Merge pull request #6341 from oasisprotocol/martin/feature/cli/prune-offline

go/oasis-node/cmd/storage: Add command for offline pruning

Author: martintomazic

.changelog/6321.feature.md

@@ -0,0 +1,7 @@
+go/oasis-node: Add new command for offline pruning of consensus databases
+
+A new experimental command `oasis-node storage prune-experimental`
+was added, enabling offline pruning as specified in the configuration.
+
+Operators are encouraged to run this command whenever they change pruning
+configuration, to ensure node is healthy when it starts.

docs/oasis-node/cli.md

@@ -336,7 +336,7 @@ oasis1qqncl383h8458mr9cytatygctzwsx02n4c5f8ed7
 
 ### compact-experimental
 
-Run
+Run (when the node is not running):
 
 ```sh
 oasis-node storage compact-experimental --config /path/to/config/file
@@ -364,3 +364,26 @@ may stay constant or not be reclaimed for a very long time.
 
 This command gives operators manual control to release disk space during
 maintenance periods.
+
+### prune-experimental
+
+Run (when the node is not running):
+
+```sh
+oasis-node storage prune-experimental --config /path/to/config/file
+```
+
+to trigger manual pruning of consensus database instances:
+
+```sh
+{"caller":"storage.go:433","level":"info","module":"cmd/storage", \
+"msg":"Starting consensus databases pruning. This may take a while...", \
+"ts":"2025-10-23T11:02:11.129822974Z"}
+```
+
+Operators should run this whenever they change pruning configuration, e.g. when
+enabling it for the first time, or later changing it to retain less data. This
+way they guarantee the node is healthy when it starts.
+
+Following successful pruning, to release disk space, they are encouraged to run
+[the compaction command](#compact-experimental).

go/consensus/cometbft/abci/prune.go

@@ -200,6 +200,8 @@ func (p *genericPruner) canPrune(v int64) error {
 	return nil
 }
 
+// Warning: When registering new handler DO NOT forget to update the logic for
+// "oasis-node storage prune" command as well.
 func (p *genericPruner) RegisterHandler(handler consensus.StatePruneHandler) {
 	p.Lock()
 	defer p.Unlock()

go/consensus/cometbft/db/init.go

@@ -2,23 +2,61 @@
 package db
 
 import (
+	"fmt"
+
 	dbm "github.com/cometbft/cometbft-db"
+	cmtconfig "github.com/cometbft/cometbft/config"
 	"github.com/cometbft/cometbft/node"
+	cmtnode "github.com/cometbft/cometbft/node"
+	"github.com/cometbft/cometbft/state"
 
 	"github.com/oasisprotocol/oasis-core/go/consensus/cometbft/db/badger"
 )
 
-// GetBackendName returns the currently configured CometBFT database backend.
-func GetBackendName() string {
+// BackendName returns the currently configured CometBFT database backend.
+func BackendName() string {
 	return badger.BackendName
 }
 
-// GetProvider returns the currently configured CometBFT DBProvider.
-func GetProvider() (node.DBProvider, error) {
+// Provider returns the currently configured CometBFT DBProvider.
+func Provider() (node.DBProvider, error) {
 	return badger.DBProvider, nil
 }
 
 // New constructs a new CometBFT DB with the configured backend.
 func New(fn string, noSuffix bool) (dbm.DB, error) {
 	return badger.New(fn, noSuffix)
 }
+
+// OpenBlockstoreDB opens a CometBFT managed blockstore DB.
+//
+// This function is a hack as CometBFT does not expose a way to access the underlying databases.
+func OpenBlockstoreDB(provider cmtnode.DBProvider, cfg *cmtconfig.Config) (dbm.DB, error) {
+	// NOTE: DBContext uses a full CometBFT config but the only thing that is actually used
+	// is the data dir field.
+	db, err := provider(&cmtnode.DBContext{ID: "blockstore", Config: cfg})
+	if err != nil {
+		return nil, fmt.Errorf("failed to open blockstore: %w", err)
+	}
+
+	return db, nil
+}
+
+// OpenStateDB opens a CometBFT managed state DB.
