Add concurrent evaluation for ruler

CHANGELOG.md

@@ -10,6 +10,7 @@
 * [FEATURE] Query Frontend/Scheduler: Add query priority support. #5605
 * [FEATURE] Tracing: Add `kuberesolver` to resolve endpoints address with `kubernetes://` prefix as Kubernetes service. #5731
 * [FEATURE] Tracing: Add `tracing.otel.round-robin` flag to use `round_robin` gRPC client side LB policy for sending OTLP traces. #5731
+* [FEATURE] Ruler: Add `ruler.concurrent-evals-enabled` flag to enable concurrent evaluation within a single rule group for independent rules. Maximum concurrency can be configured via `ruler.max-concurrent-evals`. #5766
 * [ENHANCEMENT] Store Gateway: Added `-store-gateway.enabled-tenants` and `-store-gateway.disabled-tenants` to explicitly enable or disable store-gateway for specific tenants. #5638
 * [ENHANCEMENT] Compactor: Add new compactor metric `cortex_compactor_start_duration_seconds`. #5683
 * [ENHANCEMENT] Upgraded Docker base images to `alpine:3.18`. #5684

docs/configuration/config-file-reference.md

@@ -3968,6 +3968,16 @@ alertmanager_client:
 # CLI flag: -ruler.resend-delay
 [resend_delay: <duration> | default = 1m]
 
+# If enabled, rules from a single rule group can be evaluated concurrently if
+# there is no dependency between each other. Max concurrency for each rule group
+# is controlled via ruler.max-concurrent-evals flag.
+# CLI flag: -ruler.concurrent-evals-enabled
+[concurrent_evals_enabled: <boolean> | default = false]
+
+# Max concurrency for a single rule group to evaluate independent rules.
+# CLI flag: -ruler.max-concurrent-evals
+[max_concurrent_evals: <int> | default = 1]
+
 # Distribute rule evaluation using ring backend
 # CLI flag: -ruler.enable-sharding
 [enable_sharding: <boolean> | default = false]

go.mod

@@ -46,7 +46,7 @@ require (
 	github.com/prometheus/client_model v0.5.0
 	github.com/prometheus/common v0.46.0
 	// Prometheus maps version 2.x.y to tags v0.x.y.
-	github.com/prometheus/prometheus v0.49.2-0.20240202164002-aa845f7c12ce
+	github.com/prometheus/prometheus v0.49.2-0.20240205174859-6005ac6f9dc6
 	github.com/segmentio/fasthash v1.0.3
 	github.com/sony/gobreaker v0.5.0
 	github.com/spf13/afero v1.9.5

go.sum

@@ -1320,8 +1320,8 @@ github.com/prometheus/procfs v0.8.0/go.mod h1:z7EfXMXOkbkqb9IINtpCn86r/to3BnA0ua
 github.com/prometheus/procfs v0.9.0/go.mod h1:+pB4zwohETzFnmlpe6yd2lSc+0/46IYZRB/chUwxUZY=
 github.com/prometheus/procfs v0.12.0 h1:jluTpSng7V9hY0O2R9DzzJHYb2xULk9VTR1V1R/k6Bo=
 github.com/prometheus/procfs v0.12.0/go.mod h1:pcuDEFsWDnvcgNzo4EEweacyhjeA9Zk3cnaOZAZEfOo=
-github.com/prometheus/prometheus v0.49.2-0.20240202164002-aa845f7c12ce h1:gHCPxX6dJZJOZh/nYy0DmnTu2PbgWjs8hY0eLgofPfA=
-github.com/prometheus/prometheus v0.49.2-0.20240202164002-aa845f7c12ce/go.mod h1:FvE8dtQ1Ww63IlyKBn1V4s+zMwF9kHkVNkQBR1pM4CU=
+github.com/prometheus/prometheus v0.49.2-0.20240205174859-6005ac6f9dc6 h1:E1dnG12fSlUeHST75LpGqPpd/YCOSNqKD2CUmm3Em90=
+github.com/prometheus/prometheus v0.49.2-0.20240205174859-6005ac6f9dc6/go.mod h1:FvE8dtQ1Ww63IlyKBn1V4s+zMwF9kHkVNkQBR1pM4CU=
 github.com/redis/rueidis v1.0.14-go1.18 h1:dGir5z8w8X1ex7JWO/Zx2FMBrZgQ8Yjm+lw9fPLSNGw=
 github.com/redis/rueidis v1.0.14-go1.18/go.mod h1:HGekzV3HbmzFmRK6j0xic8Z9119+ECoGMjeN1TV1NYU=
 github.com/remyoudompheng/bigfft v0.0.0-20200410134404-eec4a21b6bb0/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=

