Implement 'batch mode' for persisting allocations on the client.

Fixes #9047, see problem details there.

As a solution, we use BoltDB's 'Batch' mode that combines multiple
parallel writes into small number of transactions. See
https://github.com/boltdb/bolt#batch-read-write-transactions for
more information.

client/allocrunner/alloc_runner.go

@@ -888,31 +888,33 @@ func (ar *allocRunner) PersistState() error {
 	defer ar.destroyedLock.Unlock()
 
 	if ar.destroyed {
-		err := ar.stateDB.DeleteAllocationBucket(ar.id)
+		err := ar.stateDB.DeleteAllocationBucket(ar.id, cstate.WithBatchMode())
 		if err != nil {
 			ar.logger.Warn("failed to delete allocation bucket", "error", err)
 		}
 		return nil
 	}
 
 	// persist network status, wrapping in a func to release state lock as early as possible
-	if err := func() error {
+	err := func() error {
 		ar.stateLock.Lock()
 		defer ar.stateLock.Unlock()
 		if ar.state.NetworkStatus != nil {
-			if err := ar.stateDB.PutNetworkStatus(ar.id, ar.state.NetworkStatus); err != nil {
+			err := ar.stateDB.PutNetworkStatus(ar.id, ar.state.NetworkStatus, cstate.WithBatchMode())
+			if err != nil {
 				return err
 			}
 		}
 		return nil
-	}(); err != nil {
+	}()
+	if err != nil {
 		return err
 	}
 
 	// TODO: consider persisting deployment state along with task status.
 	// While we study why only the alloc is persisted, I opted to maintain current
 	// behavior and not risk adding yet more IO calls unnecessarily.
-	return ar.stateDB.PutAllocation(ar.Alloc())
+	return ar.stateDB.PutAllocation(ar.Alloc(), cstate.WithBatchMode())
 }
 
 // Destroy the alloc runner by stopping it if it is still running and cleaning

client/state/db_test.go

@@ -4,6 +4,7 @@ import (
 	"io/ioutil"
 	"os"
 	"reflect"
+	"sync"
 	"testing"
 
 	trstate "github.com/hashicorp/nomad/client/allocrunner/taskrunner/state"
@@ -57,7 +58,7 @@ func testDB(t *testing.T, f func(*testing.T, StateDB)) {
 	}
 }
 
-// TestStateDB asserts the behavior of GetAllAllocations, PutAllocation, and
+// TestStateDB_Allocations asserts the behavior of GetAllAllocations, PutAllocation, and
 // DeleteAllocationBucket for all operational StateDB implementations.
 func TestStateDB_Allocations(t *testing.T) {
 	t.Parallel()
@@ -137,6 +138,102 @@ func TestStateDB_Allocations(t *testing.T) {
 	})
 }
 
