feat(cli): add shard CLI command (#1785)

* stub out shard CLI command * prune blocks before and after desired range * update rollback to use patched cometbft * temp override for local patched versions * handle pruning cometbft & block store state * include docs & support -1 for "latest" * update changelog * add --only-app-state flag to match cosmos-sdk prune cmd * give -1 magic number a name & reuse home from ctx * refactor to only open state.db & blockstore.db once * write rollback progress to one line * prevent attempting rollback of future blocks * make shard inclusive of endblock * use tagged cosmo-sdk & cometbft versions
2024-11-10 10:05:18 +00:00 · 2024-01-05 11:25:53 -08:00 · 2024-01-05 11:25:53 -08:00 · 58621577ae
commit 58621577ae
parent f00727fe85
6 changed files with 245 additions and 8 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -36,6 +36,9 @@ Ref: https://keepachangelog.com/en/1.0.0/

 ## [Unreleased]

+### Features
+
+- (cli) [#1785] Add `shard` CLI command to support creating partitions of data for standalone nodes

 ## [v0.25.0]

@ -317,6 +320,7 @@ the [changelog](https://github.com/cosmos/cosmos-sdk/blob/v0.38.4/CHANGELOG.md).
 - [#257](https://github.com/Kava-Labs/kava/pulls/257) Include scripts to run
  large-scale simulations remotely using aws-batch

+[#1785]: https://github.com/Kava-Labs/kava/pull/1785
 [#1784]: https://github.com/Kava-Labs/kava/pull/1784
 [#1776]: https://github.com/Kava-Labs/kava/pull/1776
 [#1770]: https://github.com/Kava-Labs/kava/pull/1770
--- a/cmd/kava/cmd/root.go
+++ b/cmd/kava/cmd/root.go
@ -123,5 +123,6 @@ func addSubCmds(rootCmd *cobra.Command, encodingConfig params.EncodingConfig, de
 		newQueryCmd(),
 		newTxCmd(),
 		keyCommands(app.DefaultNodeHome),
+		newShardCmd(opts),
 	)
 }
--- a/cmd/kava/cmd/shard.go
+++ b/cmd/kava/cmd/shard.go
@ -0,0 +1,232 @@
+package cmd
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/spf13/cobra"
+
+	dbm "github.com/cometbft/cometbft-db"
+
+	"github.com/cosmos/cosmos-sdk/client"
+	"github.com/cosmos/cosmos-sdk/client/flags"
+	pruningtypes "github.com/cosmos/cosmos-sdk/pruning/types"
+	"github.com/cosmos/cosmos-sdk/server"
+	"github.com/cosmos/cosmos-sdk/store/rootmulti"
+
+	tmconfig "github.com/tendermint/tendermint/config"
+	"github.com/tendermint/tendermint/node"
+	tmstate "github.com/tendermint/tendermint/state"
+	"github.com/tendermint/tendermint/store"
+
+	ethermintserver "github.com/evmos/ethermint/server"
+)
+
+const (
+	flagShardStartBlock   = "start"
+	flagShardEndBlock     = "end"
+	flagShardOnlyAppState = "only-app-state"
+	// TODO: --preserve flag for creating & operating on a copy?
+
+	// allow using -1 to mean "latest" (perform no rollbacks)
+	shardEndBlockLatest = -1
+)
+
+func newShardCmd(opts ethermintserver.StartOptions) *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "shard --home <path-to-home-dir> --start <start-block> --end <end-block> [--only-app-state]",
+		Short: "Strip all blocks from the database outside of a given range",
+		Long: `shard opens a local kava home directory's databases and removes all blocks outside a range defined by --start and --end. The range is inclusive of the end block.
+
+It works by first rolling back the latest state to the block before the end block, and then by pruning all state before the start block.
+
+Setting the end block to -1 signals to keep the latest block (no rollbacks).
+
+The --only-app-state flag can be used to skip the pruning of the blockstore and cometbft state. This matches the functionality of the cosmos-sdk's "prune" command. Note that rolled back blocks will still affect all stores.
