Files
seaweedfs/weed/s3api/iceberg/handlers_commit.go
T
Nguyễn Lộc Phúcandchrislusf b62bbb99fc fix(iceberg): serve a registered table's metadata on load and commit (#12159)
* iceberg: serve a registered table's metadata on load and commit

RegisterTable saved only the metadata location, and loads and commits
never read the file there. They served a placeholder with no schema, no
snapshots and a new random table UUID on every load, so a registered
table read as empty, an INSERT failed against the empty schema, and
every other commit failed its assert-table-uuid requirement. Metadata
uploaded through the S3 gateway, as a migrated table's is, is stored in
chunks that registration did not read, so it registered an empty table.

Read the file following its chunks, reject one that is not Iceberg
metadata with a 400, and keep its content and table UUID in the record
the way CreateTable does.

Records registered before this change hold only the location. Loads and
commits now read the file there, and the first commit stores it in the
record under the file's table UUID.

* iceberg: transactions read a location-only registration's metadata file

* iceberg: cover transactions on a location-only registration

* iceberg: refresh the default name mapping on transaction commits

* iceberg: cover name-mapping refresh in transactions

* iceberg: roll a transaction back to a registration's metadata file

A transaction that failed after flipping a table registered before the
record kept its metadata could not undo the flip. The rollback restored
the record's prior metadata, which for such a table is only a location,
and UpdateTable keeps the full metadata a flip stored when the restore
sends none. Loads went on serving the change the client was told had
failed, and a client retrying the transaction could apply it twice.

Restore the metadata loads served from the file instead, under the
file's table UUID.

* iceberg: register oversized metadata by its location

Registration sends the metadata file's content to the S3 Tables
manager, which refuses a request over its 10MB body limit, so a table
whose metadata is larger could no longer be registered at all.

Register such a file by its location and table UUID alone; loads read
it from that location, as they do for a record that never kept its
metadata. Export the limit as s3tables.MaxRequestBodySize so the
catalog can check against it. A commit to such a table still sends the
whole metadata to UpdateTable and stays over the limit.

* iceberg: commit oversized metadata by its location

* iceberg: let maintenance follow a location-only table record

* admin: read a table's metadata file when the record keeps none

* iceberg: bound the metadata file a location-only record reads

* iceberg: bound only chunked metadata reads in the worker

Inline content is what the catalog itself wrote and the REST load path
accepts it at any size, so rejecting it in loadMetadata stranded tables
over 64 MiB from every maintenance scan. Cap only the chunked read, where
the recorded location can name any object in the bucket.

---------

Co-authored-by: chrislusf <chrislusf@users.noreply.github.com>
2026-10-28 08:15:42 -07:00

