package main

import (
	"bufio"
	"bytes"
	"context"
	"database/sql"
	"encoding/base64"
	"encoding/json"
	"fmt"
	"io"
	"log"
	"mime/multipart"
	"net/http"
	"os"
	"path"
	"path/filepath"
	"strings"
	"time"
	"unicode/utf8"
	"unsafe"

	golake "github.com/tidbcloud/lake-go"
	"github.com/google/uuid"
	"github.com/klauspost/compress/zstd"
	"github.com/pkg/errors"
	"github.com/sirupsen/logrus"
	"github.com/xitongsys/parquet-go/writer"

	"github.com/xitongsys/parquet-go-source/local"

	"github.com/tidbcloud/lake-ingest-kafka/config"
	"github.com/tidbcloud/lake-ingest-kafka/message"
)

var (
	ErrUploadStageFailed   = errors.New("upload stage failed")
	ErrCopyIntoFailed      = errors.New("copy into failed")
	ErrStreamingLoadFailed = errors.New("streaming load failed")
)

func init() {
	// UUIDs are row identifiers, not security tokens. Pooling removes one
	// crypto/rand syscall per row on the raw ingest path.
	uuid.EnableRandPool()
}

type LakeIngester interface {
	IngestData(messageBatch *message.MessagesBatch) error
	IngestParquetData(messageBatch *message.MessagesBatch) error
	CreateRawTargetTable() error
	Close() error
}

type lakeIngester struct {
	lakeIngesterCfg *config.Config
	statsRecorder       *LakeIngesterStatsRecorder
	db                  *sql.DB
	httpClient          *http.Client
}

type kafkaRecordMetadata struct {
	Topic       string `json:"topic"`
	Partition   int    `json:"partition"`
	Offset      int64  `json:"offset"`
	Key         string `json:"key"`
	KeyEncoding string `json:"key_encoding,omitempty"`
	CreateTime  string `json:"create_time"`
}

type ndjsonRecord struct {
	UUID           string              `json:"uuid"`
	Koffset        int64               `json:"koffset"`
	Kpartition     int                 `json:"kpartition"`
	RecordMetadata kafkaRecordMetadata `json:"record_metadata"`
	AddTime        string              `json:"add_time"`
	RawData        json.RawMessage     `json:"raw_data"`
}

func byteSliceToString(bs []byte) string {
	if len(bs) == 0 {
		return ""
	}
	return unsafe.String(unsafe.SliceData(bs), len(bs))
}

func NewLakeIngester(cfg *config.Config) LakeIngester {
	stats := NewLakeIntesterStatsRecorder()
	db, err := sql.Open("lake", cfg.LakeDSN)
	if err != nil {
		log.Fatalf("open lake failed: %v", err)
	}
	return &lakeIngester{
		lakeIngesterCfg: cfg,
		statsRecorder:       stats,
		db:                  db,
		httpClient:          &http.Client{Timeout: 5 * time.Minute},
	}
}

func (ig *lakeIngester) Close() error {
	if ig.db != nil {
		return ig.db.Close()
	}
	return nil
}

func buildStreamingLoadURL(cfg *golake.Config) string {
	scheme := "https"
	if cfg.SSLMode == "disable" {
		scheme = "http"
	}
	return fmt.Sprintf("%s://%s/v1/streaming_load", scheme, cfg.Host)
}

func (ig *lakeIngester) streamingLoad(ctx context.Context, batchJsonData []string) (int, error) {
	startTime := time.Now()

	dbCfg, err := golake.ParseDSN(ig.lakeIngesterCfg.LakeDSN)
	if err != nil {
		return 0, errors.Wrap(err, "parse lake dsn failed")
	}

	endpoint := buildStreamingLoadURL(dbCfg)

	pr, pw := io.Pipe()
	mw := multipart.NewWriter(pw)
	bytesCh := make(chan int, 1)

	go func() {
		defer pw.Close()
		part, partErr := mw.CreateFormFile("upload", "batch.ndjson")
		if partErr != nil {
			pw.CloseWithError(partErr)
			bytesCh <- 0
			return
		}
		n := 0
		for _, line := range batchJsonData {
			written, writeErr := io.WriteString(part, line)
			if writeErr != nil {
				pw.CloseWithError(writeErr)
				bytesCh <- n
				return
			}
			n += written
			written, writeErr = io.WriteString(part, "\n")
			if writeErr != nil {
				pw.CloseWithError(writeErr)
				bytesCh <- n
				return
			}
			n += written
		}
		mw.Close()
		bytesCh <- n
	}()