pkg/compactor/compactor_test.go

@@ -7,6 +7,7 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
 	"os"
 	"path"
 	"path/filepath"
@@ -30,7 +31,6 @@ import (
 	"github.com/thanos-io/thanos/pkg/block"
 	"github.com/thanos-io/thanos/pkg/block/metadata"
 	"github.com/thanos-io/thanos/pkg/compact"
-	thanos_testutil "github.com/thanos-io/thanos/pkg/testutil/e2eutil"
 	"gopkg.in/yaml.v2"
 
 	"github.com/cortexproject/cortex/pkg/ring"
@@ -1065,8 +1065,19 @@ func TestCompactor_ShouldSkipOutOrOrderBlocks(t *testing.T) {
 	b1 := createTSDBBlock(t, bucketClient, "user-1", 10, 20, map[string]string{"__name__": "Teste"})
 	b2 := createTSDBBlock(t, bucketClient, "user-1", 20, 30, map[string]string{"__name__": "Teste"})
 
-	err := thanos_testutil.PutOutOfOrderIndex(path.Join(tmpDir, "user-1", b1.String()), 10, 20)
+	// Read bad index file.
+	indexFile, err := os.Open("testdata/out_of_order_chunks/index")
 	require.NoError(t, err)
+	indexFileStat, err := indexFile.Stat()
+	require.NoError(t, err)
+
+	dir := path.Join(tmpDir, "user-1", b1.String())
+	outputFile, err := os.OpenFile(path.Join(dir, "index"), os.O_RDWR|os.O_TRUNC, 0755)
+	require.NoError(t, err)
+
+	n, err := io.Copy(outputFile, indexFile)
+	require.NoError(t, err)
+	require.Equal(t, indexFileStat.Size(), n)
 
 	cfg := prepareConfig()
 	cfg.SkipBlocksWithOutOfOrderChunksEnabled = true
@@ -1097,7 +1108,7 @@ func TestCompactor_ShouldSkipOutOrOrderBlocks(t *testing.T) {
 
 	// Wait until a run has completed.
 	cortex_testutil.Poll(t, 5*time.Second, true, func() interface{} {
-		if _, err := os.Stat(path.Join(tmpDir, "user-1", b1.String(), "no-compact-mark.json")); err == nil {
+		if _, err := os.Stat(path.Join(dir, "no-compact-mark.json")); err == nil {
 			return true
 		}
 		return false

pkg/ruler/compat.go

@@ -325,17 +325,19 @@ func DefaultTenantManagerFactory(cfg Config, p Pusher, q storage.Queryable, engi
 		metricsQueryFunc := MetricsQueryFunc(engineQueryFunc, totalQueries, failedQueries)
 
 		return rules.NewManager(&rules.ManagerOptions{
-			Appendable:      NewPusherAppendable(p, userID, overrides, totalWrites, failedWrites),
-			Queryable:       q,
-			QueryFunc:       RecordAndReportRuleQueryMetrics(metricsQueryFunc, queryTime, logger),
-			Context:         user.InjectOrgID(ctx, userID),
-			ExternalURL:     cfg.ExternalURL.URL,
-			NotifyFunc:      SendAlerts(notifier, cfg.ExternalURL.URL.String()),
-			Logger:          log.With(logger, "user", userID),
-			Registerer:      reg,
-			OutageTolerance: cfg.OutageTolerance,
-			ForGracePeriod:  cfg.ForGracePeriod,
-			ResendDelay:     cfg.ResendDelay,
+			Appendable:             NewPusherAppendable(p, userID, overrides, totalWrites, failedWrites),
+			Queryable:              q,
+			QueryFunc:              RecordAndReportRuleQueryMetrics(metricsQueryFunc, queryTime, logger),
+			Context:                user.InjectOrgID(ctx, userID),
+			ExternalURL:            cfg.ExternalURL.URL,
+			NotifyFunc:             SendAlerts(notifier, cfg.ExternalURL.URL.String()),
+			Logger:                 log.With(logger, "user", userID),
+			Registerer:             reg,
+			OutageTolerance:        cfg.OutageTolerance,
+			ForGracePeriod:         cfg.ForGracePeriod,
+			ResendDelay:            cfg.ResendDelay,
+			ConcurrentEvalsEnabled: cfg.ConcurrentEvalsEnabled,
+			MaxConcurrentEvals:     cfg.MaxConcurrentEvals,
 		})
 	}
 }

pkg/ruler/ruler.go

@@ -44,8 +44,9 @@ var (
 	supportedShardingStrategies = []string{util.ShardingStrategyDefault, util.ShardingStrategyShuffle}
 
