featurebase/ctl/backup.go

// Copyright 2022 Molecula Corp. (DBA FeatureBase).
// SPDX-License-Identifier: Apache-2.0
package ctl

import (
	"context"
	"crypto/tls"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"io/fs"
	"os"
	"path/filepath"
	"time"

	pilosa "github.com/featurebasedb/featurebase/v3"
	"github.com/featurebasedb/featurebase/v3/authn"
	"github.com/featurebasedb/featurebase/v3/disco"
	"github.com/featurebasedb/featurebase/v3/encoding/proto"
	"github.com/featurebasedb/featurebase/v3/logger"
	"github.com/featurebasedb/featurebase/v3/server"
	"github.com/ricochet2200/go-disk-usage/du"
	"golang.org/x/sync/errgroup"
)

// TODO(rdp): add refresh token to this as well

// BackupCommand represents a command for backing up a FeatureBase node.
type BackupCommand struct { // nolint: maligned
	tlsConfig *tls.Config

	// Destination host and port.
	Host string `json:"host"`

	// Optional Index filter
	Index string `json:"index"`

	// Path to write the backup to.
	OutputDir string

	// If true, skips file sync.
	NoSync bool

	// Number of concurrent backup goroutines running at a time.
	Concurrency int

	// Amount of time after first failed request to continue retrying.
	RetryPeriod time.Duration `json:"retry-period"`

	// Response Header Timeout for HTTP Requests
	HeaderTimeoutStr string
	HeaderTimeout    time.Duration `json:"header-timeout"`

	// Host:port on which to listen for pprof.
	Pprof string `json:"pprof"`

	// Reusable client.
	client *pilosa.InternalClient

	// Standard input/output
	logDest logger.Logger

	TLS server.TLSConfig

	AuthToken        string
	IgnoreSpaceCheck bool
}

// Logger returns the command's associated Logger to maintain CommandWithTLSSupport interface compatibility
func (cmd *BackupCommand) Logger() logger.Logger {
	return cmd.logDest
}

// NewBackupCommand returns a new instance of BackupCommand.
func NewBackupCommand(logdest logger.Logger) *BackupCommand {
	return &BackupCommand{
		logDest:       logdest,
		Concurrency:   1,
		RetryPeriod:   time.Minute,
		HeaderTimeout: time.Second * 3,
		Pprof:         "localhost:0",
	}
}

// Run executes the main program execution.
func (cmd *BackupCommand) Run(ctx context.Context) (err error) {
	logger := cmd.Logger()
	close, err := startProfilingServer(cmd.Pprof, logger)
	if err != nil {
		return fmt.Errorf("starting profiling server: %w", err)
	}
	defer close()

	// Validate arguments.
	if cmd.OutputDir == "" {
		return fmt.Errorf("%w: -o flag required", UsageError)
	} else if cmd.Concurrency <= 0 {
		return fmt.Errorf("%w: concurrency must be at least one", UsageError)
	}
	if cmd.HeaderTimeoutStr != "" {
		if dur, err := time.ParseDuration(cmd.HeaderTimeoutStr); err != nil {
			return fmt.Errorf("%w: could not parse '%s' as a duration: %v", UsageError, cmd.HeaderTimeoutStr, err)
		} else {
			cmd.HeaderTimeout = dur
		}
	}

	// Parse TLS configuration for node-specific clients.
	tls := cmd.TLSConfiguration()
	if cmd.tlsConfig, err = server.GetTLSConfig(&tls, cmd.Logger()); err != nil {
		return fmt.Errorf("parsing tls config: %w", err)
	}

	// Create a client to the server.
	client, err := commandClient(cmd, pilosa.WithClientRetryPeriod(cmd.RetryPeriod), pilosa.ClientResponseHeaderTimeoutOption(cmd.HeaderTimeout))
	if err != nil {
		return fmt.Errorf("creating client: %w", err)
	}
	cmd.client = client

	if cmd.AuthToken != "" {
		ctx = authn.WithAccessToken(ctx, "Bearer "+cmd.AuthToken)
	}

	// Determine the field type in order to correctly handle the input data.
	indexes, err := cmd.client.Schema(ctx)
	if err != nil {
		return fmt.Errorf("getting schema: %w", err)
	}
	if cmd.Index != "" {
		for _, idx := range indexes {
			if idx.Name == cmd.Index {
				indexes = make([]*pilosa.IndexInfo, 0)
				indexes = append(indexes, idx)
				break
			}
		}
		if len(indexes) <= 0 {
			return fmt.Errorf("index not found to back up")
		}
	}

	schema := &pilosa.Schema{Indexes: indexes}

	// Ensure output directory doesn't exist; then create output directory.
	if _, err := os.Stat(cmd.OutputDir); !errors.Is(err, fs.ErrNotExist) {
		return fmt.Errorf("output directory already exists")
	} else if err := os.MkdirAll(cmd.OutputDir, 0o750); err != nil {
		return err
	}

	if !cmd.IgnoreSpaceCheck {
		// Ensure there is enough free space
		if err := cmd.checkFreeSpace(ctx); err != nil {
			return fmt.Errorf("not enough disk space available: %w", err)
		}
	}

