iceberg: add sort-aware compaction rewrite (#8666)

* iceberg: add sort-aware compaction rewrite

* iceberg: share filtered row iteration in compaction

* iceberg: rely on table sort order for sort rewrites

* iceberg: harden sort compaction planning

* iceberg: include rewrite strategy in planning config hash

compactionPlanningConfigHash now incorporates RewriteStrategy and
SortMaxInputBytes so cached planning results are invalidated when
sort strategy settings change. Also use the bytesPerMB constant in
compactionNoEligibleMessage.
This commit is contained in:
Chris Lu
2026-03-17 00:57:32 -07:00
committed by GitHub
parent e5c0889473
commit 55e988a7ee
8 changed files with 950 additions and 135 deletions

View File

@@ -22,6 +22,7 @@ const (
defaultDeleteMinInputFiles = 2
defaultDeleteMaxGroupSizeMB = 256
defaultDeleteMaxOutputFiles = 8
defaultRewriteStrategy = "binpack"
defaultMinManifestsToRewrite = 5
minManifestsToRewrite = 2
defaultOperations = "all"
@@ -43,6 +44,8 @@ const (
MetricDurationMs = "duration_ms"
)
const bytesPerMB int64 = 1024 * 1024
// Config holds parsed worker config values.
type Config struct {
SnapshotRetentionHours int64
@@ -60,7 +63,6 @@ type Config struct {
ApplyDeletes bool
Where string
RewriteStrategy string
SortFields string
SortMaxInputBytes int64
}
@@ -72,19 +74,18 @@ func ParseConfig(values map[string]*plugin_pb.ConfigValue) Config {
MaxSnapshotsToKeep: readInt64Config(values, "max_snapshots_to_keep", defaultMaxSnapshotsToKeep),
OrphanOlderThanHours: readInt64Config(values, "orphan_older_than_hours", defaultOrphanOlderThanHours),
MaxCommitRetries: readInt64Config(values, "max_commit_retries", defaultMaxCommitRetries),
TargetFileSizeBytes: readInt64Config(values, "target_file_size_mb", defaultTargetFileSizeMB) * 1024 * 1024,
TargetFileSizeBytes: readSizeMBConfig(values, "target_file_size_mb", defaultTargetFileSizeMB),
MinInputFiles: readInt64Config(values, "min_input_files", defaultMinInputFiles),
DeleteTargetFileSizeBytes: readInt64Config(values, "delete_target_file_size_mb", defaultDeleteTargetFileSizeMB) * 1024 * 1024,
DeleteTargetFileSizeBytes: readSizeMBConfig(values, "delete_target_file_size_mb", defaultDeleteTargetFileSizeMB),
DeleteMinInputFiles: readInt64Config(values, "delete_min_input_files", defaultDeleteMinInputFiles),
DeleteMaxFileGroupSizeBytes: readInt64Config(values, "delete_max_file_group_size_mb", defaultDeleteMaxGroupSizeMB) * 1024 * 1024,
DeleteMaxFileGroupSizeBytes: readSizeMBConfig(values, "delete_max_file_group_size_mb", defaultDeleteMaxGroupSizeMB),
DeleteMaxOutputFiles: readInt64Config(values, "delete_max_output_files", defaultDeleteMaxOutputFiles),
MinManifestsToRewrite: readInt64Config(values, "min_manifests_to_rewrite", defaultMinManifestsToRewrite),
Operations: readStringConfig(values, "operations", defaultOperations),
ApplyDeletes: readBoolConfig(values, "apply_deletes", true),
Where: strings.TrimSpace(readStringConfig(values, "where", "")),
RewriteStrategy: strings.TrimSpace(strings.ToLower(readStringConfig(values, "rewrite_strategy", "binpack"))),
SortFields: strings.TrimSpace(readStringConfig(values, "sort_fields", "")),
SortMaxInputBytes: readInt64Config(values, "sort_max_input_mb", 0) * 1024 * 1024,
RewriteStrategy: strings.TrimSpace(strings.ToLower(readStringConfig(values, "rewrite_strategy", defaultRewriteStrategy))),
SortMaxInputBytes: readSizeMBConfig(values, "sort_max_input_mb", 0),
}
// Clamp the fields that are always defaulted by worker config parsing.
@@ -98,15 +99,6 @@ func ParseConfig(values map[string]*plugin_pb.ConfigValue) Config {
cfg.MaxCommitRetries = defaultMaxCommitRetries
}
cfg = applyThresholdDefaults(cfg)
if cfg.RewriteStrategy == "" {
cfg.RewriteStrategy = "binpack"
}
if cfg.RewriteStrategy != "binpack" && cfg.RewriteStrategy != "sort" {
cfg.RewriteStrategy = "binpack"
}
if cfg.SortMaxInputBytes < 0 {
cfg.SortMaxInputBytes = 0
}
return cfg
}
@@ -132,12 +124,34 @@ func applyThresholdDefaults(cfg Config) Config {
if cfg.DeleteMaxOutputFiles <= 0 {
cfg.DeleteMaxOutputFiles = defaultDeleteMaxOutputFiles
}
if cfg.RewriteStrategy == "" {
cfg.RewriteStrategy = defaultRewriteStrategy
}
if cfg.RewriteStrategy != "binpack" && cfg.RewriteStrategy != "sort" {
cfg.RewriteStrategy = defaultRewriteStrategy
}
if cfg.SortMaxInputBytes < 0 {
cfg.SortMaxInputBytes = 0
}
if cfg.MinManifestsToRewrite < minManifestsToRewrite {
cfg.MinManifestsToRewrite = minManifestsToRewrite
}
return cfg
}
func readSizeMBConfig(values map[string]*plugin_pb.ConfigValue, field string, fallbackMB int64) int64 {
mb := readInt64Config(values, field, fallbackMB)
if mb <= 0 {
return 0
}
maxMB := int64(^uint64(0)>>1) / bytesPerMB
if mb > maxMB {
glog.V(1).Infof("readSizeMBConfig: clamping %q from %d MB to %d MB", field, mb, maxMB)
mb = maxMB
}
return mb * bytesPerMB
}
// parseOperations returns the ordered list of maintenance operations to execute.
// Order follows Iceberg best practices: compact → rewrite_position_delete_files
// → expire_snapshots → remove_orphans → rewrite_manifests.