Stage 4A (RFC-004 §D1/§D5): graph.<id> Create — and its paired schema Create,
which the init carries — classify Applied and execute first in the run,
sequentially and sidecar-fenced: sidecar written before Omnigraph::init at
the derived root, rewritten with the post-init manifest pin, deleted only
after the final state CAS lands. Dependent queries and policies no longer
block on a graph create in the same plan — creates run first, so they apply
in the same run; a create failure demotes them to blocked
(dependency_not_applied) and stops further graph-moving work (loud partials),
with the sidecar left for the sweep to classify. Graphs with a kept recovery
sidecar (rows 5/6) classify Blocked/cluster_recovery_pending, and the sweep's
Drifted/Error statuses are never clobbered by a generic Blocked.
Schema source is re-read and digest-verified under the lock before the init
(the write_resource_payload TOCTOU posture). Plan previews the same
dispositions. e2e fallout updated: a fresh multi-graph config now converges
in one apply; a destroyed root is re-created as an EMPTY graph by the next
apply (declarative convergence — visible in plan, called out in docs); the
new cluster_e2e_declared_graph_created_by_apply pins the no-manual-init flow.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
RFC-004 §D2/§D3 for the graph_create kind. RecoverySidecar records intent
under __cluster/recoveries/{ulid}.json; the roll-forward-only sweep runs at
the start of apply/refresh/import under the state lock and classifies each
survivor by observation: root absent -> intent removed (row 1); outcome
already recorded -> retired (row 2); create completed but state stale ->
ledger rolled forward with a recovery_records audit entry (row 4); partial
root -> Error/graph_create_incomplete, kept, never auto-deleted (row 5);
unexpected schema -> Drifted/actual_applied_state_pending, kept (row 6).
Sweep mutations ride the command's existing CAS write; completed sidecars
are deleted only after that write lands. Read-only status/plan warn
(cluster_recovery_pending) without acting. The apply payload gate now counts
only payload-phase errors so kept-sidecar diagnostics don't abort the run
before their statuses persist.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Mechanical conversion ahead of Stage 4A graph create (which calls the async
Omnigraph::init from inside apply): the fn signature, the CLI dispatch arm,
and every test caller (#[test] -> #[tokio::test]). Zero behavior change; all
60 lib tests and 3 failpoint tests green before and after.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Engineering and docs roles both resolve to @aaltshuler; every path
(catch-all, crates/**, docs/**, repo-level docs) now requires their review.
CODEOWNERS and the doc tables regenerated from codeowners-roles.yml via
render-codeowners.py.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* docs(cluster): RFC-004 — graph & schema apply design (Phase 4)
The design the implementation spec's exit criteria require before
graph-moving cluster apply ships. Core positions:
- Cluster recovery is roll-forward-only: the engine's own sidecars make every
graph-level operation atomic within the graph, so the cluster never rolls a
graph back — its sidecars (__cluster/recoveries/{ulid}.json) classify and
record, converging the ledger to observable reality (axiom 5) or surfacing
a loud pending-repair condition. Eight-row decision matrix, every row
testable with the Stage 3B failpoint harness.
- Irreversible operations (graph delete, allow_data_loss schema apply)
consume digest-bound approval artifacts written by a new cluster approve
command and retired into state.approval_records (axiom 11). A stale
approval can never authorize a different change.
- cluster apply gains an actor, threaded to apply_schema_as so engine Cedar
enforcement and commit attribution work unchanged; the cluster adds no
policy engine of its own.
- Deterministic ordering (creates -> schema applies -> catalog -> deletes),
per-resource apply groups, cross-graph atomicity explicitly not promised.
- Staged 4A graph create / 4B schema apply / 4C graph delete, each gated on
per-matrix-row failpoint tests.
Answers exit criteria 2 and 4 fully, 1/5/6 partially; 3/7/8/9 deferred to
their phases (coverage table in the RFC). Linked from the dev index and the
implementation spec's Phase 4 section.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* docs(cluster): RFC-004 review fixes — graph_delete sweep rows, state_cas_base contract
Two greptile findings: (1) D3 row 2 could not be evaluated for graph_delete
(no manifest to version-check after prefix removal) and 'root absent, state
already tombstoned' fell into the stale row — split into rows 7 (delete's
analog of row 2) and 7b (the roll-forward), with expected_manifest_version
documented as always null for the delete kind. (2) state_cas_base is now
explicitly audit/diagnostics-only — the sweep never consults it; independent
state mutations are handled by the ordinary CAS like any concurrent write.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
ScopedFailPoint::with_callback gives cfg_callback the same Drop-based cleanup
as cfg actions; a panic while the point is active no longer leaks the callback
into the process-global registry where it would fire under later tests
(greptile review, PR #167).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The failpoints feature added fail = { workspace = true, optional = true } to
the crate manifest; the lockfile edge belongs with it (--locked CI gate).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The apply-side coverage the implementation spec's hard gate requires before
Phase 4 graph-moving apply:
- crash after the payload phase: state.json byte-identical, blobs inert on
disk, lock released, no phantom statuses, nothing acknowledged; a plain
re-run repairs via skip-if-exists blob reuse.
- CAS race: a cfg_callback rewrites state.json at the exact read->write
window (the state.lock:false concurrent-writer scenario); apply surfaces
state_cas_mismatch, acknowledges nothing, reports the persisted status
snapshot, leaves the concurrent writer's state on disk; a re-run converges.
CI's failpoints step now runs both the engine and cluster suites.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Optional failpoints feature (dep:fail + fail/failpoints, deliberately NOT
enabling omnigraph/failpoints), a maybe_fail/ScopedFailPoint module returning
Diagnostic-typed injected errors, and two call sites in apply_config_dir:
cluster_apply.after_payload_phase (the crash point: blobs on disk, state
untouched) and cluster_apply.before_state_write (routes through the
persisted-statuses revert contract; a cfg_callback here can mutate state.json
to make the CAS check fail organically). Feature off compiles to Ok(()) —
zero behavior change. Tests live in a separate integration binary because the
fail registry is process-global. Also refresh the crate description (stale
'read-only' since Stage 3A).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
status warns read-only -> refresh persists drift and drops the digest ->
apply republishes the blob -> status clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Closes the Stage 3A product gap where a deleted or corrupted blob under
__cluster/resources/ went unnoticed forever (status reported converged and
apply could not repair it because the digests matched).
verify_catalog_payloads checks every query/policy digest in state against its
content-addressed blob (existence + full sha256 re-hash; graph/schema/unknown
addresses have no payloads and are skipped). status reports findings read-only
(warnings catalog_payload_missing/_mismatch; error catalog_payload_read_error
— an unverifiable catalog must not report healthy). refresh closes the
self-heal loop: missing/mismatched blobs mark the resource drifted and remove
its digest from state so the next plan proposes a create and the next apply
republishes; unreadable blobs keep the digest (no spurious republish), mark
error, and exit non-zero. Verification runs before graph observation so the
recomputed graph composite already excludes removed query digests.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Four lifecycle compositions over the spawned binary that pin spec claims no
single-command test proves:
- Lost ledger: delete state.json -> re-import from the live graph -> re-apply
converges onto the same content-addressed blobs (axiom 5's reconstructable-
state resilience edge, end to end).
