scheduler: fragment queue and querier pick-up coordination

docs/configuration/config-file-reference.md

@@ -3480,6 +3480,10 @@ grpc_client_config:
   # using default gRPC client connect timeout 20s.
   # CLI flag: -querier.frontend-client.connect-timeout
   [connect_timeout: <duration> | default = 5s]
+
+# Name of network interface to read address from.
+# CLI flag: -querier.instance-interface-names
+[instance_interface_names: <list of string> | default = [eth0 en0]]
 ```
 
 ### `ingester_config`

pkg/cortex/modules.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"flag"
 	"fmt"
-
 	"log/slog"
 	"net/http"
 	"runtime"
@@ -414,6 +413,9 @@ func (t *Cortex) initQuerier() (serv services.Service, err error) {
 
 	t.Cfg.Worker.MaxConcurrentRequests = t.Cfg.Querier.MaxConcurrent
 	t.Cfg.Worker.TargetHeaders = t.Cfg.API.HTTPRequestHeadersToLog
+
+	t.Cfg.Worker.ListenPort = t.Cfg.Server.GRPCListenPort
+
 	return querier_worker.NewQuerierWorker(t.Cfg.Worker, httpgrpc_server.NewServer(internalQuerierRouter), util_log.Logger, prometheus.DefaultRegisterer)
 }
 
@@ -813,7 +815,7 @@ func (t *Cortex) initQueryScheduler() (services.Service, error) {
 		tenant.WithDefaultResolver(tenantfederation.NewRegexValidator())
 	}
 
-	s, err := scheduler.NewScheduler(t.Cfg.QueryScheduler, t.Overrides, util_log.Logger, prometheus.DefaultRegisterer)
+	s, err := scheduler.NewScheduler(t.Cfg.QueryScheduler, t.Overrides, util_log.Logger, prometheus.DefaultRegisterer, t.Cfg.Querier.DistributedExecEnabled)
 	if err != nil {
 		return nil, errors.Wrap(err, "query-scheduler init")
 	}

pkg/distributed_execution/fragment_key.go

@@ -0,0 +1,33 @@
+package distributed_execution
+
+// FragmentKey uniquely identifies a fragment of a distributed logical query plan.
+// It combines a queryID (to identify the overall query) and a fragmentID
+// (to identify the specific fragment within that query).
+type FragmentKey struct {
+	// QueryID  identifies the distributed query this fragment belongs to
+	queryID uint64
+
+	// FragmentID identifies this specific fragment within the query
+	fragmentID uint64
+}
+
+// MakeFragmentKey creates a new FragmentKey with the given queryID and fragmentID.
+// It's used to track and identify fragments during distributed query execution.
+func MakeFragmentKey(queryID uint64, fragmentID uint64) FragmentKey {
+	return FragmentKey{
+		queryID:    queryID,
+		fragmentID: fragmentID,
+	}
+}
+
+// GetQueryID returns the queryID for the current key
+// This ID is shared across all fragments of the same distributed query.
+func (f FragmentKey) GetQueryID() uint64 {
+	return f.queryID
+}
+
+// GetFragmentID returns the ID for this specific fragment
+// within its parent query.
+func (f FragmentKey) GetFragmentID() uint64 {
+	return f.fragmentID
+}

pkg/distributed_execution/plan_fragments/fragmenter.go

@@ -0,0 +1,49 @@
+package plan_fragments
+
+import "github.com/thanos-io/promql-engine/logicalplan"
+
+type Fragmenter interface {
+	Fragment(node logicalplan.Node) ([]Fragment, error)
+}
+
+type DummyFragmenter struct {
+}
+
+func (f *DummyFragmenter) Fragment(node logicalplan.Node) ([]Fragment, error) {
+	// simple logic without distributed optimizer
+	return []Fragment{
+		{
+			Node:       node,
+			FragmentID: uint64(1),
+			ChildIDs:   []uint64{},
+			IsRoot:     true,
+		},
+	}, nil
+}
+
+type Fragment struct {
+	Node       logicalplan.Node
+	FragmentID uint64
+	ChildIDs   []uint64
+	IsRoot     bool
+}
+
+func (s *Fragment) IsEmpty() bool {
+	if s.Node != nil {
+		return false
+	}
+	if s.FragmentID != 0 {
+		return false
+	}
+	if s.IsRoot {
+		return false
+	}
+	if len(s.ChildIDs) != 0 {
+		return false
+	}
+	return true
+}
+
+func NewDummyFragmenter() Fragmenter {
+	return &DummyFragmenter{}
+}

pkg/distributed_execution/plan_fragments/fragmenter_test.go

@@ -0,0 +1,46 @@
+package plan_fragments
+
+import (
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/cortexproject/cortex/pkg/util/logical_plan"
+)
+
+func TestFragmenter(t *testing.T) {
+	type testCase struct {
+		name              string