	// Backup schema.
	if err := cmd.backupSchema(ctx, schema); err != nil {
		return fmt.Errorf("cannot back up schema: %w", err)
	} else if err := cmd.backupIDAllocData(ctx); err != nil {
		return fmt.Errorf("cannot back up id alloc data: %w", err)
	}

	// Backup data for each index.
	for _, ii := range schema.Indexes {
		if err := cmd.backupIndexData(ctx, ii); err != nil {
			return err
		}
	}
	// Backup translation data. This has to happen separately, because
	// otherwise a field which uses foreign key translation can reasonably
	// contain values which got created for the foreign index after we
	// backed up that index.
	for _, ii := range schema.Indexes {
		if err := cmd.backupIndexTranslation(ctx, ii); err != nil {
			return err
		}
	}

	// Wait for the OS to persist all directories.
	err = cmd.syncDirectories(ctx)
	if err != nil {
		return fmt.Errorf("syncing directories: %w", err)
	}

	return nil
}

// backupSchema writes the schema to the archive.
func (cmd *BackupCommand) backupSchema(ctx context.Context, schema *pilosa.Schema) error {
	logger := cmd.Logger()
	logger.Printf("backing up schema")

	buf, err := json.MarshalIndent(schema, "", "\t")
	if err != nil {
		return fmt.Errorf("marshaling schema: %w", err)
	}

	if err := os.WriteFile(filepath.Join(cmd.OutputDir, "schema"), buf, 0o600); err != nil {
		return fmt.Errorf("writing schema: %w", err)
	}

	return nil
}

func (cmd *BackupCommand) backupIDAllocData(ctx context.Context) error {
	logger := cmd.Logger()
	logger.Printf("backing up id alloc data")

	rc, err := cmd.client.IDAllocDataReader(ctx)
	if err != nil {
		return fmt.Errorf("fetching id alloc data reader: %w", err)
	}
	defer rc.Close()

	f, err := os.Create(filepath.Join(cmd.OutputDir, "idalloc"))
	if err != nil {
		return err
	}
	defer f.Close()

	if _, err := io.Copy(f, rc); err != nil {
		return err
	} else if err := cmd.syncFile(f); err != nil {
		return err
	}
	return f.Close()
}

// backupIndexTranslation backs up both field and index-wide key translation for
// the given index. it has to run after the index's data has been backed up,
// but also after the data for any index which might have a foreign-key
// relation to this index has been backed up.
func (cmd *BackupCommand) backupIndexTranslation(ctx context.Context, ii *pilosa.IndexInfo) error {
	logger := cmd.Logger()
	logger.Printf("backing up index translation: %q", ii.Name)
	if ii.Options.Keys {
		if err := cmd.backupIndexTranslateData(ctx, ii.Name); err != nil {
			return err
		}
	}

	// Back up field translation data.
	for _, fi := range ii.Fields {
		if !fi.Options.Keys {
			continue
		}
		if err := cmd.backupFieldTranslateData(ctx, ii.Name, fi.Name); err != nil {
			return fmt.Errorf("cannot backup field translation data for field %q on index %q: %w", fi.Name, ii.Name, err)
		}
	}

	return nil
}

// backupIndexData backs up all shard data for a given index.
func (cmd *BackupCommand) backupIndexData(ctx context.Context, ii *pilosa.IndexInfo) error {
	logger := cmd.Logger()
	logger.Printf("backing up index data: %q", ii.Name)
	shards, err := cmd.client.AvailableShards(ctx, ii.Name)
	if err != nil {
		return fmt.Errorf("cannot find available shards for index %q: %w", ii.Name, err)
	}

	// Back up all bitmap data for the index.
	ch := make(chan uint64, len(shards))
	for _, shard := range shards {
		ch <- shard
	}
	close(ch)

	g, ctx := errgroup.WithContext(ctx)
	for i := 0; i < cmd.Concurrency; i++ {
		g.Go(func() error {
			for {
				select {
				case <-ctx.Done():
					return ctx.Err()
				case shard, ok := <-ch:
					if !ok {
						return nil
					} else if err := cmd.backupShard(ctx, ii.Name, shard); err != nil {
						return fmt.Errorf("cannot backup shard %d on index %q: %w", shard, ii.Name, err)
					}
				}
			}
		})
	}
	return g.Wait()
}