- Out-of-band schema apply (the Sarah/Bob violation): refresh marks
graph/schema Drifted with schema_mismatch, status and plan surface it, and
cluster apply refuses to silently correct it — state keeps the LIVE schema
digest (drift correction is gated, axiom 8).
- Destroyed graph root: refresh records graph_missing drift and drops
graph/schema digests while preserving query/policy; plan proposes deferred
creates only; apply moves nothing and the catalog stays intact.
- Two graphs (one live, one not yet created) + a graph-spanning policy + a
cluster-scoped policy: a single apply yields all four dispositions at once
(applied/derived/deferred/blocked, deterministically ordered), then the
second graph appears, refresh observes it, and apply converges.
Helpers: init_named_cluster_graph generalizes init_cluster_derived_graph;
write_multi_graph_cluster_fixture builds the two-graph config.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Encode the omnigraph.yaml ↔ cluster.yaml coexistence rules that were implicit
across the specs:
- cluster-axioms.md: new axiom 15 — every fact has exactly one owner at a time;
coexistence is a mode switch, never a merge; omnigraph.yaml's job description
shrinks to the permanent per-operator layer. Added review-tension bullet.
- cluster-config-specs.md: "Migration model" subsection (three coexistence
windows: no-conflict, Phase-5 mode switch, bridges-with-sunsets) and a
"per-operator layer" completeness table (connection, credential reference,
active context, ergonomics, personal aliases) with its global-config-dir
destination per the RFC-002 direction.
- cluster-config-implementation-spec.md: Compatibility Stance #7–#9 (single
ownership, shrinking role, bridges carry sunsets); Phase 5 boot is an
exclusive XOR mode switch; fixed the duplicated recoveries/recovery dirs in
the Phase-1 storage layout.
- docs/user/cluster-config.md: "Relationship to omnigraph.yaml" section in
current-reality terms (cluster catalog is inspectable, not live).
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Two review findings (greptile, PR #165):
- ApplyOutput.resource_statuses on a failed state write now carries the
pre-apply on-disk snapshot instead of the in-memory mutations that were
never persisted, so automation reading the field independently of `ok`
cannot see phantom applied/blocked statuses. Regression test forces the
state write to fail via a read-only __cluster dir (unix-only, skips when
permissions are not enforced).
- Human-mode `cluster apply` prints the classified changes list on failure
too, so an operator debugging a partial apply without --json sees what was
attempted.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* MR-854: convert engine call sites to &dyn TableStorage; demote legacy methods
Phase 1b: every db.table_store.X(...) call site converts to
db.storage().X(...), reaching the storage layer through the sealed
TableStorage trait (returns &dyn TableStorage). Opaque SnapshotHandle
and StagedHandle replace bare lance::Dataset and Transaction in the
threaded values.
Phase 9: the inherent inline-commit methods on TableStore
(append_batch, merge_insert_batch{,es}, overwrite_batch,
create_btree_index, create_inverted_index) demote from pub to
pub(crate). Their only remaining direct users are table_store.rs
itself and the bulk loader's LoadMode::{Append, Overwrite, Merge}
concurrent fast-paths in loader::write_batch_to_dataset (no
two-phase shape in Lance 4.0.0 — closes after lance#6658 and #6666).
Docs:
- invariants.md \u00a7VI.23: drop "at the writer-trait surface"
qualifier; staged primitives are now the only engine surface.
- runs.md: residual matrix shrinks to delete_where and
create_vector_index (the two upstream-blocked residuals).
- forbidden_apis.rs: replace transitional language with the
current allow-list shape (table_store.rs + loader concurrent
fast-path only).
Files touched:
- changes/mod.rs, db/omnigraph.rs (+export/optimize/schema_apply/
table_ops.rs), exec/{merge,mod,mutation,staging}.rs,
loader/mod.rs, storage_layer.rs, table_store.rs,
tests/forbidden_apis.rs, docs/{invariants,runs}.md.
Co-Authored-By: Ragnor Comerford <ragnor.comerford@gmail.com>
* MR-854: replace test-only inline-commit append callers with local Lance helpers
After demoting TableStore::append_batch from pub to pub(crate), the
integration tests in tests/recovery.rs and tests/staged_writes.rs
that previously called store.append_batch(...) directly to simulate
HEAD-ahead-of-manifest drift can no longer access the inherent
method. Replace those calls with small in-test helpers that do a raw
Dataset::append (the same body the inherent method runs).
- tests/helpers/mod.rs gains lance_append_inline (shared helper).
- tests/staged_writes.rs gets a file-local lance_append_inline_local
(staged_writes.rs does not import helpers::).
- tests/recovery.rs drops the unused TableStore import in the one
function whose store binding became unused after the conversion.
Co-Authored-By: Ragnor Comerford <ragnor.comerford@gmail.com>
* MR-854: retrigger CI for flaky Test Workspace job
Co-Authored-By: Ragnor Comerford <ragnor.comerford@gmail.com>
* MR-854: convert remaining table_store call sites in export.rs / read_blob
Two leftover `self.table_store.X` / `db.table_store.X` call sites were
missed in the initial sweep — flagged by Devin Review on PR #86. Both
now go through the trait surface:
- `entity_from_snapshot` (db/omnigraph/export.rs): switch from
`db.table_store.open_snapshot_table` + `db.table_store.scan` to
`db.storage().open_snapshot_at_table` + `db.storage().scan`.
- `read_blob` (db/omnigraph.rs): replace
`snapshot.open(table_key)` + `self.table_store.first_row_id_for_filter`
with `self.storage().open_snapshot_at_table` +
`self.storage().first_row_id_for_filter`. The follow-up
`take_blobs` call still needs an `Arc<Dataset>` (it's a Lance blob
accessor not surfaced through the trait), so we hand off via
`SnapshotHandle::into_arc()` with a comment.
After this commit, no engine code outside `table_store.rs` reaches the
inherent `TableStore` API — the docs/runs.md and docs/invariants.md
claim is now uniformly true.
Co-Authored-By: Ragnor Comerford <ragnor.comerford@gmail.com>
* MR-854: post-rebase doc fixes (Lance 6.0.1, MR-A framing, into_dataset note)
Reviewer feedback on the rebased PR:
* docs/dev/writes.md residuals matrix: drop demoted methods from the trait-surface table (now `pub(crate)`); keep only the two genuine trait-surface residuals (`delete_where`, `create_vector_index`); reframe under MR-A (Lance v7.x bump) per docs/dev/lance.md.
* tests/forbidden_apis.rs: update transitional allow-list header to (a) drop the truncate_table mislabel (truncate_table is a Lance Dataset method, not a TableStore method — overwrite_batch's internal call), (b) reframe trait-surface residuals under MR-A / Lance #6666.
