)
+	recentMetadataQuerySplitInterval := validation.MaxDurationOrZeroPerTenant(tenantIDs, limits.RecentMetadataQuerySplitDuration)
+
+	// if either of the options are not configured, use the default metadata split interval
+	if recentMetadataQueryWindow == 0 || recentMetadataQuerySplitInterval == 0 {
+		return split
+	}
+
+	// if the query start is not before window start, it would be split using recentMetadataQuerySplitInterval
+	if windowStart := ref.Add(-recentMetadataQueryWindow); !start.Before(windowStart) {
+		split = recentMetadataQuerySplitInterval
+	}
+
+	return split
 }
 
 // GenerateCacheKey generates a cache key based on the userID, split duration and the interval of the request.
 // It also includes the label name and the provided query for label values request.
 func (i cacheKeyLabels) GenerateCacheKey(ctx context.Context, userID string, r resultscache.Request) string {
 	lr := r.(*LabelRequest)
-
-	split := SplitIntervalForTimeRange(i.iqo, i.Limits, i.MetadataQuerySplitDuration, []string{userID}, time.Now().UTC(), r.GetEnd().UTC())
+	split := metadataSplitIntervalForTimeRange(i.Limits, []string{userID}, time.Now().UTC(), r.GetStart().UTC())
 
 	var currentInterval int64
 	if denominator := int64(split / time.Millisecond); denominator > 0 {
@@ -80,7 +101,6 @@ func NewLabelsCacheMiddleware(
 	merger queryrangebase.Merger,
 	c cache.Cache,
 	cacheGenNumberLoader queryrangebase.CacheGenNumberLoader,
-	iqo util.IngesterQueryOptions,
 	shouldCache queryrangebase.ShouldCacheFn,
 	parallelismForReq queryrangebase.ParallelismForReqFn,
 	retentionEnabled bool,
@@ -90,7 +110,7 @@ func NewLabelsCacheMiddleware(
 	return queryrangebase.NewResultsCacheMiddleware(
 		logger,
 		c,
-		cacheKeyLabels{limits, transformer, iqo},
+		cacheKeyLabels{limits, transformer},
 		limits,
 		merger,
 		labelsExtractor{},
@@ -100,6 +120,7 @@ func NewLabelsCacheMiddleware(
 		},
 		parallelismForReq,
 		retentionEnabled,
+		true,
 		metrics,
 	)
 }