+
+WARNING: this is a destructive action.`,
+		Example: `Create a 1M block data shard (keeps blocks kava 1,000,000 to 2,000,000)
+$ kava shard --home path/to/.kava --start 1000000 --end 2000000
+
+Prune all blocks up to 5,000,000:
+$ kava shard --home path/to/.kava --start 5000000 --end -1
+
+Prune first 1M blocks _without_ affecting blockstore or cometBFT state:
+$ kava shard --home path/to/.kava --start 1000000 --end -1 --only-app-state`,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			// read & validate flags
+			startBlock, err := cmd.Flags().GetInt64(flagShardStartBlock)
+			if err != nil {
+				return err
+			}
+			endBlock, err := cmd.Flags().GetInt64(flagShardEndBlock)
+			if err != nil {
+				return err
+			}
+			if (endBlock == 0 || endBlock < startBlock) && endBlock != shardEndBlockLatest {
+				return fmt.Errorf("end block (%d) must be greater than start block (%d)", endBlock, startBlock)
+			}
+			onlyAppState, err := cmd.Flags().GetBool(flagShardOnlyAppState)
+			if err != nil {
+				return err
+			}
+
+			clientCtx := client.GetClientContextFromCmd(cmd)
+
+			ctx := server.GetServerContextFromCmd(cmd)
+			ctx.Config.SetRoot(clientCtx.HomeDir)
+
+			//////////////////////////////
+			// Rollback state to endBlock
+			//////////////////////////////
+
+			// connect to database
+			db, err := opts.DBOpener(ctx.Viper, clientCtx.HomeDir, server.GetAppDBBackend(ctx.Viper))
+			if err != nil {
+				return err
+			}
+
+			// close db connection when done
+			defer func() {
+				if err := db.Close(); err != nil {
+					ctx.Logger.Error("error closing db", "error", err.Error())
+				}
+			}()
+
+			// get the multistore
+			app := opts.AppCreator(ctx.Logger, db, nil, ctx.Viper)
+			cms := app.CommitMultiStore()
+			multistore, ok := cms.(*rootmulti.Store)
+			if !ok {
+				return fmt.Errorf("only sharding of rootmulti.Store type is supported")
+			}
+
+			// handle desired endblock being latest
+			latest := multistore.LatestVersion()
+			fmt.Printf("latest height: %d\n", latest)
+			if endBlock == shardEndBlockLatest {
+				endBlock = latest
+			}
+			shardSize := endBlock - startBlock + 1
+
+			// error if requesting block range the database does not have
+			if endBlock > latest {
+				return fmt.Errorf("data does not contain end block (%d): latest version is %d", endBlock, latest)
+			}
+
+			fmt.Printf("pruning data in %s down to heights %d - %d (%d blocks)\n", clientCtx.HomeDir, startBlock, endBlock, shardSize)
+
+			// set pruning options to prevent no-ops from `PruneStores`
+			multistore.SetPruning(pruningtypes.PruningOptions{KeepRecent: uint64(shardSize), Interval: 0})
+
+			// rollback application state
+			if err = multistore.RollbackToVersion(endBlock); err != nil {
+				return fmt.Errorf("failed to rollback application state: %s", err)
+			}
+
+			// open block store & cometbft state
+			blockStore, stateStore, err := openCometBftDbs(ctx.Config)
+			if err != nil {
+				return fmt.Errorf("failed to open cometbft dbs: %s", err)
+			}
+
+			// prep for outputting progress repeatedly to same line
+			needsRollback := endBlock < latest
+			progress := "rolling back blockstore & cometbft state to height %d"
+			numChars := len(fmt.Sprintf(progress, latest))
+			clearLine := fmt.Sprintf("\r%s\r", strings.Repeat(" ", numChars))
+			printRollbackProgress := func(h int64) {
+				fmt.Print(clearLine)
+				fmt.Printf(progress, h)
+			}
+
+			// rollback tendermint db
+			height := latest