+		query             string
+		start             time.Time
+		end               time.Time
+		expectedFragments int
+	}
+
+	now := time.Now()
+
+	// more tests will be added when distributed optimizer and fragmenter are implemented
+	tests := []testCase{
+		{
+			name:              "simple logical query plan - no fragmentation",
+			query:             "up",
+			start:             now,
+			end:               now,
+			expectedFragments: 1,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			lp, err := logical_plan.CreateTestLogicalPlan(tc.query, tc.start, tc.end, 0)
+			require.NoError(t, err)
+
+			fragmenter := NewDummyFragmenter()
+			res, err := fragmenter.Fragment((*lp).Root())
+
+			require.NoError(t, err)
+			require.Equal(t, tc.expectedFragments, len(res))
+		})
+	}
+}

pkg/querier/querier.go

@@ -96,7 +96,8 @@ type Config struct {
 	ParquetQueryableShardCacheSize    int    `yaml:"parquet_queryable_shard_cache_size"`
 	ParquetQueryableDefaultBlockStore string `yaml:"parquet_queryable_default_block_store"`
 	ParquetQueryableFallbackDisabled  bool   `yaml:"parquet_queryable_fallback_disabled"`
-	DistributedExecEnabled            bool   `yaml:"distributed_exec_enabled" doc:"hidden"`
+
+	DistributedExecEnabled bool `yaml:"distributed_exec_enabled" doc:"hidden"`
 }
 
 var (

pkg/querier/worker/scheduler_processor.go

@@ -31,7 +31,7 @@ import (
 	"github.com/cortexproject/cortex/pkg/util/services"
 )
 
-func newSchedulerProcessor(cfg Config, handler RequestHandler, log log.Logger, reg prometheus.Registerer) (*schedulerProcessor, []services.Service) {
+func newSchedulerProcessor(cfg Config, handler RequestHandler, log log.Logger, reg prometheus.Registerer, querierAddress string) (*schedulerProcessor, []services.Service) {
 	p := &schedulerProcessor{
 		log:            log,
 		handler:        handler,
@@ -47,6 +47,7 @@ func newSchedulerProcessor(cfg Config, handler RequestHandler, log log.Logger, r
 			Help:    "Time spend doing requests to frontend.",
 			Buckets: prometheus.ExponentialBuckets(0.001, 4, 6),
 		}, []string{"operation", "status_code"}),
+		querierAddress: querierAddress,
 	}
 
 	frontendClientsGauge := promauto.With(reg).NewGauge(prometheus.GaugeOpts{
@@ -71,6 +72,7 @@ type schedulerProcessor struct {
 	grpcConfig     grpcclient.Config
 	maxMessageSize int
 	querierID      string
+	querierAddress string
 
 	frontendPool                  *client.Pool
 	frontendClientRequestDuration *prometheus.HistogramVec
@@ -97,7 +99,7 @@ func (sp *schedulerProcessor) processQueriesOnSingleStream(ctx context.Context,
 	for backoff.Ongoing() {
 		c, err := schedulerClient.QuerierLoop(ctx)
 		if err == nil {
-			err = c.Send(&schedulerpb.QuerierToScheduler{QuerierID: sp.querierID})
+			err = c.Send(&schedulerpb.QuerierToScheduler{QuerierID: sp.querierID, QuerierAddress: sp.querierAddress})
 		}
 
 		if err != nil {

pkg/querier/worker/scheduler_processor_test.go

@@ -144,7 +144,7 @@ func Test_ToShowNotPanic_RelatedIssue6599(t *testing.T) {
 		go stat.AddFetchedChunkBytes(10)
 	}).Return(&httpgrpc.HTTPResponse{}, nil)
 
-	sp, _ := newSchedulerProcessor(cfg, requestHandler, log.NewNopLogger(), nil)
+	sp, _ := newSchedulerProcessor(cfg, requestHandler, log.NewNopLogger(), nil, "")
 	schedulerClient := &mockSchedulerForQuerierClient{}
 	schedulerClient.On("QuerierLoop", mock.Anything, mock.Anything).Return(querierLoopClient, nil)
 

pkg/querier/worker/worker.go

@@ -3,7 +3,9 @@ package worker
 import (
 	"context"
 	"flag"
+	"net"
 	"os"
+	"strconv"
 	"sync"
 	"time"
 
@@ -14,7 +16,9 @@ import (
 	"github.com/weaveworks/common/httpgrpc"
 	"google.golang.org/grpc"
 
+	"github.com/cortexproject/cortex/pkg/ring"
 	"github.com/cortexproject/cortex/pkg/util"
+	"github.com/cortexproject/cortex/pkg/util/flagext"
 	"github.com/cortexproject/cortex/pkg/util/grpcclient"
 	"github.com/cortexproject/cortex/pkg/util/services"
 )
@@ -33,6 +37,10 @@ type Config struct {
 	GRPCClientConfig grpcclient.Config `yaml:"grpc_client_config"`
 
 	TargetHeaders []string `yaml:"-"` // Propagated by config.