* crates/omnigraph/src/storage_layer.rs::SnapshotHandle::{into_arc, into_dataset}: add single-ref invariant doc — both consume Arc via try_unwrap-or-clone; sibling SnapshotHandle clones across an await point force a deep Dataset clone.
* Replace lance-4.0.0 version refs with lance-6.0.1 in active source/test/dev-doc comments (storage_layer.rs, table_store.rs, table_ops.rs, schema_apply.rs, merge.rs, recovery.rs, staged_writes.rs, consistency.rs, docs/dev/execution.md, docs/user/query-language.md). Historical refs in docs/releases/v0.4.1.md and the canonical "Lance 4.0.0 → 6.0.1 migration" line in docs/dev/lance.md left intact.
No engine code changes.
* MR-854: update docs/dev/invariants.md Storage trait row + gap entry
Reviewer feedback: the docs reorg landed; the invariant row now lives in
docs/dev/invariants.md with stable headings (no more numbered §VI.23).
Update two pieces to reflect MR-854 completion:
* Status table 'Storage trait' row: was 'full call-site migration ... incomplete';
now 'engine call sites all route through db.storage() (MR-854); inline-commit
inherent methods are pub(crate)-demoted; capability/stat surfaces are roadmap'.
* 'Known Gaps' 'Storage abstraction' entry: was 'older inherent TableStore call
sites and inline residuals remain'; now names the closed scope (MR-854 — call
sites migrated, methods demoted, loader fast-paths) and the remaining
trait-surface residuals under MR-A (Lance v7.x bump) and Lance #6666.
Cross-links to docs/dev/lance.md and docs/dev/writes.md so the framing stays
co-located with the canonical Lance surface tracking.
* MR-854: remove dead inline-commit methods from the storage surface
The loader concurrent fast-path (write_batch_to_dataset) is only reached
for LoadMode::Overwrite — Append/Merge route through MutationStaging — so
its Append/Merge arms were unreachable. Collapse it to overwrite-only and
drop the now-unused mode params, which removes the only callers of:
- TableStorage::append_batch + TableStorage::merge_insert_batches (trait)
- TableStore::merge_insert_batch + merge_insert_batches (inherent)
create_btree_index / create_inverted_index had zero callers anywhere
(scalar index builds use the stage_* primitives). Remove both from the
trait and the inherent impl.
Inherent append_batch stays pub(crate): overwrite_batch and recovery
tests use it. Migrate the one trait-append_batch test caller
(seed_person_row) to stage_append + commit_staged. The merge_insert
FirstSeen-workaround rationale moves from the deleted merge_insert_batch
into stage_merge_insert (now the sole merge path). No behavior change.
Also corrects the inaccurate loader residual comment (the prior text
blamed Lance #6658/#6666, which are the delete and vector-index issues,
for keeping overwrite inline; a stage_overwrite primitive already exists
and schema_apply uses it).
* MR-854: seal db.storage() to staged-only; move residuals to InlineCommitResidual
Split the three remaining inline-commit writes (overwrite_batch,
delete_where, create_vector_index) off the TableStorage trait onto a new
sealed InlineCommitResidual trait, reachable only via the explicit
Omnigraph::storage_inline_residual() accessor. db.storage() now exposes
only staged primitives + reads, so engine code cannot couple a write
with a Lance HEAD advance through the default surface — MR-793 acceptance
§1 ("no public method commits as a side effect of writing") now holds by
construction, not by review + naming.
Call sites moved to storage_inline_residual(): loader overwrite
fast-path, the three mutation delete_where paths, the branch-merge
delete, and the vector-index build. Impl bodies are unchanged (same
delegation to the pub(crate) inherent methods); this is a pure surface
reshape with no behavior change.
The residual trait holds two genuinely upstream-blocked methods
(delete_where -> Lance #6658/v7.x, create_vector_index -> Lance #6666)
plus overwrite_batch, kept for the loader's cross-table bulk-overwrite
concurrency until its staged migration lands (tracked follow-up).
* MR-854 docs: describe the staged-only seal; fix stale Lance index URLs
- writes.md / invariants.md / AGENTS.md: the inline-commit residuals now
live on InlineCommitResidual behind db.storage_inline_residual(), so
acceptance §1 holds by construction rather than 'option (b)' per-method
enumeration. Drop the inaccurate 'until Lance exposes
Operation::Overwrite { fragments }' claim (that op exists; stage_overwrite
already builds it) and reframe overwrite_batch as a removable legacy
residual gated on the loader's bulk-overwrite concurrency.
- forbidden_apis.rs: rewrite the allow-list doc for the split surface.
- lance.md: the index spec pages moved from /format/table/index/ to
/format/index/ in Lance 6.x (the old paths 404). Fix all 13 URLs.
* MR-854: fix stale lance-4.0.0 comment refs flagged in review
Addresses greptile (exec/merge.rs) and aaltshuler's stale-version blocker:
update lance-4.0.0 -> 6.0.1 in the comment/doc refs within this PR's
footprint (exec/merge.rs, exec/mutation.rs, docs/dev/writes.md). Also
corrects exec/merge.rs to cite lance#6666 (not #6658) for
build_index_metadata_from_segments — that is the vector-index segment-commit
API; #6658 is the two-phase delete. (Pre-existing 4.0.0 refs in untouched
files like architecture.md/storage.md are main's incomplete migration
cleanup, left out of scope.)
* fix(storage): stage loader overwrites
* fix(storage): stage empty schema rewrites
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: Ragnor Comerford <ragnor.comerford@gmail.com>
Co-authored-by: Ragnor Comerford <hello@ragnor.co>
Three composition tests over the spawned binary against a real derived graph:
- import -> plan (dispositions) -> apply -> status -> refresh -> plan-empty,
then a query edit round-trip. Pins that refresh and apply recompute the
graph composite digest identically — divergence would silently re-open
the plan forever and no single-command test would catch it.
- The Stage 3A operator workflow across the control/data-plane boundary:
cluster apply defers a schema change, omnigraph schema apply executes it,
cluster refresh observes it, the next cluster apply re-converges.
- Held lock refuses apply, force-unlock clears it, retried apply converges.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Terraform-style: apply executes directly (cluster plan is the preview, now
annotated with apply dispositions). Human output prints per-change
dispositions, convergence, and the catalog-only caveat; --json emits the full
ApplyOutput. Exit is non-zero only on errors — deferred/blocked changes are
warnings with converged: false as the automation signal.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
apply_config_dir executes the query/policy subset of the plan: payloads are
written content-addressed under __cluster/resources/{query,policy}/... before
the state CAS (state is the publish point; orphaned blobs from a failed CAS
are inert and re-apply is the repair), then state.json is CAS-updated with
applied digests, Applied/Blocked statuses, and a revision bump. Graph/schema
changes are never executed here: schema content and graph lifecycle defer to
a later phase with loud warnings, while graph.<id> composite-digest updates
whose schema component is unchanged converge automatically via recomputation
from state's own components (without which apply could never converge).
Idempotent re-apply leaves state bytes and revision untouched.