+			for height > endBlock {
+				printRollbackProgress(height - 1)
+				height, _, err = tmstate.Rollback(blockStore, stateStore, true)
+				if err != nil {
+					return fmt.Errorf("failed to rollback tendermint state: %w", err)
+				}
+			}
+
+			if needsRollback {
+				fmt.Println()
+			} else {
+				fmt.Printf("latest store height is already %d\n", latest)
+			}
+
+			//////////////////////////////
+			// Prune blocks to startBlock
+			//////////////////////////////
+
+			// enumerate all heights to prune
+			pruneHeights := make([]int64, 0, latest-shardSize)
+			for i := int64(1); i < startBlock; i++ {
+				pruneHeights = append(pruneHeights, i)
+			}
+
+			if len(pruneHeights) > 0 {
+				// prune application state
+				fmt.Printf("pruning application state to height %d\n", startBlock)
+				if err := multistore.PruneStores(true, pruneHeights); err != nil {
+					return fmt.Errorf("failed to prune application state: %s", err)
+				}
+			}
+
+			// get starting block of block store
+			baseBlock := blockStore.Base()
+
+			// only prune if data exists, otherwise blockStore.PruneBlocks will panic
+			if !onlyAppState && baseBlock < startBlock {
+				// prune block store
+				fmt.Printf("pruning block store from %d - %d\n", baseBlock, startBlock)
+				if _, err := blockStore.PruneBlocks(startBlock); err != nil {
+					return fmt.Errorf("failed to prune block store (retainHeight=%d): %s", startBlock, err)
+				}
+
+				// prune cometbft state
+				fmt.Printf("pruning cometbft state from %d - %d\n", baseBlock, startBlock)
+				if err := stateStore.PruneStates(baseBlock, startBlock); err != nil {
+					return fmt.Errorf("failed to prune cometbft state store (%d - %d): %s", baseBlock, startBlock, err)
+				}
+			} else {
+				fmt.Printf("blockstore and cometbft state begins at block %d\n", baseBlock)
+			}
+
+			// TODO: db compaction
+
+			return nil
+		},
+	}
+
+	cmd.Flags().String(flags.FlagHome, opts.DefaultNodeHome, "The application home directory")
+	cmd.Flags().Int64(flagShardStartBlock, 1, "Start block of data shard (inclusive)")
+	cmd.Flags().Int64(flagShardEndBlock, 0, "End block of data shard (inclusive)")
+	cmd.Flags().Bool(flagShardOnlyAppState, false, "Skip pruning of blockstore & cometbft state")
+
+	return cmd
+}
+
+// inspired by https://github.com/Kava-Labs/cometbft/blob/277b0853db3f67865a55aa1c54f59790b5f591be/node/node.go#L234
+func openCometBftDbs(config *tmconfig.Config) (blockStore *store.BlockStore, stateStore tmstate.Store, err error) {
+	dbProvider := node.DefaultDBProvider
+
+	var blockStoreDB dbm.DB
+	blockStoreDB, err = dbProvider(&node.DBContext{ID: "blockstore", Config: config})
+	if err != nil {
+		return
+	}
+	blockStore = store.NewBlockStore(blockStoreDB)
+
+	stateDB, err := dbProvider(&node.DBContext{ID: "state", Config: config})
+	if err != nil {
+		return
+	}
+
+	stateStore = tmstate.NewStore(stateDB, tmstate.StoreOptions{
+		DiscardABCIResponses: config.Storage.DiscardABCIResponses,
+	})
+
+	return
+}
--- a/go.mod
+++ b/go.mod
@ -6,6 +6,7 @@ require (
 	cosmossdk.io/errors v1.0.0-beta.7
 	cosmossdk.io/math v1.0.0-beta.6.0.20230216172121-959ce49135e4
 	github.com/cenkalti/backoff/v4 v4.1.3
+	github.com/cometbft/cometbft-db v0.7.0
 	github.com/cosmos/cosmos-proto v1.0.0-beta.3
 	github.com/cosmos/cosmos-sdk v0.46.11
 	github.com/cosmos/go-bip39 v1.0.0
@ -62,7 +63,6 @@ require (
 	github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e // indirect
 	github.com/cockroachdb/apd/v2 v2.0.2 // indirect
