feat: Introduce cow-based Update Transaction#370
Open
zhanglei1949 wants to merge 3 commits into
Open
Conversation
zhanglei1949
commented
May 20, 2026
| virtual void resize(vid_t vnum) = 0; | ||
|
|
||
| virtual size_t capacity() const = 0; | ||
|
|
Member
Author
There was a problem hiding this comment.
close() is now inherited from Module.
zhanglei1949
commented
May 20, 2026
| } | ||
|
|
||
| void dump(const std::string& filename) override { | ||
| bool is_data_unmodified() const { |
Member
Author
There was a problem hiding this comment.
TODO: Could we avoid casting data_buffer to MMapContainer*?
zhanglei1949
force-pushed
the
zl/update-txn-cow
branch
2 times, most recently
from
6cbf42f to
754e8e3
Compare
zhanglei1949
force-pushed
the
zl/update-txn-cow
branch
4 times, most recently
from
1afcc6e to
a422a14
Compare
This was referenced May 26, 2026
zhanglei1949
force-pushed
the
zl/update-txn-cow
branch
3 times, most recently
from
a257c67 to
89cdad4
Compare
This was referenced May 28, 2026
zhanglei1949
force-pushed
the
zl/update-txn-cow
branch
from
89cdad4 to
77f481d
Compare
zhanglei1949
added a commit
that referenced
this pull request
Jun 1, 2026
…dary (#422) ## What do these changes do? Harden `PropertyGraph`'s label-id and edge-triplet validation so that bad caller-supplied IDs no longer become undefined behavior in Release builds, and unify the existing private check helpers so the string and `label_t` overloads behave consistently. - Add `label_t` overloads of the private helpers and make all four overloads `const`. Both `edge_triplet_check` overloads now delegate per-vertex validation to `vertex_label_check`, so they share logic and produce equally granular errors: ```cpp Status vertex_label_check(const std::string&) const; Status vertex_label_check(label_t) const; Status edge_triplet_check(const std::string&, const std::string&, const std::string&) const; Status edge_triplet_check(label_t, label_t, label_t) const; ``` - **Mutating APIs** (`BatchAddVertices`, `BatchDeleteVertices`, `DeleteVertex` ×2, `DeleteEdge`, `BatchDeleteEdges` ×2, `AddVertex`, `UpdateVertexProperty`, `UpdateEdgeProperty`) now `RETURN_IF_NOT_OK(vertex_label_check(...) / edge_triplet_check(...))` instead of `assert(...)` or relying on the callee to catch the bad ID. - **Non-Status accessors** (`get_vertex_table` ×2, `GetVertexPropertyColumn` ×2, `GetVertexSet`, `LidNum`, `VertexNum`, `IsValidLid`, `get_lid`, `GetOid`) now call `schema_.ensure_vertex_label_valid(...)` (throws `InvalidArgumentException`) — the only viable signal channel for non-`Status` returns. - `GetVertexPropertyColumn(label, col_id)` additionally bounds-checks `col_id` against the label's property count. - Unify error wording across overloads: `"Vertex label '<name>' is not valid"` / `"Vertex label id <N> is not valid"`; same `"Edge triplet <...> is not valid"` shape. Drop the `LOG(ERROR)` calls from the existing string-based helpers (caller decides). No behavior change for valid inputs. Verified no tests assert on the old error wording. ## Background These changes were originally bundled inside the `zl/update-txn-cow` branch (PR #370 split work) but have no semantic dependency on the workspace/checkpoint or COW refactor. Splitting them out keeps those downstream PRs focused and lets this hardening land independently against `main`. ## Related issue number Fixes #421 --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
refine module_store and module_desc refactor snapshot_meta/checkpoint, remove unncessary changes remove module store minor ref ref wal apply refine refine checkpoint manage fix ci minor refine fix minor refine refine module register macro fix format rm file_names.h minor refine minor refine supporing fast dump remove some cmments TBD: add dirty flag for MutableCSR, to avoid redumping the data minor fix remove method generate_uuid fix test revert is_dirty_ flag and compute md5 for mutable_csr's nbr_list before dump fix test check whether data_buffer is modified at column level refine remove unused field last_descriptor cherry-pick c964e58 fix issues reported by aone copilot two