+//
+// This function is a hack as CometBFT does not expose a way to access the underlying databases.
+func OpenStateDB(provider cmtnode.DBProvider, cfg *cmtconfig.Config) (dbm.DB, error) {
+	// NOTE: DBContext uses a full CometBFT config but the only thing that is actually used
+	// is the data dir field.
+	db, err := provider(&cmtnode.DBContext{ID: "state", Config: cfg})
+	if err != nil {
+		return nil, fmt.Errorf("failed to open state db: %w", err)
+	}
+
+	return db, nil
+}
+
+// OpenStateStore constructs a new state store using default options.
+func OpenStateStore(stateDB dbm.DB) state.Store {
+	return state.NewStore(stateDB, state.StoreOptions{})
+}

go/consensus/cometbft/full/archive.go

@@ -7,11 +7,9 @@ import (
 	"sync"
 	"time"
 
-	dbm "github.com/cometbft/cometbft-db"
 	abcicli "github.com/cometbft/cometbft/abci/client"
 	cmtconfig "github.com/cometbft/cometbft/config"
 	cmtsync "github.com/cometbft/cometbft/libs/sync"
-	cmtnode "github.com/cometbft/cometbft/node"
 	cmtproxy "github.com/cometbft/cometbft/proxy"
 	cmtcore "github.com/cometbft/cometbft/rpc/core"
 	"github.com/cometbft/cometbft/state"
@@ -194,31 +192,27 @@ func NewArchive(ctx context.Context, cfg ArchiveConfig) (consensusAPI.Service, e
 	logger := tmcommon.NewLogAdapter(!config.GlobalConfig.Consensus.LogDebug)
 	srv.abciClient = abcicli.NewLocalClient(new(cmtsync.Mutex), srv.mux.Mux())
 
-	dbProvider, err := db.GetProvider()
-	if err != nil {
-		return nil, err
-	}
 	cmtConfig := cmtconfig.DefaultConfig()
 	_ = viper.Unmarshal(&cmtConfig)
 	cmtConfig.SetRoot(filepath.Join(srv.dataDir, tmcommon.StateDir))
 
-	// NOTE: DBContext uses a full CometBFT config but the only thing that is actually used
-	// is the data dir field.
-	srv.blockStoreDB, err = dbProvider(&cmtnode.DBContext{ID: "blockstore", Config: cmtConfig})
+	dbProvider, err := db.Provider()
+	if err != nil {
+		return nil, fmt.Errorf("failed to obtain db provider: %w", err)
+	}
+
+	srv.blockStoreDB, err = db.OpenBlockstoreDB(dbProvider, cmtConfig)
 	if err != nil {
 		return nil, err
 	}
 	srv.blockStoreDB = db.WithCloser(srv.blockStoreDB, srv.dbCloser)
 
-	// NOTE: DBContext uses a full CometBFT config but the only thing that is actually used
-	// is the data dir field.
-	var stateDB dbm.DB
-	stateDB, err = dbProvider(&cmtnode.DBContext{ID: "state", Config: cmtConfig})
+	stateDB, err := db.OpenStateDB(dbProvider, cmtConfig)
 	if err != nil {
 		return nil, err
 	}
 	stateDB = db.WithCloser(stateDB, srv.dbCloser)
-	srv.stateStore = state.NewStore(stateDB, state.StoreOptions{})
+	srv.stateStore = db.OpenStateStore(stateDB)
 
 	srv.eb = cmttypes.NewEventBus()
 	// Setup minimal CometBFT environment needed to support consensus queries.

go/consensus/cometbft/full/full.go

@@ -668,7 +668,7 @@ func (t *fullService) lazyInit() error { // nolint: gocyclo
 		return t.genesisDoc, nil
 	}
 
-	dbProvider, err := db.GetProvider()
+	dbProvider, err := db.Provider()
 	if err != nil {
 		t.Logger.Error("failed to obtain database provider",
 			"err", err,

go/oasis-node/cmd/common/common.go

@@ -2,8 +2,10 @@
 package common
 
 import (
+	"errors"
 	"fmt"
 	"io"
+	"io/fs"
 	"os"
 	"path/filepath"
 	"strings"
@@ -71,6 +73,20 @@ func InternalSocketPath() string {
 	return filepath.Join(DataDir(), InternalSocketName)
 }
 
+// IsNodeRunning returns true when the node is running.
+func IsNodeRunning() (bool, error) {
+	path := InternalSocketPath()
+
+	if _, err := os.Stat(path); err != nil {
+		if errors.Is(err, fs.ErrNotExist) {
+			return false, nil
+		}
+		return false, fmt.Errorf("stat %s: %w", path, err)
+	}
+
+	return true, nil
+}
+
 // IsNodeCmd returns true iff the current command is the ekiden node.
 func IsNodeCmd() bool {
 	return isNodeCmd