PlanChange gains optional disposition/reason fields, populated by the same
classifier in cluster plan, so plan is an honest preview of what apply will
execute, derive, defer, or block.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The v0.6.2 notes omitted the MR-770 `__run__` cleanup entirely, and the v0.6.1
notes wrongly claimed it shipped in v0.6.1. The code (the `migrate_v2_to_v3`
`__manifest` sweep + `is_internal_run_branch`/`run_registry.rs` removal) first
appears at the v0.6.2 tag via #132 and is absent at v0.6.1.
- v0.6.2: add the MR-770 highlight, correct the manifest-stamp note to describe
the v2→v3 auto-migration on first read-write open (with the read-only caveat),
and mention the cleanup in the intro.
- v0.6.1: replace the two over-claiming `__run__` lines with corrections that
point to v0.6.2.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(unique): collision-free tuple key shared by intake and merge, loud on un-keyable types
Hardening on top of #133. That PR introduced a shared
`loader::composite_unique_key(parts)` joining per-column scalars with U+001F
and routed both intake and branch-merge through it, closing the original
'|' vs U+001F separator drift. This takes the shared keying the rest of the
way to correct-by-design:
- Collision-free by construction: the key is now the tuple of per-column
scalar strings (Vec<String>) keyed directly, no separator, so no data value
(not even a literal U+001F) can forge a collision.
- One scalar converter across both paths: intake used an explicit type-match,
merge used Arrow's array_value_to_string. Both now derive the key through
composite_unique_key(group_columns, row), so they can't drift on conversion.
- Loud on un-keyable types: the scalar converter returned None for any Arrow
type it didn't recognize, and the caller treated None as null-exempt, so a
@unique on a column type it couldn't reduce (list, blob) was silently
un-enforced. It now returns Err, surfacing the constraint it can't enforce
instead of weakening it in silence.
Tests:
- consistency::composite_unique_key_is_consistent_across_intake_and_merge pins
that intake and merge key the tuple identically (load-on-branch then merge
of values containing '|').
- loader unit tests pin tuple keying + null exemption and the loud error on an
un-keyable (binary) column.
Docs: invariants truth-matrix updated; stale loader/mod.rs line pointers fixed.
Scope unchanged: intra-batch / merge-candidate-set only; cross-version
uniqueness against committed rows stays a documented gap.
* fix(unique): cover all string encodings; make format_tuple private (PR #160 review)
Addresses two Greptile P2 comments on PR #160:
- unique_key_scalar handled only StringArray (Utf8). The loud-on-unknown-type
behavior turned any legal string column that read back as LargeUtf8 or
Utf8View into a hard write failure (the old code silently returned None). Add
LargeStringArray and StringViewArray arms so a legal string column is keyable
in every physical Arrow encoding; the Err path now fires only for a genuinely
un-keyable logical type (list/blob/vector), never a legal value in an
unenumerated encoding.
- format_tuple was pub(crate) but only used within loader/mod.rs; make it a
private fn (matches the old format_unique_columns it replaced, minimal
exposed surface).
New unit test unique_key_scalar_handles_all_string_encodings pins that Utf8 /
LargeUtf8 / Utf8View all render rather than error.
* perf(engine): route Expand node hydration through the id BTREE via structured filter
hydrate_nodes built an `id IN (...)` SQL string applied via Scanner::filter,
which DataFusion evaluates with InListEval (O(N×M)) rather than using the id
BTREE scalar index — measured at 72× the indexed cost on a 100k-node hop
(MR-376). Build the id IN-list as a structured DataFusion Expr, AND it with
the pushable destination filters, and apply via Scanner::filter_expr (the same
path execute_node_scan already uses); Lance then compiles it to
scalar-index-search -> take.
Destination-filter pushability is now decided by ir_filter_to_expr (structured)
instead of ir_filter_to_sql, so list-contains (array_has) pushes down too.
Removes the now-dead string-filter helpers build_lance_filter, ir_filter_to_sql,
and ir_expr_to_sql; literal_to_sql stays (still used by the mutation delete path).
* feat(engine): add TableStore::scan_edges_by_endpoint for indexed neighbor lookup
Static helper returning edge rows that match a set of endpoint keys on src/dst,
projected to [key_col, opposite_col], via a structured `key_col IN (keys)`
filter_expr. Lance routes it through the persisted BTREE on the endpoint column
(index-search -> take), so cost scales with the frontier size rather than |E|.
Unused until execute_expand's indexed mode lands; isolated in its own commit so
the storage-layer primitive is reviewable on its own.
* feat(engine): add BTREE-indexed Expand traversal path
Split execute_expand into a dispatcher over execute_expand_csr (the existing
in-memory CSR BFS, unchanged) and a new execute_expand_indexed that serves each
hop by batching the frontier into one scan_edges_by_endpoint call against the
persisted src/dst BTREE (index-search -> take), then fans out per source row.
Both share expand_hydrate_and_align — the destination hydration + alignment +
hconcat + in-memory non-pushable filters — which now aligns by string id (a
HashMap) instead of a dense row-id vec, so one tail serves both modes.
Mode selection is OMNIGRAPH_TRAVERSAL_MODE for now (default csr); the
frontier-size auto policy and lazy CSR build follow. AntiJoin stays on CSR.
tests/traversal_indexed.rs (its own #[serial] binary, so env writes never race a
reader) asserts the indexed path matches CSR for one-hop, multi-hop, cross-type,
and no-match cases, and that a freshly-appended unindexed edge is still found
(partial index coverage — fast_search=false unindexed-fragment scan).
* feat(engine): frontier-size Expand dispatcher + lazy CSR build
Replace the env-only mode switch with an auto policy: Expand uses the
BTREE-indexed path when the source frontier is small and the hop count bounded
(OMNIGRAPH_EXPAND_INDEXED_MAX_FRONTIER=1024, OMNIGRAPH_EXPAND_INDEXED_MAX_HOPS=6),
else the in-memory CSR. OMNIGRAPH_TRAVERSAL_MODE=indexed|csr still forces a mode.
Make the CSR index lazy: thread a GraphIndexHandle (memoizing OnceCell over a
Cached/Direct/None builder) through execute_query/execute_pipeline/
execute_rrf_query/execute_anti_join instead of a pre-built Option<&GraphIndex>.
A query served entirely by the indexed path with no AntiJoin never pays the
O(|E|) CSR build — the perf win of Tier 3. AntiJoin still realizes the index
(its negation uses CSR has_neighbors).
Net effect: selective traversals (the common case) skip the whole-graph CSR
build and resolve neighbors from the persisted, incrementally-maintained
src/dst BTREE. Existing traversal/aggregation/end_to_end/search suites now run
the indexed path by default and stay green.
Docs: constants.md (new env knobs), query-language.md (Expand dual path),
indexes.md (graph index is lazy + the indexed alternative).
* test(engine): bench indexed vs CSR selective traversal
Add a selective single-source knows{1,2} comparison to bench_expand: per growing
|E|, time the cold query in csr vs indexed mode (fresh db each, so CSR pays its
O(|E|) build) and assert both modes return identical rows — a guard against the
scalar-index physical_rows silent fallback dropping unindexed-fragment rows. The
existing dense hop1/2/3 latency bench is unchanged.