422 lines
16 KiB
Go

package iceberg
import (
"encoding/json"
"errors"
"fmt"
"net/http"
"path"
"strings"
"github.com/apache/iceberg-go/table"
"github.com/google/uuid"
"github.com/gorilla/mux"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3_constants"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3tables"
)
// handleUpdateTable commits updates to a table.
// Implements the Iceberg REST Catalog commit protocol.
func (s *Server) handleUpdateTable(w http.ResponseWriter, r *http.Request) {
vars := mux.Vars(r)
namespace := parseNamespace(vars["namespace"])
tableName := vars["table"]
if len(namespace) == 0 || tableName == "" {
writeError(w, http.StatusBadRequest, "BadRequestException", "Namespace and table name are required")
return
}
bucketName := getBucketFromPrefix(r)
bucketARN := buildTableBucketARN(bucketName)
// Extract identity from context
identityName := s3_constants.GetIdentityNameFromContext(r)
// Parse commit request and keep statistics updates separate because iceberg-go v0.4.0
// does not decode set/remove-statistics update actions yet.
var raw struct {
Identifier *TableIdentifier `json:"identifier,omitempty"`
Requirements json.RawMessage `json:"requirements"`
Updates []json.RawMessage `json:"updates"`
}
if err := json.NewDecoder(r.Body).Decode(&raw); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Invalid request body: "+err.Error())
return
}
var req CommitTableRequest
req.Identifier = raw.Identifier
var statisticsUpdates []statisticsUpdate
if len(raw.Requirements) > 0 {
normalized, err := normalizeRequirements(raw.Requirements)
if err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Invalid requirements: "+err.Error())
return
}
if err := json.Unmarshal(normalized, &req.Requirements); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Invalid requirements: "+err.Error())
return
}
}
if len(raw.Updates) > 0 {
var err error
req.Updates, statisticsUpdates, err = parseCommitUpdates(raw.Updates)
if err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Invalid updates: "+err.Error())
return
}
}
// Manifest repair runs once, as soon as the table location is known; on
// commit retries the updates already reference the repaired files. Repair
// is best effort end to end: the originals parsed already, so a repair
// that fails to re-parse is discarded rather than failing the commit.
manifestsRepaired := false
repairManifests := func(location string) {
if manifestsRepaired {
return
}
manifestsRepaired = true
repaired, changed := s.repairAddSnapshotManifests(r.Context(), location, raw.Updates)
if !changed {
return
}
repairedUpdates, repairedStatistics, err := parseCommitUpdates(repaired)
if err != nil {
glog.Warningf("Iceberg: repaired updates failed to parse, keeping originals: %v", err)
return
}
raw.Updates = repaired
req.Updates = repairedUpdates
statisticsUpdates = repairedStatistics
}
maxCommitAttempts := 3
generatedLegacyUUID := uuid.New()
stageCreateEnabled := isStageCreateEnabled()
for attempt := 1; attempt <= maxCommitAttempts; attempt++ {
getReq := &s3tables.GetTableRequest{
TableBucketARN: bucketARN,
Namespace: namespace,
Name: tableName,
}
var getResp s3tables.GetTableResponse
err := s.filerClient.WithFilerClient(false, func(client filer_pb.SeaweedFilerClient) error {
mgrClient := s3tables.NewManagerClient(client)
return s.tablesManager.Execute(r.Context(), mgrClient, "GetTable", getReq, &getResp, identityName)
})
if err != nil {
if isS3TablesNotFound(err) {
location := fmt.Sprintf("s3://%s/%s", bucketName, path.Join(flattenNamespacePath(namespace), tableName))
tableUUID := generatedLegacyUUID
baseMetadataVersion := 0
baseMetadataLocation := ""
var baseMetadata table.Metadata
var latestMarker *stageCreateMarker
if stageCreateEnabled {
var markerErr error
latestMarker, markerErr = s.loadLatestStageCreateMarker(r.Context(), bucketName, namespace, tableName)
if markerErr != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to load stage-create marker: "+markerErr.Error())
return
}
}
if latestMarker != nil {
if latestMarker.Location != "" {
location = strings.TrimSuffix(latestMarker.Location, "/")
}
if latestMarker.TableUUID != "" {
if parsedUUID, parseErr := uuid.Parse(latestMarker.TableUUID); parseErr == nil {
tableUUID = parsedUUID
}
}
stagedMetadataLocation := latestMarker.StagedMetadataLocation
if stagedMetadataLocation == "" {
stagedMetadataLocation = fmt.Sprintf("%s/metadata/v1.metadata.json", strings.TrimSuffix(location, "/"))
}
stagedLocation := tableLocationFromMetadataLocation(stagedMetadataLocation)
stagedFileName := path.Base(stagedMetadataLocation)
stagedBucket, stagedPath, parseLocationErr := parseS3Location(stagedLocation)
if parseLocationErr != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Invalid staged metadata location: "+parseLocationErr.Error())
return
}