	insertSQL := fmt.Sprintf(
		// _databend_load is the server-side stage placeholder (STAGE_PLACEHOLDER in the
		// query engine); it is protocol, not branding.
		"INSERT INTO %s FROM @_databend_load FILE_FORMAT=(type=NDJSON missing_field_as=FIELD_DEFAULT COMPRESSION=AUTO)",
		ig.lakeIngesterCfg.LakeTable,
	)

	req, err := http.NewRequestWithContext(ctx, http.MethodPut, endpoint, pr)
	if err != nil {
		return 0, errors.Wrap(ErrStreamingLoadFailed, err.Error())
	}
	// Wire protocol: these header names are fixed by the server and must not be
	// rebranded. The gateway rejects streaming load without X-DATABEND-WAREHOUSE.
	req.Header.Set("Content-Type", mw.FormDataContentType())
	req.Header.Set("X-Databend-SQL", insertSQL)
	if dbCfg.User != "" {
		req.SetBasicAuth(dbCfg.User, dbCfg.Password)
	}
	if dbCfg.Warehouse != "" {
		req.Header.Set("X-Databend-Warehouse", dbCfg.Warehouse)
	}

	resp, err := ig.httpClient.Do(req)
	bytesWritten := <-bytesCh
	if err != nil {
		return 0, errors.Wrap(ErrStreamingLoadFailed, err.Error())
	}
	defer resp.Body.Close()

	body, _ := io.ReadAll(resp.Body)
	if resp.StatusCode >= 400 {
		return 0, errors.Wrap(ErrStreamingLoadFailed,
			fmt.Sprintf("status=%d body=%s", resp.StatusCode, string(body)))
	}

	logrus.Infof("streaming load %s, rows=%d bytes=%d cost=%s",
		ig.lakeIngesterCfg.LakeTable, len(batchJsonData), bytesWritten, time.Since(startTime))
	return bytesWritten, nil
}

// sanitizeUTF8 removes or replaces invalid UTF-8 sequences from a string
func sanitizeUTF8(s string) string {
	if utf8.ValidString(s) {
		return s
	}

	// Method 1: Use strings.ToValidUTF8 to replace invalid sequences with empty string
	// This is the most reliable approach using Go standard library
	var b strings.Builder
	b.Grow(len(s))

	for len(s) > 0 {
		r, size := utf8.DecodeRuneInString(s)
		if r == utf8.RuneError && size == 1 {
			// Invalid UTF-8 byte, skip it
			s = s[1:]
			continue
		}
		b.WriteRune(r)
		s = s[size:]
	}

	return b.String()
}

func sanitizeKafkaKey(key string) (string, string) {
	if key == "" {
		return "", ""
	}
	if utf8.ValidString(key) {
		return key, ""
	}
	return base64.StdEncoding.EncodeToString([]byte(key)), "base64"
}

func prepareRawData(data string, offset int64) (json.RawMessage, error) {
	payload := data
	if !utf8.ValidString(payload) {
		payload = sanitizeUTF8(payload)
	}

	raw := bytes.TrimSpace([]byte(payload))
	if len(raw) == 0 {
		return json.RawMessage([]byte("null")), nil
	}

	if !json.Valid(raw) {
		return nil, fmt.Errorf("invalid json payload at offset %d", offset)
	}

	return json.RawMessage(raw), nil
}

func (ig *lakeIngester) buildNDJSONRecord(msg message.MessageData) (ndjsonRecord, error) {
	return ig.buildNDJSONRecordWithAddTime(msg, time.Now().Format(time.RFC3339Nano))
}

func (ig *lakeIngester) buildNDJSONRecordWithAddTime(msg message.MessageData, addTime string) (ndjsonRecord, error) {
	rawData, err := prepareRawData(msg.Data, msg.DataOffset)
	if err != nil {
		return ndjsonRecord{}, err
	}
	keyValue, encoding := sanitizeKafkaKey(msg.Key)