* feat(engine): surface silent scalar-index fallback in indexed traversal (C6)
Add TableStore::key_column_index_coverage — a metadata-only check (no IO) of
whether a `key_col IN (...)` scan will be served by the persisted BTREE or
silently fall back to a full filtered scan, mirroring Lance's own decision:
no BTREE on the column, or any fragment missing physical_rows (which disables
scalar indices for the whole scan, lance dataset/scanner.rs create_filter_plan).
execute_expand_indexed calls it once per traversal and tracing::warn!s on
Degraded, so the perf cliff is observable instead of hidden behind a bench oracle.
Detection-only: results are correct either way (the scan returns all rows). Closes
the "no silent failures" gap the traversal best-practice audit flagged as the top
deviation, and adds an IndexCoverage value a future cost-based planner can consume.
* perf(engine): dense-id BFS on the indexed traversal path (C3)
execute_expand_indexed ran its per-source BFS in string space
(Vec<HashSet<String>>, HashMap<String,Vec<String>>, ~4 String clones per neighbor
occurrence). Intern node ids to u32 once via a per-traversal TypeIndex (no
GraphIndex/CSR build — laziness preserved) and run visited/seen/frontier/
neighbor-map in dense u32 space, mirroring the CSR path; de-intern only for the
per-hop IN-list and the emitted dst ids handed to the hydrate+align tail.
Behavior-preserving — the traversal_indexed CSR-vs-indexed equivalence tests are
the guard (results are identical, the key type just changes String -> u32).
* refactor(engine): thread the opened edge dataset into indexed Expand
Hoist the edge-dataset open and the C6 index-coverage warning out of
execute_expand_indexed into execute_expand, threading the opened dataset in
as a parameter so it is opened exactly once. Extract the endpoint-column
mapping (endpoint_columns) and the coverage warning (warn_on_degraded_coverage)
as helpers.
Behavior-preserving: same dataset, same warning, same dispatch decision. This
only relocates the open so the upcoming cost-based chooser can consult index
coverage before dispatch without opening the dataset twice.
* feat(engine): cost-based Expand dispatch chooser (C5)
Replace the fixed frontier<=1024 && hops<=6 dispatch threshold with a pure,
IO-free cost model. choose_expand_mode compares the indexed path's
frontier-relative work (hops * frontier * fanout, or hops * |E| when BTREE
coverage is degraded) against the cost of building the whole-graph CSR
(BUILD_FACTOR * |E|), from cheap manifest row counts. Under good coverage this
reduces to a selectivity ratio independent of |E|, preserving the flat-in-|E|
indexed win for selective traversals while routing dense / deep / high-fanout
or degraded-and-expensive traversals to CSR.
execute_expand decides cardinality-first and only opens the edge dataset to
confirm coverage when it leans indexed (no open on a clearly-CSR traversal).
The two env knobs become hard ceilings layered on the model; the
OMNIGRAPH_TRAVERSAL_MODE override still forces a path; the chosen mode is
traced. Results are unchanged across modes — only the path differs.
Adds inline crossover unit tests and extends the traversal_indexed both_modes
harness with an auto pass asserting the chooser is result-preserving across
every traversal shape. Documents the new flag semantics in
docs/user/{constants,query-language}.md.
* test(engine): pin Lance scalar-index coverage + system-column/deletion-metadata surface
Add three Lance surface guards de-risking a future persisted-adjacency cache:
- a compile-only guard pinning the fragment physical_rows + index-detail
surface that key_column_index_coverage mirrors (the C6 fallback);
- a runtime probe confirming a scalar BTREE on the system column
_row_last_updated_at_version is not buildable via the normal create-index
path (the column is not in the user schema), so a version-column range delta
is not viable as drafted;
- a runtime probe confirming per-fragment deletion metadata
(deletion_file.num_deleted_rows) is available as cheap O(fragments) metadata,
the primitive a fragment-coverage delete model would rely on.
The probes turn the two largest substrate assumptions into green/red CI facts
before any cache work begins.
* test(engine): regression for cross-type id-collision in indexed traversal
A node id is unique only within a type, so a Person and a Company can share an
id string. A variable-length traversal over a cross-type edge (WorksAt) must
structurally stop after one hop. This test builds a graph where 'shared' is both
a Person and a Company id and asserts worksAt{1,2} returns only the one-hop
company. It fails today: the indexed path's single string interner de-interns
the hop-1 Company id back to the colliding Person id and runs a hop-2 scan that
matches that Person's edges, emitting a spurious second-hop company (indexed
["other","shared"] vs csr ["shared"]).
* fix(engine): structurally cap cross-type Expand at one hop
A cross-type edge cannot chain (e.g. a Company is not a WorksAt source), so a
variable-length traversal over one is structurally single-hop. Both traversal
paths now enforce this by capping max hops at 1 when from_type != to_type,
instead of relying on the hop-2 scan returning empty.
That reliance was a correctness hole on the indexed path: it interns every
endpoint string into one dense id space, so a cross-type id-string collision (a
Person and a Company sharing an id) let hop 2 de-intern a destination id back to
the colliding source-type id and match its edges, emitting rows the CSR path
never produces. With the cap the cross-type second-hop scan never runs, so the
shared interner can no longer alias across types. Turns the regression test
green (indexed == csr == ["shared"]).
* perf(engine): set-oriented filtered anti-join, remove per-row dispatch
execute_anti_join's filtered slow path sliced the outer batch to one row at a
time and re-ran the inner pipeline per row, so each 1-row inner Expand dispatched
to the indexed path — one Lance scan per outer row, while the CSR realized up
front sat unused.
Replace it with a set-oriented anti-semi-join: tag each outer row with a
synthetic index column, run the inner pipeline once over the whole frontier (the
tag survives Expand's hconcat and Filter's row-drop), then exclude outer rows
whose tag survived. The inner Expand now runs as a single set-at-a-time traversal
over the full frontier; config is read once per operator, not per row (the env
nit is mooted). A produced-but-untagged inner batch fails loudly rather than
silently keeping every row. Results are unchanged (the predicated-negation tests
exercise the path over a multi-row outer with dst-filters).
* test(engine): drop flaky wall-clock budget from the merge truth table
The 30s wall-clock assertion in merge_pair_truth_table flakes under parallel
test load: it tripped at ~31s in the full --test-threads=4 gate while passing at
~20s in isolation. A fixed time budget in a correctness test depends on machine
and parallelism, not correctness; elapsed is still logged for visibility, and a
real merge-perf regression belongs in a bench. The cell-count correctness
assertions (81 / 36 / 45) are unchanged.
* fix(engine): total deterministic ORDER via entity-key tie-break + NULL contract
apply_ordering used an unstable lexsort with no tie-break, so rows with equal
user-sort keys came out in a run-dependent order (the input order depends on
scan parallelism / upstream hashing) — making ORDER ... LIMIT non-deterministic,
a latent deny-list violation (no nondeterministic result ordering).