if err := confineMetadataLocation(stagedBucket, stagedPath, bucketName); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", err.Error())
return
}
stagedMetadataBytes, loadErr := s.loadMetadataFile(r.Context(), stagedBucket, stagedPath, stagedFileName)
if loadErr != nil {
if !errors.Is(loadErr, filer_pb.ErrNotFound) {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to load staged metadata: "+loadErr.Error())
return
}
} else if len(stagedMetadataBytes) > 0 {
stagedMetadata, parseErr := table.ParseMetadataBytes(stagedMetadataBytes)
if parseErr != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to parse staged metadata: "+parseErr.Error())
return
}
// Staged metadata is only a template for table creation; commit starts from version 1.
baseMetadata = stagedMetadata
baseMetadataLocation = ""
baseMetadataVersion = 0
if stagedMetadata.TableUUID() != uuid.Nil {
tableUUID = stagedMetadata.TableUUID()
}
}
}
hasAssertCreate := hasAssertCreateRequirement(req.Requirements)
hasStagedTemplate := baseMetadata != nil
if !(stageCreateEnabled && (hasAssertCreate || hasStagedTemplate)) {
writeError(w, http.StatusNotFound, "NoSuchTableException", fmt.Sprintf("Table does not exist: %s", tableName))
return
}
// From here the commit creates the table, writing its metadata
// file before the create that authorizes it.
if authErr := s.authorizeCreateTable(r.Context(), bucketARN, namespace, tableName, identityName); authErr != nil {
writeManagerError(w, authErr)
return
}
for _, requirement := range req.Requirements {
validateAgainst := table.Metadata(nil)
if hasStagedTemplate && requirement.GetType() != requirementAssertCreate {
validateAgainst = baseMetadata
}
if requirementErr := requirement.Validate(validateAgainst); requirementErr != nil {
writeError(w, http.StatusConflict, "CommitFailedException", "Requirement failed: "+requirementErr.Error())
return
}
}
if baseMetadata == nil {
var buildErr error
if baseMetadata, buildErr = newTableMetadata(tableUUID, location, nil, nil, nil, nil); buildErr != nil {
glog.Errorf("Iceberg: CommitTable placeholder metadata for %s: %v", tableName, buildErr)
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to build current metadata")
return
}
}
createBucket, createPath, createLocErr := parseS3Location(location)
if createLocErr != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Invalid table location: "+createLocErr.Error())
return
}
if err := confineMetadataLocation(createBucket, createPath, bucketName); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", err.Error())
return
}
repairManifests(location)
result, reqErr := s.finalizeCreateOnCommit(r.Context(), createOnCommitInput{
bucketARN: bucketARN,
markerBucket: bucketName,
namespace: namespace,
tableName: tableName,
identityName: identityName,
location: location,
tableUUID: tableUUID,
baseMetadata: baseMetadata,
baseMetadataLoc: baseMetadataLocation,
baseMetadataVer: baseMetadataVersion,
updates: req.Updates,
statisticsUpdates: statisticsUpdates,
})
if reqErr != nil {
writeError(w, reqErr.status, reqErr.errType, reqErr.message)
return
}
writeJSON(w, http.StatusOK, result)
return
}
glog.V(1).Infof("Iceberg: CommitTable GetTable error: %v", err)
writeManagerError(w, err)
return
}
if reqErr := s.authorizeCommit(r.Context(), bucketARN, namespace, tableName, identityName); reqErr != nil {
writeError(w, reqErr.status, reqErr.errType, reqErr.message)
return
}
location := tableLocationFromMetadataLocation(getResp.MetadataLocation)
if location == "" {
location = fmt.Sprintf("s3://%s/%s", bucketName, path.Join(flattenNamespacePath(namespace), tableName))
}
locBucket, locPath, locErr := parseS3Location(location)
if locErr != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Invalid table location: "+locErr.Error())
return
}
if err := confineMetadataLocation(locBucket, locPath, bucketName); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", err.Error())
return
}
tableUUID := uuid.Nil
if getResp.Metadata != nil && getResp.Metadata.Iceberg != nil && getResp.Metadata.Iceberg.TableUUID != "" {
if parsed, parseErr := uuid.Parse(getResp.Metadata.Iceberg.TableUUID); parseErr == nil {
tableUUID = parsed
}
}
currentBytes, err := s.currentMetadataBytes(r.Context(), getResp, bucketName)
if err != nil {
glog.Errorf("Iceberg: CommitTable load current metadata for %s: %v", tableName, err)
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to load current metadata")
return
}
var currentMetadata table.Metadata
if len(currentBytes) > 0 {
currentMetadata, err = table.ParseMetadataBytes(currentBytes)
if err != nil {
glog.Errorf("Iceberg: Failed to parse current metadata for %s: %v", tableName, err)
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to parse current metadata")
return
}
// A record that kept only a metadata location carries no UUID of
// its own; the one in its metadata file is the table's.