+// TestStateDB_Batch asserts the behavior of PutAllocation, PutNetworkStatus and
+// DeleteAllocationBucket in batch mode, for all operational StateDB implementations.
+func TestStateDB_Batch(t *testing.T) {
+	t.Parallel()
+
+	testDB(t, func(t *testing.T, db StateDB) {
+		require := require.New(t)
+
+		// For BoltDB, get initial tx_id
+		var getTxID func() int
+		var prevTxID int
+		if boltStateDB, ok := db.(*BoltStateDB); ok {
+			boltdb := boltStateDB.DB().BoltDB()
+			getTxID = func() int {
+				tx, err := boltdb.Begin(true)
+				require.NoError(err)
+				defer tx.Rollback()
+				return tx.ID()
+			}
+			prevTxID = getTxID()
+		}
+
+		// Write 1000 allocations and network statuses in batch mode
+		var allocs []*structs.Allocation
+		for i := 0; i < 1000; i++ {
+			allocs = append(allocs, mock.Alloc())
+		}
+		var wg sync.WaitGroup
+		for _, alloc := range allocs {
+			wg.Add(1)
+			go func(alloc *structs.Allocation) {
+				require.NoError(db.PutNetworkStatus(alloc.ID, mock.AllocNetworkStatus(), WithBatchMode()))
+				require.NoError(db.PutAllocation(alloc, WithBatchMode()))
+				wg.Done()
+			}(alloc)
+		}
+		wg.Wait()
+
+		// Check BoltDB actually combined PutAllocation calls into much fewer transactions.
+		// The actual number of transactions depends on how fast the goroutines are spawned,
+		// with every 10ms period saved in a separate transaction (see boltdb MaxBatchDelay
+		// and MaxBatchSize parameters).
+		if getTxID != nil {
+			numTransactions := getTxID() - prevTxID
+			require.Less(numTransactions, 10)
+			prevTxID = getTxID()
+		}
+
+		// Retrieve allocs and make sure they are the same (order can differ)
+		readAllocs, errs, err := db.GetAllAllocations()
+		require.NoError(err)
+		require.NotNil(readAllocs)
+		require.Len(readAllocs, len(allocs))
+		require.NotNil(errs)
+		require.Empty(errs)
+
+		readAllocsById := make(map[string]*structs.Allocation)
+		for _, readAlloc := range readAllocs {
+			readAllocsById[readAlloc.ID] = readAlloc
+		}
+		for _, alloc := range allocs {
+			readAlloc, ok := readAllocsById[alloc.ID]
+			if !ok {
+				t.Fatalf("no alloc with ID=%q", alloc.ID)
+			}
+			if !reflect.DeepEqual(readAlloc, alloc) {
+				pretty.Ldiff(t, readAlloc, alloc)
+				t.Fatalf("alloc %q unequal", alloc.ID)
+			}
+		}
+
+		// Delete all allocs in batch mode
+		for _, alloc := range allocs {
+			wg.Add(1)
+			go func(alloc *structs.Allocation) {
+				require.NoError(db.DeleteAllocationBucket(alloc.ID, WithBatchMode()))
+				wg.Done()
+			}(alloc)
+		}
+		wg.Wait()
+
+		// Check BoltDB combined DeleteAllocationBucket calls into much fewer transactions.
+		if getTxID != nil {
+			numTransactions := getTxID() - prevTxID
+			require.Less(numTransactions, 10)
+			prevTxID = getTxID()
+		}
+
+		// Check all allocs were deleted.
+		readAllocs, errs, err = db.GetAllAllocations()
+		require.NoError(err)
+		require.Empty(readAllocs)
+		require.Empty(errs)
+	})
+}
+
 // TestStateDB_TaskState asserts the behavior of task state related StateDB
 // methods.
 func TestStateDB_TaskState(t *testing.T) {

client/state/errdb.go

@@ -5,6 +5,7 @@ import (
 
 	"github.com/hashicorp/nomad/client/allocrunner/taskrunner/state"
 	dmstate "github.com/hashicorp/nomad/client/devicemanager/state"
+	"github.com/hashicorp/nomad/client/dynamicplugins"
 	driverstate "github.com/hashicorp/nomad/client/pluginmanager/drivermanager/state"
 	"github.com/hashicorp/nomad/nomad/structs"
 )
@@ -27,7 +28,7 @@ func (m *ErrDB) GetAllAllocations() ([]*structs.Allocation, map[string]error, er
 	return m.Allocs, nil, nil
 }
 
-func (m *ErrDB) PutAllocation(alloc *structs.Allocation) error {
+func (m *ErrDB) PutAllocation(alloc *structs.Allocation, opts ...WriteOption) error {
 	return fmt.Errorf("Error!")
 }
 
@@ -43,7 +44,7 @@ func (m *ErrDB) GetNetworkStatus(allocID string) (*structs.AllocNetworkStatus, e
 	return nil, fmt.Errorf("Error!")
 }
 
-func (m *ErrDB) PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus) error {
+func (m *ErrDB) PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus, opts ...WriteOption) error {
 	return fmt.Errorf("Error!")
 }
 
@@ -63,14 +64,22 @@ func (m *ErrDB) DeleteTaskBucket(allocID, taskName string) error {
 	return fmt.Errorf("Error!")
 }
 
-func (m *ErrDB) DeleteAllocationBucket(allocID string) error {
+func (m *ErrDB) DeleteAllocationBucket(allocID string, opts ...WriteOption) error {
 	return fmt.Errorf("Error!")
 }
 
 func (m *ErrDB) PutDevicePluginState(ps *dmstate.PluginState) error {
 	return fmt.Errorf("Error!")
 }
 
+func (m *ErrDB) GetDynamicPluginRegistryState() (*dynamicplugins.RegistryState, error) {
+	return nil, fmt.Errorf("Error!")
+}
+
+func (m *ErrDB) PutDynamicPluginRegistryState(state *dynamicplugins.RegistryState) error {
+	return fmt.Errorf("Error!")
+}
+
 // GetDevicePluginState stores the device manager's plugin state or returns an
 // error.
 func (m *ErrDB) GetDevicePluginState() (*dmstate.PluginState, error) {
@@ -88,3 +97,6 @@ func (m *ErrDB) PutDriverPluginState(ps *driverstate.PluginState) error {
 func (m *ErrDB) Close() error {
 	return fmt.Errorf("Error!")
 }
+
+// Ensure *ErrDB implements StateDB
+var _ StateDB = (*ErrDB)(nil)

client/state/interface.go

@@ -24,9 +24,9 @@ type StateDB interface {
 	// If a single error is returned then both allocations and the map will be nil.
 	GetAllAllocations() ([]*structs.Allocation, map[string]error, error)
 