Append the bound entities' key columns (<var>.id, unique per row) in canonical
name-sorted order as ascending tie-breaks, giving a total, reproducible order
(and a deterministic top-N when ties straddle the LIMIT cutoff). NULL placement
(nulls_first = !descending) is unchanged and now documented as the contract.
New tests/ordering.rs locks descending, multi-key precedence, the deterministic
key tie-break (data loaded in a different order than the expected output, so it
proves the tie sorts by key not by load order), and NULL placement under ASC/DESC.
docs/user/query-language.md documents the total-order + NULL contract.
* test(engine): property-based query-correctness invariants over generated graphs
Adds a proptest harness (new dev-dep) that generates small graphs whose Person
and Company keys are drawn from a shared 5-key alphabet, so cross-type id
collisions, cycles, and self-loops arise by search rather than from one
hand-built fixture. Three invariants:
- prop_expand_indexed_eq_csr: csr == indexed == auto over knows{1,3} (same-type,
cycles) and worksAt{1,2} (cross-type, collision-prone) from every start.
- prop_results_subset_of_existing_nodes: no phantom rows (catches over-emission
even if both modes are wrong identically).
- prop_antijoin_partitions_persons: not{worksAt} and its complement are disjoint
and cover all persons.
Verified the guard bites: neutering the cross-type hop cap makes
prop_expand_indexed_eq_csr fail and proptest shrinks it to persons["c","e"] /
companies["b","c"] — the cross-type collision class the hand-built fixture
only sampled once. Tests are sync + #[serial] (per-case runtime; the mode test
writes OMNIGRAPH_TRAVERSAL_MODE).
* test(engine): cover cycle/self-loop termination + nested anti-join (C5 edge cases)
- variable_hops_terminate_and_dedup_on_cycle: a 3-cycle a->b->c->a traversed with
knows{1,5} (ceiling above the cycle length) terminates and emits each node once
(the c->a back-edge hits the seeded source); both_modes confirms indexed == csr.
Uses a bounded range deliberately — unbounded {1,} is a typecheck error, not a
runtime path.
- variable_hops_handle_self_loop: a->a self-loop does not loop forever and does
not re-emit the seeded source.
- nested_anti_join_double_negation: not { worksAt; not { name = Acme } } recurses
through execute_pipeline, yielding [Alice,Charlie,Diana] (people with no non-Acme
employer) — distinct from plain unemployed [Charlie,Diana].
* test(engine): execution goldens for typed-literal filters (C4 gap #4)
New literal_filters.rs covers filtering by F64/F32/Bool/Date/DateTime LITERALS
across both arms: standalone comparisons ($m.score > 1.5, $m.ratio <= 0.25,
$m.active = true, $m.born >= date(...), $m.seen < datetime(...)) exercise the
in-memory comparison path, and inline bindings (Metric { active: true },
Metric { score: 3.0 }) exercise Lance filter_expr pushdown. Seeds partition each
predicate so a dropped/miscast filter returns all rows. (Param-bound scalars and
list-column contains are covered elsewhere.)
* test(engine): full rank-order goldens for nearest + bm25 (gap #2)
Existing search tests stopped at top-1 (nearest) or non-empty (bm25), so a
regression corrupting ranks 2..k or reversing the sort direction passed CI
silently. Pin the FULL ordered slug list: nearest([0.1,0.2,0.3,0.4]) ->
[ml-intro, nlp-guide, rl-intro] (ml-intro exact at dist 0, rest by ascending
L2); bm25(Learning) -> [rl-intro, ml-intro, dl-basics] (descending score).
nearest/bm25 skip apply_ordering (is_search_ordered) and return Lance native
order, so result_slugs row order == rank order; values resolved by running and
confirmed stable across runs.
* test(engine): search fuzzy/match_text characterization + RRF non-default pairings
- match_text_matches_exact_set_excludes_unrelated: match_text(body,'neural') ==
[dl-basics] exactly (not just contains).
- fuzzy_does_not_match_under_default_tokenizer: characterizes that fuzzy() is
inert with the default tokenizer here (search/match_text work, fuzzy returns
nothing); turns red — to be promoted to a real golden — if fuzzy starts matching.
- rrf_fuses_two_fts_fields / rrf_fuses_two_vector_queries: RRF fuses arms other
than the default nearest+bm25 (bm25 title+body; two vector queries), proving
primary_var resolves and fusion runs. New fixtures/search.gq queries +
two_vector_params helper. Orders resolved by running, confirmed stable.
* test(engine): anti-join fast-vs-slow path equivalence harness
anti_join_fast_and_slow_paths_agree: the CSR has_neighbors fast path
(not { $p worksAt $_ }) and the set-oriented inner-pipeline replay (same
negation forced slow by an always-true $c.name != "" dst filter) must produce
the same result ([Charlie, Diana]). Closes the second real engine fork explicitly.
* test(engine): regression for nested slow-path anti-join tag collision
A nested not { ... not { ... } } where both levels hit the set-oriented slow
path collides on the fixed __antijoin_outer_row correlation column: the inner
call appends a duplicate, and column_by_name reads the OUTER tag. Fan-out (p1
works at two companies) makes inner row indices diverge from outer tags, so the
bug returns the wrong person set. Fails on current code (left ["p2","p4"] vs
right ["p3","p4"]).
* fix(engine): collision-free anti-join correlation tag for nested negation
The set-oriented anti-join tagged the outer batch with a fixed column name and
read it back by name. Under a nested slow-path anti-join the enclosing tag rides
through the inner pipeline, so the inner call produced a duplicate field; Arrow
permits duplicate names and column_by_name returns the first, so the inner
negation mis-correlated against the outer row indices.
Choose a tag name not already present in the batch (suffix-incremented), so each
nesting level reads its own correlation column. Turns the fan-out regression
green; the existing nested/fast-vs-slow/proptest anti-join invariants still pass.
* fix(engine): cap cross-type hops in the Expand cost model
gather_cost_inputs fed the requested max_hops into choose_expand_mode even though
execute_expand_indexed runs at most one hop for a cross-type edge. So a cross-type
variable-length expand (e.g. worksAt{1,5}) had its indexed cost scaled by 5 while
only one hop runs, skewing the chooser toward CSR (an unnecessary whole-graph
build) near the crossover. Results were unaffected (modes are equivalent); this
is a plan-accuracy fix.
Add cost_effective_hops(requested, same_type) — caps to 1 for cross-type — and
apply it in gather_cost_inputs so the estimate matches what executes. Unit test
covers the cap and the crossover consequence (capped 1 hop stays indexed where
the requested 5 would have flipped to CSR).
* perf(engine): realize anti-join CSR lazily + reuse a warm CSR in the chooser
Two CSR build/reuse fixes flagged on the set-oriented anti-join work (results
unchanged — plan/perf accuracy):
- execute_anti_join called graph_index.get() (the O(|E|) whole-graph CSR build)
unconditionally, but only the bulk fast path consumes it; a filtered/nested
slow-path anti-join's inner Expand picks its own access path. Gate the build on
a pure shape predicate (bulk_anti_join_applies) so a selective anti-join over a
large graph no longer pays a build it won't use.
