This is an automated email from the ASF dual-hosted git repository.
JingsongLi pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/paimon.git
The following commit(s) were added to refs/heads/master by this push:
new 5f700f4aff [hotfix] Remove useless design docs
5f700f4aff is described below
commit 5f700f4affa01b122187b41b537cc76855d4d113
Author: JingsongLi <[email protected]>
AuthorDate: Wed Jul 15 23:55:23 2026 +0800
[hotfix] Remove useless design docs
---
.../designs/2026-07-15-pk-vector-dv-aware-build.md | 59 -----------------
...026-07-15-primary-key-vector-parallel-search.md | 75 ----------------------
2 files changed, 134 deletions(-)
diff --git a/docs/designs/2026-07-15-pk-vector-dv-aware-build.md
b/docs/designs/2026-07-15-pk-vector-dv-aware-build.md
deleted file mode 100644
index fa8873a844..0000000000
--- a/docs/designs/2026-07-15-pk-vector-dv-aware-build.md
+++ /dev/null
@@ -1,59 +0,0 @@
-# DV-aware Primary-key Vector Index Build Design
-
-## Problem Statement
-
-Primary-key vector index maintenance currently reads compact source files
without deletion-vector
-filtering and therefore indexes physical row positions that were already
deleted when an ANN build
-was scheduled. Query-time deletion-vector filtering preserves correctness, but
rebuilding the same
-active source cannot reclaim those holes.
-
-## Chosen Approach
-
-Capture the deletion vectors visible to the bucket writer when a pending ANN
build is created. Pass
-an immutable per-source snapshot into the existing `excludedPosition` hook
while preserving each
-vector's physical ordinal (`fileOffset + rowPosition`).
-
-## Design Details
-
-### Maintenance integration
-
-The bucket writer already restores and updates `BucketedDvMaintainer` before
primary-key index
-maintenance prepares its commit. Its deletion-vector factory will be passed to
the primary-key
-index maintainer and then to the vector maintainer.
-
-### Asynchronous snapshot semantics
-
-`PendingBuild` will clone relevant deletion vectors synchronously when the
build is scheduled. The
-executor must not observe later mutations of the writer's deletion-vector
state. Deletions created
-after scheduling remain covered by query-time filtering.
-
-### Row-id stability
-
-The vector reader continues to read every physical row without applying a
filtering reader.
-`PkVectorAnnSegmentFile` skips excluded vectors but writes included ids as
-`fileOffset + physicalRowPosition`, preserving source metadata and positional
reads.
-
-### Fully deleted sources
-
-If every source row is excluded, publish a zero-row segment backed by an empty
marker file. This
-keeps the source covered without passing deleted row ids to an ANN writer or
retrying the same
-build indefinitely. Search recognizes the zero-row segment and returns an
empty result without
-opening the marker as an ANN payload.
-
-### Tests
-
-1. A deletion present before scheduling is absent from the newly built ANN
payload.
-2. Mutating the writer deletion vector after scheduling does not change the
pending build snapshot.
-3. A fully deleted source publishes a searchable zero-row segment.
-4. Existing query-time deletion-vector filtering remains unchanged.
-
-## Open Questions
-
-None for this scoped change.
-
-## Out of Scope
-
-- Paimon vindex changes.
-- Stale asynchronous build discard/retry policy.
-- Source-file stale-ratio scheduling.
-- Removing query-time deletion-vector filtering.
diff --git a/docs/designs/2026-07-15-primary-key-vector-parallel-search.md
b/docs/designs/2026-07-15-primary-key-vector-parallel-search.md
deleted file mode 100644
index 8135f63310..0000000000
--- a/docs/designs/2026-07-15-primary-key-vector-parallel-search.md
+++ /dev/null
@@ -1,75 +0,0 @@
-<!--
-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.
--->
-
-# Primary-Key Vector Parallel Search Design
-
-## Problem Statement
-
-`PrimaryKeyVectorRead` searches bucket splits serially, and
-`PrimaryKeyVectorBucketSearch` waits for each ANN segment before starting the
next one. A query
-over multiple buckets or segments therefore accumulates their local-search
latency.
-
-## Chosen Approach
-
-Use the existing global-index read executor and `global-index.thread-num` to
compose bucket,
-segment, and uncovered-file searches with `CompletableFuture`. Submit only
leaf searches to the
-executor and wait once after all searches have been started. This avoids
blocking an executor task
-while it waits for more work submitted to the same executor.
-
-## Design Details
-
-### Asynchronous Search
-
-- Add asynchronous ANN segment search while retaining synchronous wrappers for
compatibility.
-- Compose all segment futures within a bucket instead of joining each segment
immediately.
-- Compose all bucket futures in `PrimaryKeyVectorRead` and preserve the
existing deterministic
- global Top-K merge.
-- Submit FULL and DETAIL exact searches for uncovered files as independent
leaf tasks.
-
-### Resource and Error Handling
-
-- Attach index-reader closure directly to the future returned by the index
reader.
-- Close exact-search readers within their leaf tasks.
-- Do not submit additional work from completion callbacks.
-- Preserve the existing top-level `Failed to search primary-key vector index.`
error contract.
-
-### Compatibility
-
-- Keep JDK 8 syntax and existing synchronous search entry points.
-- Do not add a new option or executor.
-- Keep bucket preparation, deletion-vector loading, residual-filter
evaluation, reranking, and
- result ordering unchanged.
-
-### Verification
-
-- Prove multiple ANN segment searches can be in flight together.
-- Prove multiple bucket searches are composed concurrently and merge
deterministically.
-- Prove FULL and DETAIL exact fallback searches uncovered files concurrently.
-- Verify exceptional completion closes readers and is propagated.
-- Verify `global-index.thread-num=1` completes without deadlock.
-
-## Open Questions
-
-None.
-
-## Out of Scope
-
-- Fair scheduling across buckets when a semaphore-limited executor blocks
submissions.
-- Parallel bucket metadata, deletion-vector, residual-filter, or rerank reads.
-- Changes to `GlobalIndexReadThreadPool` or its executor wrappers.
