anoopj commented on code in PR #3564: URL: https://github.com/apache/iceberg-rust/pull/3564#discussion_r4429649709
########## crates/iceberg/src/transaction/snapshot/manifest_filter.rs: ########## @@ -0,0 +1,227 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! Rewrites existing manifests to mark files as deleted. + +use std::collections::{HashMap, HashSet}; + +use uuid::Uuid; + +use super::{manifest_path, new_manifest_writer}; +use crate::error::{Result, invalid_data}; +use crate::io::object_cache::ManifestKey; +use crate::spec::{DataFile, FormatVersion, ManifestContentType, ManifestFile}; +use crate::table::Table; + +/// Removes a fixed set of files from the manifests of one content type. +/// +/// One manager lives for one logical execution of an action and is driven +/// once per attempt via [`ManifestFilterManager::filter`]. A manifest that +/// references none of the files passes through untouched. One that does is +/// rewritten with those entries marked `Deleted` and the rest `Existing`. +/// +/// A rewrite is reused across attempts while its inputs are unchanged. Its +/// entries carry the sequence numbers and row ids the source's entries +/// inherited from the manifest list when they were read, so the reuse key is +/// the source's [`ManifestKey`], not its path alone: a transaction-local +/// source manifest keeps its path across attempts but inherits new values +/// after a rebase. +pub(crate) struct ManifestFilterManager { + content: ManifestContentType, + /// Paths of the files to remove. Fixed for the execution. + files_to_remove: HashSet<String>, + /// Source manifest → rewritten result. + filtered: HashMap<ManifestKey, FilteredManifest>, + /// Format version the rewrites were written under. Per spec, v1 entries + /// cannot inherit sequence numbers, so an upgrade invalidates them. + format_version: Option<FormatVersion>, +} + +/// One rewritten manifest and the files it marked as deleted. +#[derive(Clone)] +struct FilteredManifest { + manifest: ManifestFile, + removed_files: Vec<DataFile>, +} + +/// The result of one filter pass. +pub(crate) struct Filtered { + /// The resulting manifests, in input order. + pub manifests: Vec<ManifestFile>, + /// The files that were live and are now marked deleted, including their + /// metrics as read from the source manifests. + pub removed_files: Vec<DataFile>, +} + +impl ManifestFilterManager { + pub(crate) fn new(content: ManifestContentType, files_to_remove: &[DataFile]) -> Self { + Self { + content, + files_to_remove: files_to_remove + .iter() + .map(|f| f.file_path().to_string()) + .collect(), + filtered: HashMap::new(), + format_version: None, + } + } + + /// Applies the removals to `manifests`, rewriting those of this manager's + /// content type that reference a file to remove. + /// + /// Fails if a file to remove is not live in any manifest: it was never in + /// the table, or a concurrent commit already removed it. + /// + /// Rewritten manifests take their paths from `next_manifest_number` and are + /// recorded in `owned_artifacts` before being written, so a failed write + /// still leaves its path tracked for cleanup. + pub(crate) async fn filter( + &mut self, + table: &Table, + snapshot_id: i64, + commit_uuid: Uuid, + manifests: Vec<ManifestFile>, + next_manifest_number: &mut u64, + owned_artifacts: &mut HashSet<String>, + ) -> Result<Filtered> { + if self.files_to_remove.is_empty() { + return Ok(Filtered { + manifests, + removed_files: Vec::new(), + }); + } + + let format_version = table.metadata().format_version(); + if self.format_version != Some(format_version) { + self.filtered.clear(); + self.format_version = Some(format_version); + } + + let mut out = Vec::with_capacity(manifests.len()); + let mut removed_files: Vec<DataFile> = Vec::new(); + + for source in manifests { + if source.content != self.content { + out.push(source); + continue; + } + + let key = ManifestKey::new(&source); + let result = match self.filtered.get(&key) { + Some(cached) => cached.clone(), + None => { + let result = self + .rewrite( + table, + snapshot_id, + commit_uuid, + &source, + next_manifest_number, + owned_artifacts, + ) + .await?; + self.filtered.insert(key, result.clone()); + result + } + }; + + removed_files.extend(result.removed_files); + out.push(result.manifest); + } + + let found: HashSet<&str> = removed_files.iter().map(|f| f.file_path()).collect(); Review Comment: Some context: The goal of the `failMissingPaths()` check is to validate that the files the operation intends to delete still exist when committing (ie no double deletions), and it is required for snapshot isolation. The spec says: > Delete operations must verify that specific files to delete are still in the table. I would let the action decide whether missing removal targets fail. A example is probably a merge-on-read DELETE (*not* a `MERGE` or `UPDATE`) with v2 positional deletes. Consider: • Existing D0 deletes position {1}. • A writes D1 = {1, 2} and requests removal of D0. • B writes D2 = {1, 3} and also requests removal of D0. If `A` succeeded, `B` can still go through because, its removal of the already-absent D0 is harmless. So an unconditional failure here will make `B` fail here. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