- gather_cost_inputs hardcoded csr_cached=false, so once an earlier op realized
the CSR, later Expands still cost it as a cold build and could pick per-hop
indexed scans over reusing the warm in-memory CSR. Add GraphIndexHandle::
is_built() and thread it through so the chooser reuses a materialized CSR.
Anti-join, cross-type, proptest-equivalence, and chooser unit tests stay green.
* test(engine): RAII traversal-mode guard in proptest equivalence
prop_expand_indexed_eq_csr set/cleared OMNIGRAPH_TRAVERSAL_MODE manually; a panic
between set and clear (e.g. a query unwrap on a generated case) would leak the
forced mode into proptest's shrink/subsequent cases and mask the divergence under
test. Replace with a ModeGuard that clears on drop (including on unwind), scoping
the forced mode to a single query.
* test(engine): regression for multi-hop anti-join hop bounds
The bulk anti-join fast path answers via has_neighbors (one-hop existence), so
not { $p knows{2,2} $x } wrongly drops a node with a 1-hop neighbor but no
2-hop path. On a->b (sink) and c->d->e, only c has a 2-hop path; the query should
keep [a,b,d,e]. Fails on current code (left ["b","e"] — only the sinks).
* fix(engine): restrict anti-join bulk fast path to one-hop expands
bulk_anti_join_applies accepted any single Expand, but try_bulk_anti_join_mask
decides via the CSR has_neighbors one-hop existence check — wrong for multi-hop
negations. Require min_hops==1 && max_hops==1 in the predicate; anything else
falls to the slow path, whose inner Expand runs the real bounded traversal.
Turns the multi-hop regression green; one-hop anti-joins unchanged.
* fix(engine): IndexCoverage reports Degraded for uncovered fragments
key_column_index_coverage checked BTREE-exists + physical_rows but not that the
index actually covers the current fragments. Since edge-index creation is skipped
once a BTREE exists, fragments appended later stay unindexed while coverage still
reported Indexed — so the cost chooser priced a partly-full scan as fully indexed.
Compare the BTREE's fragment_bitmap (public on lance_table IndexMetadata) against
the dataset's current fragment ids; report Degraded when any are uncovered. A None
bitmap means Lance can't report coverage — don't over-degrade. Results are
unaffected (the scan returns unindexed-fragment rows either way); this corrects
the cost signal.
Test: a freshly-loaded edge BTREE is Indexed; after appending an edge the new
fragment is uncovered → Degraded. Surface guard pins IndexMetadata.fragment_bitmap.
* docs: clarify the Expand frontier ceiling bounds the initial dispatch frontier
The cap is applied at dispatch on the initial frontier; per-hop fan-out
(union_dense) is not hard-capped. Correct the constants.md and query-language.md
claims: the ceilings bound the initial-dispatch frontier/hops, the cost model
estimates total indexed work as ~hops*frontier*fanout (pricing dense fan-out
toward CSR), and per-hop work is not a hard bound. Drops the overstated 'hard
caps bound indexed work' / 'cost ∝ frontier' wording.
* fix(loader): enforce composite @unique(a, b) as a true composite key
Node/edge composite uniqueness constraints were flattened into a single
list of property names, so @unique(a, b) was enforced as independent
single-field checks @unique(a) AND @unique(b) at intake. Preserve the
constraint grouping and check each group as a composite key, mirroring
the merge-path enforcement. Error messages now name the full composite.
MR-983
* docs: clarify unit-separator comment in composite unique check
* docs: fix separator reference in composite unique comment (merge.rs also uses U+001F)
* fix(merge): align composite @unique key separator with intake (U+001F)
The branch-merge path (update_unique_constraints) joined composite key
columns with '|', while intake joins with U+001F. The same @unique(a, b)
was keyed two different ways, and '|'-join can raise phantom merge
conflicts for values containing '|' (e.g. ('x|y','z') vs ('x','y|z')).
Factor the tuple-join into one shared helper (loader::composite_unique_key)
so the intake and merge paths cannot drift again. Add branching regression
tests for edge @unique(src, dst) on the merge path.
Refs MR-983.
---------
Co-authored-by: Ragnor Comerford <ragnor.comerford@gmail.com>
Co-authored-by: Andrew Altshuler <andrew@collectivelab.io>
* docs(invariants): note the non-atomic manifest->commit-graph publish gap
Every graph publish commits __manifest then appends _graph_commits as two
separate writes; a crash between them leaves the manifest ahead of the commit
DAG. Live reads + durability are unaffected (reads resolve via the manifest) and
recovery does not repair it; impact is bounded to commit history / time-travel
by commit id / merge-base completeness. Pre-existing across all publishes, not
the optimize reconcile specifically. Documented as a Known Gap; the fix is a
commit-graph reconcilable from the manifest, not a recovery sidecar.
* fix(maintenance): route uncovered drift through repair
* fix(maintenance): harden repair review feedback
require_code_owner_reviews + count=1 with no bypass meant EVERY PR needed a
code-owner approval — including code owners' own PRs, which can't be
self-approved, so an owner's PR deadlocked on the other owner (forcing admin
overrides). Intended behavior: review is required only for non-owners.
Add bypass_pull_request_allowances for the two engineering owners (ragnorc,
aaltshuler): they merge their own PRs after CI without a second review;
non-owners still require a code-owner approval. CI status checks remain
required for everyone. Applied live via scripts/apply-branch-protection.sh.
Note: the bypass list mirrors codeowners-roles.yml engineering members by hand
(render-codeowners.py doesn't generate it) — keep in sync on owner changes.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The `test_windows_binaries` job ran a full Windows --release build +
smoke test on every code PR. It was a non-required (non-blocking) check,
so it never gated a merge — it only burned the slowest/most expensive
runner (windows-latest, --release, 75-min ceiling) on every code change.
Windows binary validation is already covered (better) on release tags:
release.yml's `smoke_windows_installer` (on v* tags) builds the release
binaries, installs via scripts/install.ps1, and smoke-runs
`omnigraph.exe version` + `omnigraph-server.exe --help` — the same smoke
test plus the real installer path. Nothing `needs:` the removed job.
Trade-off (accepted): a PR that breaks the Windows build or install.ps1
syntax is now caught at release-cut rather than at PR time. install.ps1
and platform-specific code change rarely; the cost savings on every PR
outweigh the earlier signal.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The maintenance.rs row referenced `optimize_reconciles_preexisting_manifest_head_drift`,
which never existed (leftover from the reconcile-drift heuristic removed in #141).
The actual second test is `optimize_defers_when_recovery_sidecar_is_pending`.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(optimize): cover manifest publish + HEAD-drift reconcile
Red against the pre-fix optimize, which ran compact_files without
publishing the compacted version to __manifest:
- maintenance: optimize must publish so the manifest table_version
tracks the compacted Lance HEAD and a later schema apply succeeds;
and must reconcile a pre-existing manifest-behind-HEAD drift (forged
via raw Lance compaction) so strict writes commit again.