	record := ndjsonRecord{
		UUID:       uuid.New().String(),
		Koffset:    msg.DataOffset,
		Kpartition: msg.Partition,
		RecordMetadata: kafkaRecordMetadata{
			Topic:       ig.lakeIngesterCfg.KafkaTopic,
			Partition:   msg.Partition,
			Offset:      msg.DataOffset,
			Key:         keyValue,
			KeyEncoding: encoding,
			CreateTime:  msg.CreateTime.Format(time.RFC3339Nano),
		},
		AddTime: addTime,
		RawData: rawData,
	}
	return record, nil
}

func (ig *lakeIngester) reWriteTheJsonData(messagesBatch *message.MessagesBatch) ([]string, error) {
	batchJsonData := messagesBatch.Messages
	afterHandleJsonData := make([]string, 0, len(batchJsonData))
	addTime := time.Now().Format(time.RFC3339Nano)

	for i := 0; i < len(batchJsonData); i++ {
		record, err := ig.buildNDJSONRecordWithAddTime(batchJsonData[i], addTime)
		if err != nil {
			return nil, err
		}

		jsonData, err := json.Marshal(record)
		if err != nil {
			return nil, fmt.Errorf("failed to marshal record: %w", err)
		}

		afterHandleJsonData = append(afterHandleJsonData, byteSliceToString(jsonData))
	}
	return afterHandleJsonData, nil
}

func (ig *lakeIngester) reWriteParquetJsonData(messagesBatch *message.MessagesBatch) ([]string, error) {
	records, err := ig.buildParquetRecords(messagesBatch)
	if err != nil {
		return nil, err
	}
	batchJsonData := messagesBatch.Messages
	afterHandleJsonData := make([]string, 0, len(batchJsonData))

	for i := range records {
		jsonData, err := json.Marshal(records[i])
		if err != nil {
			return nil, fmt.Errorf("failed to marshal parquet record: %w", err)
		}

		afterHandleJsonData = append(afterHandleJsonData, byteSliceToString(jsonData))
	}
	return afterHandleJsonData, nil
}

func (ig *lakeIngester) buildParquetRecords(messagesBatch *message.MessagesBatch) ([]*message.RecordForParquet, error) {
	batchJsonData := messagesBatch.Messages
	records := make([]*message.RecordForParquet, 0, len(batchJsonData))
	addTime := time.Now().Format(time.RFC3339Nano)

	for i := 0; i < len(batchJsonData); i++ {
		record, err := ig.buildNDJSONRecordWithAddTime(batchJsonData[i], addTime)
		if err != nil {
			return nil, err
		}

		metadataJSON, err := json.Marshal(record.RecordMetadata)
		if err != nil {
			return nil, fmt.Errorf("failed to marshal record metadata: %w", err)
		}

		records = append(records, &message.RecordForParquet{
			UUID:           record.UUID,
			KOffset:        record.Koffset,
			KPartition:     int32(record.Kpartition),
			RecordMetadata: byteSliceToString(metadataJSON),
			AddTime:        record.AddTime,
			RawData:        byteSliceToString(record.RawData),
		})
	}

	return records, nil
}

func (ig *lakeIngester) IngestParquetData(messageBatch *message.MessagesBatch) error {
	l := logrus.WithFields(logrus.Fields{"ingest_lake": "IngestParquetData"})
	startTime := time.Now()
	if messageBatch.Empty() {
		return nil
	}

	records, err := ig.buildParquetRecords(messageBatch)
	if err != nil {
		l.Errorf("re-write the json data failed: %v", err)
		return err
	}
	fileName, bytesSize, err := ig.generateParquetFileFromRecords(records)
	if err != nil {
		l.Errorf("generate parquet file failed: %v", err)
		return err
	}

	stage, err := ig.uploadToStage(fileName)
	if err != nil {
		l.Errorf("upload to stage failed: %v", err)
		return err
	}

	err = ig.replaceInto(stage)
	if err != nil {
		l.Errorf("replace into failed: %v", err)
		return err
	}

	ig.statsRecorder.RecordMetric(bytesSize, len(records))
	stats := ig.statsRecorder.Stats(time.Since(startTime))
	log.Printf("ingest %d rows (%f rows/s), %d bytes (%f bytes/s)", len(records), stats.RowsPerSecond, bytesSize, stats.BytesPerSecond)
	return nil
}

func (ig *lakeIngester) IngestData(messageBatch *message.MessagesBatch) error {
	if messageBatch.Empty() {
		return nil
	}

	l := logrus.WithFields(logrus.Fields{"ingest_lake": "IngestData", "lastOffset": messageBatch.LastMessageOffset})
	startTime := time.Now()
	var batchJsonData []string