 	// Validation errors.
-	errInvalidShardingStrategy = errors.New("invalid sharding strategy")
-	errInvalidTenantShardSize  = errors.New("invalid tenant shard size, the value must be greater than 0")
+	errInvalidShardingStrategy   = errors.New("invalid sharding strategy")
+	errInvalidTenantShardSize    = errors.New("invalid tenant shard size, the value must be greater than 0")
+	errInvalidMaxConcurrentEvals = errors.New("invalid max concurrent evals, the value must be greater than 0")
 )
 
 const (
@@ -95,7 +96,7 @@ type Config struct {
 	RulePath string `yaml:"rule_path"`
 
 	// URL of the Alertmanager to send notifications to.
-	// If your are configuring the ruler to send to a Cortex Alertmanager,
+	// If you are configuring the ruler to send to a Cortex Alertmanager,
 	// ensure this includes any path set in the Alertmanager external URL.
 	AlertmanagerURL string `yaml:"alertmanager_url"`
 	// Whether to use DNS SRV records to discover Alertmanager.
@@ -118,6 +119,9 @@ type Config struct {
 	// Minimum amount of time to wait before resending an alert to Alertmanager.
 	ResendDelay time.Duration `yaml:"resend_delay"`
 
+	ConcurrentEvalsEnabled bool  `yaml:"concurrent_evals_enabled"`
+	MaxConcurrentEvals     int64 `yaml:"max_concurrent_evals"`
+
 	// Enable sharding rule groups.
 	EnableSharding   bool          `yaml:"enable_sharding"`
 	ShardingStrategy string        `yaml:"sharding_strategy"`
@@ -149,6 +153,10 @@ func (cfg *Config) Validate(limits validation.Limits, log log.Logger) error {
 	if err := cfg.ClientTLSConfig.Validate(log); err != nil {
 		return errors.Wrap(err, "invalid ruler gRPC client config")
 	}
+
+	if cfg.ConcurrentEvalsEnabled && cfg.MaxConcurrentEvals <= 0 {
+		return errInvalidMaxConcurrentEvals
+	}
 	return nil
 }
 
@@ -188,6 +196,8 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
 	f.DurationVar(&cfg.OutageTolerance, "ruler.for-outage-tolerance", time.Hour, `Max time to tolerate outage for restoring "for" state of alert.`)
 	f.DurationVar(&cfg.ForGracePeriod, "ruler.for-grace-period", 10*time.Minute, `Minimum duration between alert and restored "for" state. This is maintained only for alerts with configured "for" time greater than grace period.`)
 	f.DurationVar(&cfg.ResendDelay, "ruler.resend-delay", time.Minute, `Minimum amount of time to wait before resending an alert to Alertmanager.`)
+	f.BoolVar(&cfg.ConcurrentEvalsEnabled, "ruler.concurrent-evals-enabled", false, `If enabled, rules from a single rule group can be evaluated concurrently if there is no dependency between each other. Max concurrency for each rule group is controlled via ruler.max-concurrent-evals flag.`)
+	f.Int64Var(&cfg.MaxConcurrentEvals, "ruler.max-concurrent-evals", 1, `Max concurrency for a single rule group to evaluate independent rules.`)
 
 	f.Var(&cfg.EnabledTenants, "ruler.enabled-tenants", "Comma separated list of tenants whose rules this ruler can evaluate. If specified, only these tenants will be handled by ruler, otherwise this ruler can process rules from all tenants. Subject to sharding.")
 	f.Var(&cfg.DisabledTenants, "ruler.disabled-tenants", "Comma separated list of tenants whose rules this ruler cannot evaluate. If specified, a ruler that would normally pick the specified tenant(s) for processing will ignore them instead. Subject to sharding.")