Merge pull request #5203 from oasisprotocol/kostko/stable/22.2.x/back…

…port-5200

[BACKPORT/22.2.x] Move halt epoch from genesis to node-local configuration

.changelog/5200.feature.md

@@ -0,0 +1 @@
+Move halt epoch from genesis to node-local configuration

go/consensus/tendermint/abci/mux.go

@@ -67,12 +67,12 @@ var (
 
 // ApplicationConfig is the configuration for the consensus application.
 type ApplicationConfig struct { // nolint: maligned
-	DataDir         string
-	StorageBackend  string
-	Pruning         PruneConfig
-	HaltEpochHeight beacon.EpochTime
-	HaltBlockHeight uint64
-	MinGasPrice     uint64
+	DataDir        string
+	StorageBackend string
+	Pruning        PruneConfig
+	HaltEpoch      beacon.EpochTime
+	HaltHeight     uint64
+	MinGasPrice    uint64
 
 	DisableCheckpointer       bool
 	CheckpointerCheckInterval time.Duration
@@ -440,7 +440,7 @@ func (mux *abciMux) BeginBlock(req types.RequestBeginBlock) types.ResponseBeginB
 		// On transition, trigger halt hooks.
 		mux.logger.Info("BeginBlock: halt mode transition, emitting empty block",
 			"block_height", blockHeight,
-			"epoch", mux.state.haltEpochHeight,
+			"epoch", currentEpoch,
 		)
 		mux.logger.Debug("dispatching halt hooks before halt point")
 		mux.dispatchHaltHooks(blockHeight, currentEpoch, nil)

go/consensus/tendermint/abci/state.go

@@ -64,9 +64,9 @@ type applicationState struct { // nolint: maligned
 
 	timeSource beacon.Backend
 
-	haltMode        bool
-	haltEpochHeight beacon.EpochTime
-	haltBlockHeight uint64
+	haltMode   bool
+	haltEpoch  beacon.EpochTime
+	haltHeight uint64
 
 	minGasPrice        quantity.Quantity
 	ownTxSigner        signature.PublicKey
@@ -259,13 +259,13 @@ func (s *applicationState) shouldHalt(ctx *api.Context) bool {
 	// If there is a halt block height configured, and the height(+1)
 	// is greater than or equal to the height point, transition into
 	// the halting state.
-	if s.haltBlockHeight != 0 && uint64(h) >= s.haltBlockHeight {
+	if s.haltHeight != 0 && uint64(h) >= s.haltHeight {
 		s.logger.Info("shouldHalt: reached halt block height",
 			"block_height", h,
-			"halt_block_height", s.haltBlockHeight,
+			"halt_block_height", s.haltHeight,
 		)
 		s.haltMode = true
-	} else {
+	} else if s.haltEpoch > 0 {
 		// Otherwise, check to see if the epoch will be equal to
 		// the halt epoch (if any).
 		currentEpoch, err := s.GetEpoch(ctx, h)
@@ -276,7 +276,7 @@ func (s *applicationState) shouldHalt(ctx *api.Context) bool {
 			)
 			return false
 		}
-		s.haltMode = currentEpoch == s.haltEpochHeight
+		s.haltMode = currentEpoch == s.haltEpoch
 	}
 
 	return s.haltMode
@@ -589,8 +589,8 @@ func newApplicationState(ctx context.Context, upgrader upgrade.Backend, cfg *App
 		prunerNotifyCh:     channels.NewRingChannel(1),
 		pruneInterval:      cfg.Pruning.PruneInterval,
 		upgrader:           upgrader,
-		haltEpochHeight:    cfg.HaltEpochHeight,
-		haltBlockHeight:    cfg.HaltBlockHeight,
+		haltEpoch:          cfg.HaltEpoch,
+		haltHeight:         cfg.HaltHeight,
 		minGasPrice:        minGasPrice,
 		ownTxSigner:        cfg.OwnTxSigner,
 		ownTxSignerAddress: staking.NewAddress(cfg.OwnTxSigner),

go/consensus/tendermint/full/archive.go

@@ -165,7 +165,6 @@ func NewArchive(
 			Strategy:      abci.PruneNone,
 			PruneInterval: time.Hour * 100, // Irrelevant as pruning is disabled.
 		},
-		HaltEpochHeight:     srv.genesis.HaltEpoch,
 		OwnTxSigner:         srv.identity.NodeSigner.Public(),
 		DisableCheckpointer: true,
 		InitialHeight:       uint64(srv.genesis.Height),