// checkFreeSpace checks if there is enough space in the output directory to backup data
func (cmd *BackupCommand) checkFreeSpace(ctx context.Context) (err error) {
	freeSpace := du.NewDiskUsage(cmd.OutputDir).Free()
	var usage pilosa.DiskUsage
	if cmd.Index == "" {
		usage, err = cmd.client.GetDiskUsage(ctx)
	} else {
		usage, err = cmd.client.GetIndexUsage(ctx, cmd.Index)
	}
	if err != nil {
		return fmt.Errorf("getting size of data to be backed up: %s", err)
	}

	if freeSpace < uint64(usage.Usage) {
		return fmt.Errorf("not enough disk space available, free: %v, index usage: %v", freeSpace, usage.Usage)
	}
	return nil
}

// backupShard backs up a single shard from a single index.
func (cmd *BackupCommand) backupShard(ctx context.Context, indexName string, shard uint64) (err error) {
	nodes, err := cmd.client.FragmentNodes(ctx, indexName, shard)
	if err != nil {
		return fmt.Errorf("cannot determine fragment nodes: %w", err)
	} else if len(nodes) == 0 {
		return fmt.Errorf("no nodes available")
	}

	for _, node := range nodes {
		if e := cmd.backupShardNode(ctx, indexName, shard, node); e == nil {
			break
		} else if err == nil {
			err = e // save first error, try next node
		}
	}

	for _, node := range nodes {
		if e := cmd.backupShardDataframe(ctx, indexName, shard, node); e == nil {
			return nil // backup ok, exit
		} else if err == nil {
			err = e // save first error, try next node
		}
	}
	return err
}

// backupShardNode backs up a single shard from a single index on a specific node.
func (cmd *BackupCommand) backupShardNode(ctx context.Context, indexName string, shard uint64, node *disco.Node) error {
	logger := cmd.Logger()
	logger.Printf("backing up shard: index=%q id=%d", indexName, shard)

	client := pilosa.NewInternalClientFromURI(&node.URI,
		pilosa.GetHTTPClient(cmd.tlsConfig, pilosa.ClientResponseHeaderTimeoutOption(cmd.HeaderTimeout)),
		pilosa.WithClientRetryPeriod(cmd.RetryPeriod),
		pilosa.WithSerializer(proto.Serializer{}))
	rc, err := client.ShardReader(ctx, indexName, shard)
	if err != nil {
		return fmt.Errorf("fetching shard reader: %w", err)
	}
	defer rc.Close()

	filename := filepath.Join(cmd.OutputDir, "indexes", indexName, "shards", fmt.Sprintf("%04d", shard))
	if err := os.MkdirAll(filepath.Dir(filename), 0o750); err != nil {
		return err
	}

	f, err := os.Create(filename)
	if err != nil {
		return err
	}
	defer f.Close()

	if _, err := io.Copy(f, rc); err != nil {
		return err
	} else if err := cmd.syncFile(f); err != nil {
		return err
	}
	return f.Close()
}

func (cmd *BackupCommand) backupIndexTranslateData(ctx context.Context, name string) error {
	partitionN := disco.DefaultPartitionN

	ch := make(chan int, partitionN)
	for partitionID := 0; partitionID < partitionN; partitionID++ {
		ch <- partitionID
	}
	close(ch)

	g, ctx := errgroup.WithContext(ctx)
	for i := 0; i < cmd.Concurrency; i++ {
		g.Go(func() error {
			for {
				select {
				case <-ctx.Done():
					return ctx.Err()
				case partitionID, ok := <-ch:
					if !ok {
						return nil
					} else if err := cmd.backupIndexPartitionTranslateData(ctx, name, partitionID); err != nil {
						return fmt.Errorf("cannot backup index translation data for partition %d on %q: %w", partitionID, name, err)
					}
				}
			}
		})
	}
	return g.Wait()
}

func (cmd *BackupCommand) backupIndexPartitionTranslateData(ctx context.Context, name string, partitionID int) error {
	logger := cmd.Logger()
	logger.Printf("backing up index translation data: %s/%d", name, partitionID)

	rc, err := cmd.client.IndexTranslateDataReader(ctx, name, partitionID)
	if err != nil {
		return fmt.Errorf("fetching translate data reader: %w", err)
	}
	defer rc.Close()