+
+	InstanceInterfaceNames []string `yaml:"instance_interface_names"`
+	ListenPort             int      `yaml:"-"`
+	InstanceAddr           string   `yaml:"instance_addr" doc:"hidden"`
 }
 
 func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
@@ -46,6 +54,10 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
 	f.StringVar(&cfg.QuerierID, "querier.id", "", "Querier ID, sent to frontend service to identify requests from the same querier. Defaults to hostname.")
 
 	cfg.GRPCClientConfig.RegisterFlagsWithPrefix("querier.frontend-client", "", f)
+
+	cfg.InstanceInterfaceNames = []string{"eth0", "en0"}
+	f.Var((*flagext.StringSlice)(&cfg.InstanceInterfaceNames), "querier.instance-interface-names", "Name of network interface to read address from.")
+	f.StringVar(&cfg.InstanceAddr, "querier.instance-addr", "", "IP address to advertise in the ring.")
 }
 
 func (cfg *Config) Validate(log log.Logger) error {
@@ -109,7 +121,14 @@ func NewQuerierWorker(cfg Config, handler RequestHandler, log log.Logger, reg pr
 		level.Info(log).Log("msg", "Starting querier worker connected to query-scheduler", "scheduler", cfg.SchedulerAddress)
 
 		address = cfg.SchedulerAddress
-		processor, servs = newSchedulerProcessor(cfg, handler, log, reg)
+
+		ipAddr, err := ring.GetInstanceAddr(cfg.InstanceAddr, cfg.InstanceInterfaceNames, log)
+		if err != nil {
+			return nil, err
+		}
+		querierAddr := net.JoinHostPort(ipAddr, strconv.Itoa(cfg.ListenPort))
+
+		processor, servs = newSchedulerProcessor(cfg, handler, log, reg, querierAddr)
 
 	case cfg.FrontendAddress != "":
 		level.Info(log).Log("msg", "Starting querier worker connected to query-frontend", "frontend", cfg.FrontendAddress)

pkg/scheduler/fragment_table/fragment_table.go

@@ -0,0 +1,70 @@
+package fragment_table
+
+import (
+	"sync"
+
+	"github.com/cortexproject/cortex/pkg/distributed_execution"
+)
+
+type FragmentTable struct {
+	mappings map[distributed_execution.FragmentKey]string
+	mu       sync.RWMutex
+}
+
+func NewFragmentTable() *FragmentTable {
+	return &FragmentTable{
+		mappings: make(map[distributed_execution.FragmentKey]string),
+	}
+}
+
+func (f *FragmentTable) AddMapping(queryID uint64, fragmentID uint64, addr string) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	key := distributed_execution.MakeFragmentKey(queryID, fragmentID)
+	f.mappings[key] = addr
+}
+
+func (f *FragmentTable) GetAllChildAddresses(queryID uint64, fragmentIDs []uint64) ([]string, bool) {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	addresses := make([]string, 0, len(fragmentIDs))
+
+	for _, fragmentID := range fragmentIDs {
+		key := distributed_execution.MakeFragmentKey(queryID, fragmentID)
+		if addr, ok := f.mappings[key]; ok {
+			addresses = append(addresses, addr)
+		} else {
+			return nil, false
+		}
+	}
+	return addresses, true
+}
+
+func (f *FragmentTable) GetChildAddr(queryID uint64, fragmentID uint64) (string, bool) {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+
+	key := distributed_execution.MakeFragmentKey(queryID, fragmentID)
+	if addr, ok := f.mappings[key]; ok {
+		return addr, true
+	}
+	return "", false
+}
+
+func (f *FragmentTable) ClearMappings(queryID uint64) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+
+	keysToDelete := make([]distributed_execution.FragmentKey, 0)
+	for key := range f.mappings {
+		if key.GetQueryID() == queryID {
+			keysToDelete = append(keysToDelete, key)
+		}
+	}
+
+	for _, key := range keysToDelete {
+		delete(f.mappings, key)
+	}
+}