fold check remove some comment revert changes to indexer_ in VertexTable minor stash changes to cow fix tests fix format Committed-by: xiaolei.zl from Dev container minor changes fix test fixing fix version manager and tests fix test refactor: consolidate Schema vertex/edge label check API Unify the naming convention of label/property existence and validity checks on Schema (and the schema-check helpers it surfaces through PropertyGraph). Renames (snake_case `is_*` family): - contains_vertex_label / vertex_label_valid -> is_vertex_label_valid - contains_edge_label / edge_label_valid -> is_edge_label_valid - exist(...) -> is_edge_triplet_valid(...) - has_edge_label(...) -> has_edge_triplet(...) - IsVertexLabelSoftDeleted -> is_vertex_label_soft_deleted (then removed) - IsEdgeLabelSoftDeleted -> is_edge_label_soft_deleted (then removed) - IsVertexPropertySoftDeleted -> is_vertex_property_soft_deleted (then removed) - IsEdgePropertySoftDeleted -> is_edge_property_soft_deleted (then removed) Removed (zero production callers): - Schema::is_*_soft_deleted (8 overloads) + corresponding test cases - Schema::edge_triplet_valid (weak tomb-only check; merged into the stronger is_edge_triplet_valid) - PropertyGraph::edge_triplet_exist (dead code, no callers) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> fix format some renaming fixing append changes rm cache file refactor: extract read-side view layer over PropertyGraph storage Add VertexTableView / EdgeTableView / GraphView / TableView wrapping the existing storage components without back-pointers to the owning storage. View objects capture raw pointers to the heap-stable parts of each storage component (shared_ptr-managed indexer / v_ts / Csr / Table) at construction; no mutation flows through the underlying storage handle they were built from. Required two small additions on the underlying side: - RefColumnBase::set(idx, value), implemented on TypedRefColumn<T> and TypedRefColumn<string_view>, so TableView can write through a column ref. - VertexTable holds a stable shared_ptr<RefColumnBase> wrapping indexer_->get_keys(), so VertexTableView can borrow a raw pointer for primary-key-by-name lookups without owning it. Friend declarations added on EdgeTable / VertexTable / PropertyGraph so the sub-view ctors can capture private state without exposing it publicly. This PR introduces the view types only; no caller is migrated. ReadTransaction / InsertTransaction / StorageReadInterface migration is left for a follow-up, where the view becomes the stable observation point that the COW UpdateTransaction work in alibaba#86 needs. Fixes alibaba#400 fix test ref introduce views Remove all insert api minor format Committed-by: xiaolei.zl from Dev container format minor docs: update extensions index.md to reflect JSON built-in status and use PARQUET examples (alibaba#406) fix alibaba#405 Updates `doc/source/extensions/index.md` so that the extensions overview is consistent with `doc/source/extensions/load_json.md`, where JSON has been documented as a built-in feature since v0.1.2. 1. In the "Available Extensions" table, annotate the JSON row with `(built-in since v0.1.2)` so readers know JSON no longer requires `INSTALL` / `LOAD`. fix remove uneeded files remove unneeded views update tests remove unneeded fields for csrview refine interface explicitly call PrepareForXXX
zhanglei1949
force-pushed
the
zl/update-txn-cow
branch
from
13b7358 to
62b90ee
Compare
…ship tracking Introduce CowRef<T> wrapper that tracks exclusive ownership via a boolean flag, eliminating all shared_ptr::use_count() calls (fragile, deprecated in concurrent contexts). Replace pointer-arithmetic aliasing detection in MutableCsr with a per-vertex adjlist_owned_ bitset. Merge PrepareFor* COW-prep into PropertyGraph DML methods so callers cannot forget the protocol. Add ForkAsShared helpers to CsrBase, ColumnBase, and VertexTimestamp. Make EdgeTable/VertexTable PrepareFor* methods private. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fix #330
This PR contains two parts:
The Workspace & checkpoint implementation is based on #148