-	// PulAllocation stores an allocation or returns an error if it could
+	// PutAllocation stores an allocation or returns an error if it could
 	// not be stored.
-	PutAllocation(*structs.Allocation) error
+	PutAllocation(*structs.Allocation, ...WriteOption) error
 
 	// Get/Put DeploymentStatus get and put the allocation's deployment
 	// status. It may be nil.
@@ -36,7 +36,7 @@ type StateDB interface {
 	// Get/Put NetworkStatus get and put the allocation's network
 	// status. It may be nil.
 	GetNetworkStatus(allocID string) (*structs.AllocNetworkStatus, error)
-	PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus) error
+	PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus, opts ...WriteOption) error
 
 	// GetTaskRunnerState returns the LocalState and TaskState for a
 	// TaskRunner. Either state may be nil if it is not found, but if an
@@ -57,7 +57,7 @@ type StateDB interface {
 
 	// DeleteAllocationBucket deletes an allocation's state bucket if it
 	// exists. No error is returned if it does not exist.
-	DeleteAllocationBucket(allocID string) error
+	DeleteAllocationBucket(allocID string, opts ...WriteOption) error
 
 	// GetDevicePluginState is used to retrieve the device manager's plugin
 	// state.
@@ -78,10 +78,43 @@ type StateDB interface {
 	// GetDynamicPluginRegistryState is used to retrieve a dynamic plugin manager's state.
 	GetDynamicPluginRegistryState() (*dynamicplugins.RegistryState, error)
 
-	// PutDynamicPluginRegistryState is used to store the dynamic plugin managers's state.
+	// PutDynamicPluginRegistryState is used to store the dynamic plugin manager's state.
 	PutDynamicPluginRegistryState(state *dynamicplugins.RegistryState) error
 
 	// Close the database. Unsafe for further use after calling regardless
 	// of return value.
 	Close() error
 }
+
+// WriteOptions adjusts the way the data is persisted by the StateDB above. Default is
+// zero/false values for all fields. To provide different values, use With* functions
+// below, like this: statedb.PutAllocation(alloc, WithBatchMode())
+type WriteOptions struct {
+	// In Batch mode, concurrent writes (Put* and Delete* operations above) are
+	// coalesced into a single transaction, increasing write performance. To benefit
+	// from this mode, writes must happen concurrently in goroutines, as every write
+	// request still waits for the shared transaction to commit before returning.
+	// See https://github.com/boltdb/bolt#batch-read-write-transactions for details.
+	// This mode is only supported for BoltDB state backend and is ignored in other backends.
+	BatchMode bool
+}
+
+// WriteOption is a function that modifies WriteOptions struct above.
+type WriteOption func(*WriteOptions)
+
+// mergeWriteOptions creates a final WriteOptions struct to be used by the write methods above
+// from a list of WriteOption-s provided as variadic arguments.
+func mergeWriteOptions(opts []WriteOption) WriteOptions {
+	writeOptions := WriteOptions{} // Default WriteOptions is zero value.
+	for _, opt := range opts {
+		opt(&writeOptions)
+	}
+	return writeOptions
+}
+
+// Enable Batch mode for write requests (Put* and Delete* operations above).
+func WithBatchMode() WriteOption {
+	return func(s *WriteOptions) {
+		s.BatchMode = true
+	}
+}

client/state/memdb.go

@@ -73,7 +73,7 @@ func (m *MemDB) GetAllAllocations() ([]*structs.Allocation, map[string]error, er
 	return allocs, map[string]error{}, nil
 }
 
-func (m *MemDB) PutAllocation(alloc *structs.Allocation) error {
+func (m *MemDB) PutAllocation(alloc *structs.Allocation, opts ...WriteOption) error {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 	m.allocs[alloc.ID] = alloc
@@ -99,7 +99,7 @@ func (m *MemDB) GetNetworkStatus(allocID string) (*structs.AllocNetworkStatus, e
 	return m.networkStatus[allocID], nil
 }
 
-func (m *MemDB) PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus) error {
+func (m *MemDB) PutNetworkStatus(allocID string, ns *structs.AllocNetworkStatus, opts ...WriteOption) error {
 	m.mu.Lock()
 	m.networkStatus[allocID] = ns
 	defer m.mu.Unlock()
@@ -175,7 +175,7 @@ func (m *MemDB) DeleteTaskBucket(allocID, taskName string) error {
 	return nil
 }
 