go/consensus/tendermint/full/full.go

@@ -112,6 +112,8 @@ const (
 
 	// CfgHaltHeight is the block height at which the local node should be shutdown.
 	CfgHaltHeight = "consensus.tendermint.halt_height"
+	// CfgHaltEpoch is the epoch at which the local node should be shutdown.
+	CfgHaltEpoch = "consensus.tendermint.halt_epoch"
 )
 
 const (
@@ -579,8 +581,8 @@ func (t *fullService) lazyInit() error {
 		DataDir:                   filepath.Join(t.dataDir, tmcommon.StateDir),
 		StorageBackend:            db.GetBackendName(),
 		Pruning:                   pruneCfg,
-		HaltEpochHeight:           t.genesis.HaltEpoch,
-		HaltBlockHeight:           viper.GetUint64(CfgHaltHeight),
+		HaltEpoch:                 beaconAPI.EpochTime(viper.GetUint64(CfgHaltEpoch)),
+		HaltHeight:                viper.GetUint64(CfgHaltHeight),
 		MinGasPrice:               viper.GetUint64(CfgMinGasPrice),
 		OwnTxSigner:               t.identity.NodeSigner.Public(),
 		DisableCheckpointer:       viper.GetBool(CfgCheckpointerDisabled),
@@ -1015,7 +1017,8 @@ func init() {
 	Flags.Bool(CfgSupplementarySanityEnabled, false, "enable supplementary sanity checks (slows down consensus)")
 	Flags.Uint64(CfgSupplementarySanityInterval, 10, "supplementary sanity check interval (in blocks)")
 
-	Flags.Uint64(CfgHaltHeight, 0, "height at which to force-shutdown the node (in blocks)")
+	Flags.Uint64(CfgHaltHeight, 0, "height at which to force-shutdown the node (in blocks, zero to disable)")
+	Flags.Uint64(CfgHaltEpoch, 0, "epoch at which to force-shutdown the node (in epochs, zero to disable)")
 
 	// State sync.
 	Flags.Bool(CfgConsensusStateSyncEnabled, false, "enable state sync")

go/oasis-test-runner/oasis/network.go

@@ -29,6 +29,7 @@ import (
 	"github.com/oasisprotocol/oasis-core/go/common/quantity"
 	"github.com/oasisprotocol/oasis-core/go/consensus/api/transaction"
 	consensusGenesis "github.com/oasisprotocol/oasis-core/go/consensus/genesis"
+	tendermintFull "github.com/oasisprotocol/oasis-core/go/consensus/tendermint/full"
 	genesisFile "github.com/oasisprotocol/oasis-core/go/genesis/file"
 	genesisTestHelpers "github.com/oasisprotocol/oasis-core/go/genesis/tests"
 	governance "github.com/oasisprotocol/oasis-core/go/governance/api"
@@ -614,6 +615,13 @@ func (net *Network) startOasisNode(
 			tendermintDebugAddrBookLenient().
 			tendermintDebugAllowDuplicateIP().
 			tendermintUpgradeStopDelay(10 * time.Second)
+
+		if haltEpoch := net.cfg.HaltEpoch; haltEpoch > 0 {
+			extraArgs = extraArgs.extraArgs([]Argument{{
+				Name:   tendermintFull.CfgHaltEpoch,
+				Values: []string{strconv.FormatUint(haltEpoch, 10)},
+			}})
+		}
 	}
 	if net.cfg.UseShortGrpcSocketPaths {
 		// Keep the socket, if it was already generated!

go/oasis-test-runner/scenario/e2e/e2e.go

@@ -366,15 +366,15 @@ func (sc *E2E) DumpRestoreNetwork(
 		}
 	}
 
+	// If network is used, enable shorter per-node socket paths, because some e2e test datadir
+	// exceed maximum unix socket path length.
+	fixture.Network.UseShortGrpcSocketPaths = true
+
 	var err error
 	if sc.Net, err = fixture.Create(childEnv); err != nil {
 		return err
 	}
 
-	// If network is used, enable shorter per-node socket paths, because some e2e test datadir
-	// exceed maximum unix socket path length.
-	sc.Net.Config().UseShortGrpcSocketPaths = true
-
 	return nil
 }
 

go/oasis-test-runner/scenario/e2e/runtime/halt_restore.go

@@ -3,7 +3,6 @@ package runtime
 import (
 	"context"
 	"fmt"
-	"math"
 	"reflect"
 
 	beacon "github.com/oasisprotocol/oasis-core/go/beacon/api"
@@ -168,8 +167,7 @@ func (sc *haltRestoreImpl) Run(childEnv *env.Env) error { // nolint: gocyclo
 	// works with restored state.
 	sc.Logger.Info("starting the network again")
 