	// handle batchJsonData, if isTransform is false, then the data is already in NDJson format
	// re-write the json data into NDJson format, add the uuid, record_metadata and add_time fields
	// then insert the data into the lake table
	if !ig.lakeIngesterCfg.IsJsonTransform {
		var err error
		batchJsonData, err = ig.reWriteTheJsonData(messageBatch)
		if err != nil {
			l.Errorf("re-write the json data failed: %v, lastOffset is: %d\n", err, messageBatch.LastMessageOffset)
			return err
		}
	} else {
		batchJsonData = messageBatch.ExtractMessageData()
	}

	if ig.lakeIngesterCfg.UseStreamingLoad {
		bytesSize, err := ig.streamingLoad(context.Background(), batchJsonData)
		if err != nil {
			l.Errorf("streaming load failed: %v, lastOffset is: %d, partition is %d\n", err,
				messageBatch.LastMessageOffset, messageBatch.Messages[0].Partition)
			ingestErrors.Inc()
			return err
		}
		ig.statsRecorder.RecordMetric(bytesSize, len(batchJsonData))
		ingestRowsTotal.Add(float64(len(batchJsonData)))
		ingestBytesTotal.Add(float64(bytesSize))
		stats := ig.statsRecorder.Stats(time.Since(startTime))
		log.Printf("ingest %d rows (%f rows/s), %d bytes (%f bytes/s) in %s", len(batchJsonData), stats.RowsPerSecond, bytesSize, stats.BytesPerSecond, time.Since(startTime))
		return nil
	}

	fileName, bytesSize, err := ig.generateNDJsonFile(batchJsonData)
	if err != nil {
		l.Errorf("generate NDJson file failed: %v,lastOffset is %d\n", err, messageBatch.LastMessageOffset)
		ingestErrors.Inc()
		return err
	}

	stage, err := ig.uploadToStage(fileName)
	if err != nil {
		l.Errorf("upload to stage failed: %v, lastOffset is: %d, partition is %d\n", err,
			messageBatch.LastMessageOffset, messageBatch.Messages[0].Partition)
		ingestErrors.Inc()
		return err
	}

	err = ig.copyInto(stage)
	if err != nil {
		l.Errorf("copy into failed: %v, lastOffset is: %d, partition is %d\n", err,
			messageBatch.LastMessageOffset, messageBatch.Messages[0].Partition)
		ingestErrors.Inc()
		return err
	}
	ig.statsRecorder.RecordMetric(bytesSize, len(batchJsonData))
	ingestRowsTotal.Add(float64(len(batchJsonData)))
	ingestBytesTotal.Add(float64(bytesSize))
	stats := ig.statsRecorder.Stats(time.Since(startTime))
	log.Printf("ingest %d rows (%f rows/s), %d bytes (%f bytes/s) in %s", len(batchJsonData), stats.RowsPerSecond, bytesSize, stats.BytesPerSecond, time.Since(startTime))
	return nil
}

func (ig *lakeIngester) generateParquetFile(batchJsonData []string) (string, int, error) {
	records := make([]*message.RecordForParquet, 0, len(batchJsonData))
	for _, data := range batchJsonData {
		var record message.RecordForParquet
		err := json.Unmarshal([]byte(data), &record)
		if err != nil {
			logrus.WithFields(logrus.Fields{"ingest_lake": "generateParquetFile"}).Errorf("Unmarshal json data failed: %v", err)
			return "", 0, err
		}
		records = append(records, &record)
	}
	return ig.generateParquetFileFromRecords(records)
}

func (ig *lakeIngester) generateParquetFileFromRecords(records []*message.RecordForParquet) (string, int, error) {
	l := logrus.WithFields(logrus.Fields{"ingest_lake": "generateParquetFile"})

	tmpFile := filepath.Join(os.TempDir(), fmt.Sprintf("lake-ingest-%d-%s.parquet", time.Now().Unix(), uuid.New().String()))

	fw, err := local.NewLocalFileWriter(tmpFile)
	if err != nil {
		l.Errorf("Can't create local file writer: %v", err)
		return "", 0, err
	}

	pw, err := writer.NewParquetWriter(fw, new(message.RecordForParquet), 4)
	if err != nil {
		l.Errorf("Can't create parquet writer: %v", err)
		return "", 0, err
	}

