Module update_planner 

Metadata-only planning of Insert and Delete updates. Each plan covers a range of pending snapshots as one net change from a base snapshot to a target snapshot. Plans do not advance ingestion progress: the caller must order Delete before Insert, then acknowledge the target. This is deliberately separate from the legacy append-only planner and List/Fetch protocol.

Structs

DeleteEntry 🔒

IcebergUpdateBinding

Persist this binding before bootstrap. Never rediscover the schema or bootstrap snapshot on replay.

IcebergUpdateCursor

Last dispatched file within a deterministic range/phase stream, not a sorted path bound. Recovery replays metadata up to this marker; normal pages retain the live manifest scan. Advancing enumeration does not acknowledge application: List orders or fences phases.

IcebergUpdatePage

IcebergUpdatePlanner

IcebergUpdateTask

IcebergUpdateTaskId

Stable within a source job. snapshot_id is the target of the planned range.

ManifestDiff 🔒

Manifests are immutable, so a manifest path present in both manifest lists contributes the same entries to both snapshots and never needs to be read for the diff itself. The list-level first_row_id of an inherited data manifest may still be assigned after a V2 to V3 upgrade.

UpdateFile 🔒

Unlike the legacy split, preserves partition constants, spec, sequence numbers and name mapping. No predicates, virtual columns, or encryption keys are allowed in this read contract.

Enums

IcebergUpdatePhase

Functions

attached_deletes 🔒

delete_change 🔒

delete_tasks 🔒

Retained files are live in both snapshots, either moved between data manifests or kept in an unchanged one. The latter are looked up only when one of their delete artifacts changed. Removed files retract the rows visible in the parent.

entry_sequence 🔒

max_sequence is the sequence number of a snapshot containing the manifest.

index_deletes 🔒

Rejects multiple live delete artifacts for one data file across manifests, which must all belong to one snapshot.

insert_tasks 🔒

New files are live in added data manifests but not in removed ones. Files moved between data manifests are retained. Same-commit delete artifacts only filter Inserts.

is_compaction 🔒

load_data_files 🔒

Live, validated data files of one data manifest, without delete descriptors.

load_deletes 🔒

Live, validated delete descriptors of one delete manifest.

load_manifest 🔒

may_contain_partition 🔒

Conservative partition-summary pruning: false only when the summaries exclude partition. Iceberg applies a position delete only to data files with its spec and partition, so such a manifest cannot hold a data file that the delete applies to.

of_content 🔒

range_len 🔒

The number of pending snapshots planned as one range. A range ends before the next compaction, and a compaction forms a range of its own: it rewrites rows without changing them, while a net diff across it would retract and re-insert every rewritten row.

read_name_mapping 🔒

stream_files 🔒

Every live data file of a snapshot, for bootstrap. Delete descriptors are indexed before data enumeration, since manifests are not ordered by referenced data path. Only one data manifest and one task page are materialized at a time, in addition to the delete index.

stream_tasks 🔒

Plans one range phase as a deterministic stream; recovery replays it up to after_path. Bootstrap enumerates the whole snapshot. Incremental phases diff the manifest lists of the range base and target instead of indexing every live file of either snapshot.

task 🔒

Type Aliases

DeleteIndex 🔒

Keyed by referenced data-file path.