-	// Update halt epoch in the exported genesis so the network doesn't
-	// instantly halt.
+	// Ensure compute runtime in genesis is in expected state.
 	genesisFileProvider, err := genesis.NewFileProvider(files[0])
 	if err != nil {
 		sc.Logger.Error("failed getting genesis file provider",
@@ -185,15 +183,7 @@ func (sc *haltRestoreImpl) Run(childEnv *env.Env) error { // nolint: gocyclo
 		)
 		return err
 	}
-	genesisDoc.HaltEpoch = math.MaxUint64
-	if err = genesisDoc.WriteFileJSON(files[0]); err != nil {
-		sc.Logger.Error("failed to update genesis",
-			"err", err,
-		)
-		return err
-	}
 
-	// Ensure compute runtime in genesis is in expected state.
 	var rtList []*registry.Runtime
 	switch sc.suspendRuntime {
 	case true:
@@ -217,19 +207,20 @@ func (sc *haltRestoreImpl) Run(childEnv *env.Env) error { // nolint: gocyclo
 		return fmt.Errorf("runtime not in expected state")
 	}
 
+	// Disable halt epoch so the network doesn't instantly halt.
+	fixture.Network.HaltEpoch = 0
 	// Use the updated genesis file.
 	fixture.Network.GenesisFile = files[0]
 	// Make sure to not overwrite the entity.
 	fixture.Entities[1].Restore = true
+	// If network is used, enable shorter per-node socket paths, because some e2e test datadir
+	// exceed maximum unix socket path length.
+	fixture.Network.UseShortGrpcSocketPaths = true
 
 	if sc.Net, err = fixture.Create(childEnv); err != nil {
 		return err
 	}
 
-	// If network is used, enable shorter per-node socket paths, because some e2e test datadir exceed maximum unix
-	// socket path length.
-	sc.Net.Config().UseShortGrpcSocketPaths = true
-
 	newTestClient := sc.testClient.Clone().(*LongTermTestClient)
 	sc.runtimeImpl.testClient = newTestClient.WithMode(ModePart2).WithSeed("second_seed")
 

go/oasis-test-runner/scenario/e2e/runtime/halt_restore_nonmock.go

@@ -3,7 +3,6 @@ package runtime
 import (
 	"context"
 	"fmt"
-	"math"
 	"reflect"
 
 	genesis "github.com/oasisprotocol/oasis-core/go/genesis/file"
@@ -97,8 +96,7 @@ func (sc *haltRestoreNonMockImpl) Run(childEnv *env.Env) error { // nolint: gocy
 	// works with restored state.
 	sc.Logger.Info("starting the network again")
 
-	// Update halt epoch in the exported genesis so the network doesn't
-	// instantly halt.
+	// Update genesis file.
 	genesisFileProvider, err := genesis.NewFileProvider(files[0])
 	if err != nil {
 		sc.Logger.Error("failed getting genesis file provider",
@@ -114,7 +112,6 @@ func (sc *haltRestoreNonMockImpl) Run(childEnv *env.Env) error { // nolint: gocy
 		)
 		return err
 	}
-	genesisDoc.HaltEpoch = math.MaxUint64
 	genesisDoc.Beacon.Parameters.VRFParameters.Interval -= 5 // Reduce interval.
 	if err = genesisDoc.WriteFileJSON(files[0]); err != nil {
 		sc.Logger.Error("failed to update genesis",
@@ -123,19 +120,20 @@ func (sc *haltRestoreNonMockImpl) Run(childEnv *env.Env) error { // nolint: gocy
 		return err
 	}
 
+	// Disable halt epoch so the network doesn't instantly halt.
+	fixture.Network.HaltEpoch = 0
 	// Use the updated genesis file.
 	fixture.Network.GenesisFile = files[0]
 	// Make sure to not overwrite the entity.
 	fixture.Entities[1].Restore = true
+	// If network is used, enable shorter per-node socket paths, because some e2e test datadir
+	// exceed maximum unix socket path length.
+	fixture.Network.UseShortGrpcSocketPaths = true
 
 	if sc.Net, err = fixture.Create(childEnv); err != nil {
 		return err
 	}
 
-	// If network is used, enable shorter per-node socket paths, because some e2e test datadir
-	// exceed maximum unix socket path length.
-	sc.Net.Config().UseShortGrpcSocketPaths = true
-
 	newTestClient := sc.testClient.Clone().(*LongTermTestClient)
 	sc.runtimeImpl.testClient = newTestClient.WithMode(ModePart2).WithSeed("second_seed")