 	github.com/coinbase/rosetta-sdk-go v0.7.9 // indirect
-	github.com/cometbft/cometbft-db v0.7.0 // indirect
 	github.com/confio/ics23/go v0.9.0 // indirect
 	github.com/cosmos/btcutil v1.0.5 // indirect
 	github.com/cosmos/gogoproto v1.4.6 // indirect
@ -205,7 +205,7 @@ replace (
 	// Use rocksdb 7.9.2
 	github.com/cometbft/cometbft-db => github.com/kava-labs/cometbft-db v0.7.0-rocksdb-v7.9.2-kava.1
 	// Use cosmos-sdk fork with backported fix for unsafe-reset-all, staking transfer events, and custom tally handler support
-	github.com/cosmos/cosmos-sdk => github.com/kava-labs/cosmos-sdk v0.46.11-kava.3
+	github.com/cosmos/cosmos-sdk => github.com/kava-labs/cosmos-sdk v0.46.11-kava.4
 	// See https://github.com/cosmos/cosmos-sdk/pull/13093
 	github.com/dgrijalva/jwt-go => github.com/golang-jwt/jwt/v4 v4.4.2
 	// Use ethermint fork that respects min-gas-price with NoBaseFee true and london enabled, and includes eip712 support
@ -217,7 +217,7 @@ replace (
 	// Downgraded to avoid bugs in following commits which causes "version does not exist" errors
 	github.com/syndtr/goleveldb => github.com/syndtr/goleveldb v1.0.1-0.20210819022825-2ae1ddf74ef7
 	// Use cometbft fork of tendermint
-	github.com/tendermint/tendermint => github.com/kava-labs/cometbft v0.34.27-kava.0
+	github.com/tendermint/tendermint => github.com/kava-labs/cometbft v0.34.27-kava.1
 	// Indirect dependencies still use tendermint/tm-db
 	github.com/tendermint/tm-db => github.com/kava-labs/tm-db v0.6.7-kava.4
 )
--- a/go.sum
+++ b/go.sum
@ -802,12 +802,12 @@ github.com/julienschmidt/httprouter v1.3.0/go.mod h1:JR6WtHb+2LUe8TCKY3cZOxFyyO8
 github.com/jung-kurt/gofpdf v1.0.3-0.20190309125859-24315acbbda5/go.mod h1:7Id9E/uU8ce6rXgefFLlgrJj/GYY22cpxn+r32jIOes=
 github.com/jwilder/encoding v0.0.0-20170811194829-b4e1701a28ef/go.mod h1:Ct9fl0F6iIOGgxJ5npU/IUOhOhqlVrGjyIZc8/MagT0=
 github.com/karalabe/usb v0.0.2/go.mod h1:Od972xHfMJowv7NGVDiWVxk2zxnWgjLlJzE+F4F7AGU=
-github.com/kava-labs/cometbft v0.34.27-kava.0 h1:FUEGRkF3xtrJH+h9A5G4eA2skf7QaNoOCPaoVqHkh8k=
-github.com/kava-labs/cometbft v0.34.27-kava.0/go.mod h1:BcCbhKv7ieM0KEddnYXvQZR+pZykTKReJJYf7YC7qhw=
+github.com/kava-labs/cometbft v0.34.27-kava.1 h1:JkTspNCrz9matgrr7nsWgEkgNzDz5YwZhR5jZyxVt/0=
+github.com/kava-labs/cometbft v0.34.27-kava.1/go.mod h1:BcCbhKv7ieM0KEddnYXvQZR+pZykTKReJJYf7YC7qhw=
 github.com/kava-labs/cometbft-db v0.7.0-rocksdb-v7.9.2-kava.1 h1:EZnZAkZ+dqK+1OM4AK+e6wYH8a5xuyg4yFTR4Ez3AXk=
 github.com/kava-labs/cometbft-db v0.7.0-rocksdb-v7.9.2-kava.1/go.mod h1:mI/4J4IxRzPrXvMiwefrt0fucGwaQ5Hm9IKS7HnoJeI=
-github.com/kava-labs/cosmos-sdk v0.46.11-kava.3 h1:TOhyyW/xHso/9uIOgYdsrOWDIhXi6foORWZxVRe/wS0=
-github.com/kava-labs/cosmos-sdk v0.46.11-kava.3/go.mod h1:bSUUbmVwWkv1ZNVTWrQHa/i+73xIUvYYPsCvl5doiCs=
+github.com/kava-labs/cosmos-sdk v0.46.11-kava.4 h1:lCcClhrGCwHunwbSG/zmY8o/g8rE6GYH1x2OOPMkk0g=
+github.com/kava-labs/cosmos-sdk v0.46.11-kava.4/go.mod h1:VVdf1w8CSIhh4FnlxOGVVAlvUnvXrS8VV6DwUBCuAOs=
 github.com/kava-labs/ethermint v0.21.0-kava-v24-1 h1:XgEeXbLtGpOxSt7yFVznrjics5mRsUpMMfqlrhhH2pc=
 github.com/kava-labs/ethermint v0.21.0-kava-v24-1/go.mod h1:rdm6AinxZ4dzPEv/cjH+/AGyTbKufJ3RE7M2MDyklH0=
 github.com/kava-labs/tm-db v0.6.7-kava.4 h1:M2RibOKmbi+k2OhAFry8z9+RJF0CYuDETB7/PrSdoro=
--- a/tests/e2e/kvtool
+++ b/tests/e2e/kvtool
@ -1 +1 @@
-Subproject commit e1085562d203fd81c5dd8576170b29715b2de9ef
+Subproject commit 075328bbb54010f2dcddd29fe0a0a0cfccaf41e2