	filename := filepath.Join(cmd.OutputDir, "indexes", name, "translate", fmt.Sprintf("%04d", partitionID))
	if err := os.MkdirAll(filepath.Dir(filename), 0o750); err != nil {
		return err
	}

	f, err := os.Create(filename)
	if err != nil {
		return err
	}
	defer f.Close()

	if _, err := io.Copy(f, rc); err != nil {
		return err
	} else if err := cmd.syncFile(f); err != nil {
		return err
	}
	return f.Close()
}

func (cmd *BackupCommand) backupFieldTranslateData(ctx context.Context, indexName, fieldName string) error {
	logger := cmd.Logger()
	logger.Printf("backing up field translation data: %s/%s", indexName, fieldName)

	rc, err := cmd.client.FieldTranslateDataReader(ctx, indexName, fieldName)
	if err != nil {
		return fmt.Errorf("fetching translate data reader: %w", err)
	}
	defer rc.Close()

	filename := filepath.Join(cmd.OutputDir, "indexes", indexName, "fields", fieldName, "translate")
	if err := os.MkdirAll(filepath.Dir(filename), 0o750); err != nil {
		return err
	}

	f, err := os.Create(filename)
	if err != nil {
		return err
	}
	defer f.Close()

	if _, err := io.Copy(f, rc); err != nil {
		return err
	} else if err := cmd.syncFile(f); err != nil {
		return err
	}
	return f.Close()
}

func (cmd *BackupCommand) syncFile(f *os.File) error {
	if cmd.NoSync {
		return nil
	}
	return f.Sync()
}

func (cmd *BackupCommand) TLSHost() string { return cmd.Host }

func (cmd *BackupCommand) TLSConfiguration() server.TLSConfig { return cmd.TLS }

// syncDirectories fsyncs all directories required for the backup to be persisted to the filesystem.
func (cmd *BackupCommand) syncDirectories(ctx context.Context) error {
	if cmd.NoSync {
		return nil
	}

	syncChan := make(chan string, cmd.Concurrency)
	syncChan <- filepath.Dir(cmd.OutputDir)
	g, ctx := errgroup.WithContext(ctx)
	for i := 0; i < cmd.Concurrency; i++ {
		g.Go(func() error {
			for {
				select {
				case <-ctx.Done():
					return ctx.Err()
				case path, ok := <-syncChan:
					if !ok {
						return nil
					} else if err := cmd.syncDir(path); err != nil {
						return fmt.Errorf("cannot sync directory %q: %w", path, err)
					}
				}
			}
		})
	}

	err := filepath.Walk(cmd.OutputDir, func(path string, info os.FileInfo, err error) error {
		if err != nil {
			return err
		}

		if info.IsDir() {
			select {
			case <-ctx.Done():
				return ctx.Err()
			case syncChan <- path:
			}
		}

		return nil
	})
	close(syncChan)
	if err != nil {
		return fmt.Errorf("walking output directory tree: %w", err)
	}

	return g.Wait()
}

func (cmd *BackupCommand) syncDir(path string) error {
	logger := cmd.Logger()
	logger.Printf("syncing directory: %s", path)

	f, err := os.Open(path)
	if err != nil {
		return fmt.Errorf("opening directory for sync: %w", err)
	}
	defer f.Close()

	err = f.Sync()
	if err != nil {
		return fmt.Errorf("syncing directory: %w", err)
	}

	return f.Close()
}

func (cmd *BackupCommand) backupShardDataframe(ctx context.Context, indexName string, shard uint64, node *disco.Node) error {
	logger := cmd.Logger()
	logger.Printf("backing up dataframe shard: index=%q shard=%d", indexName, shard)

	client := pilosa.NewInternalClientFromURI(&node.URI,
		pilosa.GetHTTPClient(cmd.tlsConfig, pilosa.ClientResponseHeaderTimeoutOption(cmd.HeaderTimeout)),
	)

	resp, err := client.GetDataframeShard(ctx, indexName, shard)
	// no error if doesn't exist
	if err != nil {
		return fmt.Errorf("getting dataframe: %w", err)
	}
	defer resp.Body.Close()
	if resp.StatusCode == 404 {
		// no error if not present server maynot have it turned on
		return nil
	}
	filename := filepath.Join(cmd.OutputDir, "indexes", indexName, "dataframe", fmt.Sprintf("%04d", shard))
	if err := os.MkdirAll(filepath.Dir(filename), 0o750); err != nil {
		return err
	}

	f, err := os.Create(filename)
	if err != nil {
		return err
	}
	defer f.Close()

	if _, err := io.Copy(f, resp.Body); err != nil {
		return err
	} else if err := cmd.syncFile(f); err != nil {
		return err
	}
	return f.Close()
}