if tableUUID == uuid.Nil {
tableUUID = currentMetadata.TableUUID()
}
}
if tableUUID == uuid.Nil {
tableUUID = generatedLegacyUUID
}
if currentMetadata == nil {
currentMetadata, err = newTableMetadata(tableUUID, location, nil, nil, nil, nil)
if err != nil {
glog.Errorf("Iceberg: CommitTable placeholder metadata for %s: %v", tableName, err)
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to build current metadata")
return
}
}
for _, requirement := range req.Requirements {
if err := requirement.Validate(currentMetadata); err != nil {
writeError(w, http.StatusConflict, "CommitFailedException", "Requirement failed: "+err.Error())
return
}
}
repairManifests(location)
builder, err := table.MetadataBuilderFromBase(currentMetadata, getResp.MetadataLocation)
if err != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to create metadata builder: "+err.Error())
return
}
for _, update := range req.Updates {
if err := update.Apply(builder); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Failed to apply update: "+err.Error())
return
}
}
newMetadata, err := builder.Build()
if err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Failed to build new metadata: "+err.Error())
return
}
metadataVersion := getResp.MetadataVersion + 1
metadataFileName := fmt.Sprintf("v%d.metadata.json", metadataVersion)
newMetadataLocation := fmt.Sprintf("%s/metadata/%s", strings.TrimSuffix(location, "/"), metadataFileName)
metadataBytes, err := json.Marshal(newMetadata)
if err != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to serialize metadata: "+err.Error())
return
}
// iceberg-go does not currently support set/remove-statistics updates in MetadataBuilder.
// Patch the encoded metadata JSON and parse it back to keep the response object consistent.
metadataBytes, err = applyStatisticsUpdates(metadataBytes, statisticsUpdates)
if err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", "Failed to apply statistics updates: "+err.Error())
return
}
metadataBytes = refreshDefaultNameMapping(metadataBytes, newMetadata)
// Same spec-compliance fixup we apply on create-table; ensures
// v{N}.metadata.json files written during commit are also readable by
// strict Iceberg clients reading directly from S3, and that the
// FullMetadata persisted in S3Tables stays consistent.
metadataBytes = ensureMetadataSpecCompliance(metadataBytes)
newMetadata, err = table.ParseMetadataBytes(metadataBytes)
if err != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to parse committed metadata: "+err.Error())
return
}
metadataBucket, metadataPath, err := parseS3Location(location)
if err != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Invalid table location: "+err.Error())
return
}
if err := confineMetadataLocation(metadataBucket, metadataPath, bucketName); err != nil {
writeError(w, http.StatusBadRequest, "BadRequestException", err.Error())
return
}
metadataFileName, newMetadataLocation, err = s.stageCommitMetadata(r.Context(), metadataBucket, metadataPath, location, metadataFileName, metadataBytes)
if err != nil {
writeError(w, http.StatusInternalServerError, "InternalServerError", "Failed to save metadata file: "+err.Error())
return
}
updateReq := &s3tables.UpdateTableRequest{
TableBucketARN: bucketARN,
Namespace: namespace,
Name: tableName,
VersionToken: getResp.VersionToken,
Metadata: &s3tables.TableMetadata{
Iceberg: &s3tables.IcebergMetadata{
TableUUID: tableUUID.String(),
},
FullMetadata: metadataBytes,
},
MetadataVersion: metadataVersion,
MetadataLocation: newMetadataLocation,
}
fitUpdateTableRequest(updateReq)
err = s.filerClient.WithFilerClient(false, func(client filer_pb.SeaweedFilerClient) error {
mgrClient := s3tables.NewManagerClient(client)
return s.tablesManager.Execute(r.Context(), mgrClient, "UpdateTable", updateReq, nil, identityName)
})
if err == nil {
result := CommitTableResponse{
MetadataLocation: newMetadataLocation,
Metadata: newMetadata,
}
writeJSON(w, http.StatusOK, result)
return
}
if isS3TablesConflict(err) {
if cleanupErr := s.deleteMetadataFile(r.Context(), metadataBucket, metadataPath, metadataFileName); cleanupErr != nil {
glog.V(1).Infof("Iceberg: failed to cleanup metadata file %s on conflict: %v", newMetadataLocation, cleanupErr)
}
if attempt < maxCommitAttempts {
glog.V(1).Infof("Iceberg: CommitTable conflict for %s (attempt %d/%d), retrying", tableName, attempt, maxCommitAttempts)
sleepBeforeCommitRetry(attempt)
continue
}
writeError(w, http.StatusConflict, "CommitFailedException", "Version token mismatch")
return
}
if cleanupErr := s.deleteMetadataFile(r.Context(), metadataBucket, metadataPath, metadataFileName); cleanupErr != nil {
glog.V(1).Infof("Iceberg: failed to cleanup metadata file %s after update failure: %v", newMetadataLocation, cleanupErr)
}
logManagerError("CommitTable UpdateTable", err)
writeManagerError(w, err)
return
}
}