	for i := range records {
		if err = pw.Write(records[i]); err != nil {
			l.Errorf("Write record failed: %v", err)
			return "", 0, err
		}
	}
	if err = pw.WriteStop(); err != nil {
		l.Errorf("Write stop failed: %v", err)
		return "", 0, err
	}

	if err = fw.Close(); err != nil {
		l.Errorf("Close file writer failed: %v", err)
		return "", 0, err
	}
	fileSize := getFileSize(tmpFile)

	return tmpFile, fileSize, nil
}

func getFileSize(filename string) int {
	fi, err := os.Stat(filename)
	if err != nil {
		log.Fatal(err)
	}
	return int(fi.Size())
}

func (ig *lakeIngester) generateNDJsonFile(batchJsonData []string) (string, int, error) {
	pattern := "lake-ingest-*.ndjson"
	if ig.lakeIngesterCfg.CopyIntoUploadCompression {
		pattern = "lake-ingest-*.ndjson.zst"
	}

	outputFile, err := os.CreateTemp("/tmp", pattern)
	if err != nil {
		return "", 0, err
	}
	success := false
	defer func() {
		if !success {
			_ = os.Remove(outputFile.Name())
		}
	}()

	var writer io.Writer = outputFile
	var zstdWriter *zstd.Encoder
	if ig.lakeIngesterCfg.CopyIntoUploadCompression {
		zstdWriter, err = zstd.NewWriter(outputFile)
		if err != nil {
			_ = outputFile.Close()
			return "", 0, err
		}
		writer = zstdWriter
	}

	// Create a buffered writer for the NDJSON file.
	bufferedWriter := bufio.NewWriter(writer)
	bytesSum := 0

	for _, data := range batchJsonData {
		n, err := bufferedWriter.WriteString(data)
		if err != nil {
			_ = outputFile.Close()
			return "", 0, err
		}
		bytesSum += n
		if err := bufferedWriter.WriteByte('\n'); err != nil {
			_ = outputFile.Close()
			return "", 0, err
		}
		bytesSum++
	}
	// Flush any remaining data to the NDJSON file.
	if err = bufferedWriter.Flush(); err != nil {
		_ = outputFile.Close()
		return "", 0, err
	}
	if zstdWriter != nil {
		if err = zstdWriter.Close(); err != nil {
			_ = outputFile.Close()
			return "", 0, err
		}
	}
	if err = outputFile.Close(); err != nil {
		return "", 0, err
	}

	fileSize := bytesSum
	if ig.lakeIngesterCfg.CopyIntoUploadCompression {
		fileSize = getFileSize(outputFile.Name())
	}
	success = true
	return outputFile.Name(), fileSize, nil
}

func (ig *lakeIngester) uploadToStage(fileName string) (*golake.StageLocation, error) {
	startTime := time.Now()
	defer func() {
		err := os.RemoveAll(fileName)
		if err != nil {
			logrus.Errorf("delete batch insert file failed: %v", err)
		}
	}()

	lakeConfig, err := golake.ParseDSN(ig.lakeIngesterCfg.LakeDSN)
	if err != nil {
		return nil, err
	}
	apiClient := golake.NewAPIClientFromConfig(lakeConfig)
	fi, err := os.Stat(fileName)
	if err != nil {
		return nil, errors.Wrap(err, "get batch file size failed")
	}
	size := fi.Size()

	f, err := os.Open(fileName)
	if err != nil {
		return nil, errors.Wrap(err, "open batch file failed")
	}
	defer f.Close()
	input := bufio.NewReader(f)
	stage := &golake.StageLocation{
		Name: ig.lakeIngesterCfg.UserStage,
		Path: fmt.Sprintf("batch/%d-%s", time.Now().Unix(), filepath.Base(fileName)),
	}

	if err := apiClient.UploadToStage(context.Background(), stage, input, size); err != nil {
		return nil, errors.Wrap(ErrUploadStageFailed, err.Error())
	}

	logrus.Infof("upload to stage %s, cost: %s", stage.String(), time.Since(startTime))
	uploadStageDuration.Observe(time.Since(startTime).Seconds())
	return stage, nil
}

func execute(db *sql.DB, sql string) error {
	_, err := db.Exec(sql)
	if err != nil {
		logrus.Errorf("exec '%s' failed, err: %v", sql, err)
		return err
	}
	return nil
}