-func (m *MemDB) DeleteAllocationBucket(allocID string) error {
+func (m *MemDB) DeleteAllocationBucket(allocID string, opts ...WriteOption) error {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 

client/state/noopdb.go

@@ -23,7 +23,7 @@ func (n NoopDB) GetAllAllocations() ([]*structs.Allocation, map[string]error, er
 	return nil, nil, nil
 }
 
-func (n NoopDB) PutAllocation(*structs.Allocation) error {
+func (n NoopDB) PutAllocation(alloc *structs.Allocation, opts ...WriteOption) error {
 	return nil
 }
 
@@ -39,7 +39,7 @@ func (n NoopDB) GetNetworkStatus(allocID string) (*structs.AllocNetworkStatus, e
 	return nil, nil
 }
 
-func (n NoopDB) PutNetworkStatus(allocID string, ds *structs.AllocNetworkStatus) error {
+func (n NoopDB) PutNetworkStatus(allocID string, ds *structs.AllocNetworkStatus, opts ...WriteOption) error {
 	return nil
 }
 
@@ -59,7 +59,7 @@ func (n NoopDB) DeleteTaskBucket(allocID, taskName string) error {
 	return nil
 }
 
-func (n NoopDB) DeleteAllocationBucket(allocID string) error {
+func (n NoopDB) DeleteAllocationBucket(allocID string, opts ...WriteOption) error {
 	return nil
 }
 

client/state/state_database.go

@@ -234,8 +234,8 @@ func (s *BoltStateDB) getAllAllocations(tx *boltdd.Tx) ([]*structs.Allocation, m
 }
 
 // PutAllocation stores an allocation or returns an error.
-func (s *BoltStateDB) PutAllocation(alloc *structs.Allocation) error {
-	return s.db.Update(func(tx *boltdd.Tx) error {
+func (s *BoltStateDB) PutAllocation(alloc *structs.Allocation, opts ...WriteOption) error {
+	return s.updateWithOptions(opts, func(tx *boltdd.Tx) error {
 		// Retrieve the root allocations bucket
 		allocsBkt, err := tx.CreateBucketIfNotExists(allocationsBucketName)
 		if err != nil {
@@ -321,8 +321,8 @@ type networkStatusEntry struct {
 
 // PutDeploymentStatus stores an allocation's DeploymentStatus or returns an
 // error.
-func (s *BoltStateDB) PutNetworkStatus(allocID string, ds *structs.AllocNetworkStatus) error {
-	return s.db.Update(func(tx *boltdd.Tx) error {
+func (s *BoltStateDB) PutNetworkStatus(allocID string, ds *structs.AllocNetworkStatus, opts ...WriteOption) error {
+	return s.updateWithOptions(opts, func(tx *boltdd.Tx) error {
 		return putNetworkStatusImpl(tx, allocID, ds)
 	})
 }
@@ -493,8 +493,8 @@ func (s *BoltStateDB) DeleteTaskBucket(allocID, taskName string) error {
 }
 
 // DeleteAllocationBucket is used to delete an allocation bucket if it exists.
-func (s *BoltStateDB) DeleteAllocationBucket(allocID string) error {
-	return s.db.Update(func(tx *boltdd.Tx) error {
+func (s *BoltStateDB) DeleteAllocationBucket(allocID string, opts ...WriteOption) error {
+	return s.updateWithOptions(opts, func(tx *boltdd.Tx) error {
 		// Retrieve the root allocations bucket
 		allocations := tx.Bucket(allocationsBucketName)
 		if allocations == nil {
@@ -725,6 +725,19 @@ func (s *BoltStateDB) init() error {
 	})
 }
 
+// updateWithOptions enables adjustments to db.Update operation, including Batch mode.
+func (s *BoltStateDB) updateWithOptions(opts []WriteOption, updateFn func(tx *boltdd.Tx) error) error {
+	writeOpts := mergeWriteOptions(opts)
+
+	if writeOpts.BatchMode {
+		// In Batch mode, BoltDB opportunistically combines multiple concurrent writes into one or
+		// several transactions. See boltdb.Batch() documentation for details.
+		return s.db.Batch(updateFn)
+	} else {
+		return s.db.Update(updateFn)
+	}
+}
+
 // Upgrade bolt state db from 0.8 schema to 0.9 schema. Noop if already using
 // 0.9 schema. Creates a backup before upgrading.
 func (s *BoltStateDB) Upgrade() error {