func sqlStringLiteral(s string) string {
	return "'" + strings.ReplaceAll(s, "'", "''") + "'"
}

func stageDirAndFile(stage *golake.StageLocation) (string, string) {
	dir, file := path.Split(stage.Path)
	if dir == "" {
		return fmt.Sprintf("@%s", stage.Name), file
	}
	return (&golake.StageLocation{Name: stage.Name, Path: dir}).String(), file
}

func buildCopyIntoSQL(table string, stage *golake.StageLocation, purge, force, disableVariantCheck bool) string {
	stageDir, file := stageDirAndFile(stage)
	return fmt.Sprintf("COPY INTO %s FROM %s FILES = (%s) FILE_FORMAT = (type = NDJSON missing_field_as = FIELD_DEFAULT COMPRESSION = AUTO) "+
		"PURGE = %v FORCE = %v DISABLE_VARIANT_CHECK = %v", table, stageDir, sqlStringLiteral(file),
		purge, force, disableVariantCheck)
}

func (ig *lakeIngester) copyInto(stage *golake.StageLocation) error {
	startTime := time.Now()
	copyIntoSQL := buildCopyIntoSQL(ig.lakeIngesterCfg.LakeTable, stage,
		ig.lakeIngesterCfg.CopyPurge, ig.lakeIngesterCfg.CopyForce, ig.lakeIngesterCfg.DisableVariantCheck)

	if err := execute(ig.db, copyIntoSQL); err != nil {
		return errors.Wrap(ErrCopyIntoFailed, err.Error())
	}
	logrus.Infof("copy into %s, cost: %s", ig.lakeIngesterCfg.LakeTable, time.Since(startTime))
	copyIntoDuration.Observe(time.Since(startTime).Seconds())
	return nil
}

func (ig *lakeIngester) replaceInto(stage *golake.StageLocation) error {
	replaceIntoSQL := fmt.Sprintf("REPLACE INTO %s ON (%s,%s) SELECT * FROM %s  (FILE_FORMAT => 'parquet')",
		ig.lakeIngesterCfg.LakeTable, "koffset", "kpartition", stage.String())

	defer func() {
		if ig.lakeIngesterCfg.CopyPurge {
			err := execute(ig.db, fmt.Sprintf("REMOVE  %s", stage.String()))
			if err != nil {
				logrus.Errorf("remove stage file :%s failed: %v", stage.String(), err)
			}
		}
	}()
	err := execute(ig.db, replaceIntoSQL)
	if err != nil {
		logrus.Errorf("replace into failed: %v", err)
		return err
	}
	return nil
}

func (ig *lakeIngester) CreateRawTargetTable() error {
	createTableSQL := fmt.Sprintf("CREATE TABLE IF NOT EXISTS %s (uuid String, koffset BIGINT, kpartition int, raw_data json, record_metadata json, add_time timestamp)", ig.lakeIngesterCfg.LakeTable)

	if err := execute(ig.db, createTableSQL); err != nil {
		return err
	}

	return ig.validateRawTableSchema()
}

func (ig *lakeIngester) validateRawTableSchema() error {
	requiredColumns := map[string]bool{
		"uuid":            false,
		"koffset":         false,
		"kpartition":      false,
		"raw_data":        false,
		"record_metadata": false,
		"add_time":        false,
	}

	rows, err := ig.db.Query(fmt.Sprintf("DESC %s", ig.lakeIngesterCfg.LakeTable))
	if err != nil {
		return fmt.Errorf("failed to describe table %s: %w", ig.lakeIngesterCfg.LakeTable, err)
	}
	defer rows.Close()

	for rows.Next() {
		var field, dataType, null, defaultVal, extra string
		if err := rows.Scan(&field, &dataType, &null, &defaultVal, &extra); err != nil {
			return fmt.Errorf("failed to scan column info: %w", err)
		}
		if _, ok := requiredColumns[field]; ok {
			requiredColumns[field] = true
		}
	}

	var missing []string
	for col, found := range requiredColumns {
		if !found {
			missing = append(missing, col)
		}
	}

	if len(missing) > 0 {
		return fmt.Errorf("table %s exists but is missing required columns for raw mode: %v. "+
			"Expected columns: (uuid, koffset, kpartition, raw_data, record_metadata, add_time). "+
			"Please drop or rename the existing table, or use a different table name",
			ig.lakeIngesterCfg.LakeTable, missing)
	}

	return nil
}