- end_to_end + composite_flow: post-optimize query / strict update /
reopen in the full lifecycle (the canonical flow previously omitted
post-optimize writes as a documented "known limitation").
- failpoints: a crash between compaction and the manifest publish rolls
forward on next open.
* fix(optimize): publish compaction to manifest and reconcile HEAD drift
optimize ran Lance compact_files without publishing the new version to
__manifest, so the manifest table_version lagged the Lance HEAD: reads
stayed pinned to the pre-compaction version, and the next schema apply or
strict update/delete failed its HEAD-vs-manifest precondition with
"stale view ... refresh and retry" (open-time recovery rollback inflated
the gap on retry).
optimize now publishes each compacted table's version under the
per-(table, main) write queue, guarded by a manifest CAS and a
SidecarKind::Optimize recovery sidecar (loose-match; roll-forward is safe
because compaction is content-preserving). When a table has nothing left
to compact but its Lance HEAD is already ahead of the manifest pin
(pre-fix drift, or a recovery restore commit), optimize reconciles the
manifest forward to HEAD (metadata-only, no sidecar). Caches and the
CSR/CSC graph index are invalidated after a publish.
Docs updated (maintenance, storage, branches-commits, writes, testing).
* test(recovery): rollback convergence + optimize-defer regressions
Red against the current code, landed before the fix:
- recovery: after the open-time sweep rolls a sidecar back, the manifest
must track Lance HEAD (no residual drift) so a follow-up schema apply
succeeds — the original "+1 per retry" loop. Today roll-back restores
without publishing, so the manifest lags HEAD and the apply fails its
HEAD-vs-manifest precondition.
- maintenance: optimize must refuse while a recovery sidecar is pending —
operating on an unrecovered graph could publish a partial write the
sweep would roll back.
Also removes optimize_reconciles_preexisting_manifest_head_drift: the
ad-hoc drift reconcile it covered is replaced by recovery-side convergence.
* fix(recovery): converge manifest on roll-back; optimize defers on pending recovery
Root of PR #141's review findings and the original "+1 per retry" loop:
a Lance HEAD ahead of the manifest was ambiguous (benign content-preserving
drift vs. a partial write a sidecar will roll back), and optimize's reconcile
guessed it benign. Close the class instead of guessing:
- Recovery roll-back now PUBLISHES the restored version (via a
push_table_update_at_head helper shared with roll-forward), so the manifest
tracks the Lance HEAD after recovery — symmetric with roll-forward. This
fixes the +1 loop (after one roll-back the retry's HEAD-vs-manifest
precondition passes) and removes the only remaining source of orphaned
drift. The audit still records the logical rolled-back-to version; the
manifest is published at the restore commit (identical content).
- optimize drops the ad-hoc drift reconcile and instead REFUSES when a
__recovery sidecar is pending, so it only ever operates on a recovered
graph (manifest == HEAD); its compaction publish can no longer commit a
partial write. With the reconcile gone, the blob-skip-vs-reconcile gap is
moot.
Updates the rollback recovery-test helper (manifest == HEAD after roll-back),
the failpoints assertions, and the user/dev docs.
* test(recovery): fix rollback assertion for manifest convergence
The roll-back-publishes change makes the manifest version advance after a
SchemaApply roll-back (to the old-schema content), so the
schema_apply_without_schema_staging_rolls_back_on_next_open assertion must
be `version > pre`, not `version == pre`. This update was dropped during
the commit churn and surfaced as a CI Test Workspace failure; the
old-schema-preserved intent stays covered by count_rows + _schema.pg + the
RolledBack convergence invariant.
The loader read input line-by-line (reader.lines() + serde_json::from_str per line), so any delta where a JSON object spanned multiple lines failed with 'invalid JSON on line 1: EOF while parsing an object'. Compact JSONL worked; pretty-printed JSON never did.
Switch to a streaming value deserializer (Deserializer::from_reader().into_iter::<Value>()), which treats any whitespace (including newlines inside objects) as a separator — so both compact JSONL and pretty-printed JSON load. Error labels switch from line numbers to record numbers (line numbers are meaningless once objects span lines).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Ragnor Comerford <ragnor.comerford@gmail.com>
* feat(engine): sweep legacy __run__ branches via v2→v3 manifest migration
Pre-v0.4.0 graphs can carry stale `__run__<id>` staging branches on the
`__manifest` dataset, left by the Run state machine removed in MR-771. Lance's
`list_branches` still enumerates them, so they leak into `branch_list()` and
count as blocking branches at schema-apply time.
Add a one-time `migrate_v2_to_v3` arm to the internal-schema dispatcher: on the
first read-write open it enumerates `__manifest` branches, deletes every
`__run__*` ref, and bumps the stamp to 3. Idempotent under retry (re-enumerates
fresh each run). The `"__run__"` prefix is inlined so the migration does not
depend on the run_registry guard that MR-770 removes next.
This is the prerequisite sweep; the guard removal follows in the next commit.
* refactor(engine): remove the legacy __run__ branch guard (MR-770)
With the v2→v3 migration sweeping stale `__run__*` branches off `__manifest`
on first read-write open, the defense-in-depth `is_internal_run_branch` guard
is no longer needed.
- delete `db/run_registry.rs`; drop the module + re-export from `db/mod.rs`
- collapse `is_internal_system_branch` to the schema-apply-lock check only
- `ensure_public_branch_ref`: drop the run-ref rejection; `__run__*` is now an
ordinary branch name
- `branch_merge`: reject `is_internal_system_branch` (was run-only) so the
schema-apply lock is rejected consistently with create/delete — a small,
deliberate tightening
- update the inline schema-apply test + the writes integration tests
(`public_branch_apis_reject_internal_run_refs` →
`public_branch_apis_reject_internal_system_refs`, which also asserts
`__run__*` now creates successfully)
- docs: flip the "pending production sweep / defense-in-depth" notes to
"auto-swept by the v2→v3 migration"; document the read-only-open limitation
Known residual: the inert `_graph_runs.lance` / `_graph_run_actors.lance` bytes
remain until a `StorageAdapter::delete_prefix` primitive lands.
* fix(engine): run __run__ sweep at Omnigraph::open, not only on publish
Review (PR #132) caught a regression: removing __run__ from
`is_internal_system_branch` exposed legacy `__run__*` branches to the
schema-apply blocking-branch checks (schema_apply.rs:104 and :778) and to
`branch_list()`, but the v2→v3 sweep ran only inside the publisher's
`load_publish_state`. On a pre-v0.4.0 graph whose first write is a schema
apply, the blocking-branch check fires before any publish, so apply failed
with "found non-main branches: __run__…". The same lazy timing also created a
reverse hazard: a user-created `__run__*` branch on a still-v2 graph could be
deleted by the first publish's sweep.
Fix: run the internal-schema migration in `Omnigraph::open(ReadWrite)` (new
`manifest::migrate_on_open`), before the coordinator reads branch state. The
sweep now lands before any branch-observing code, and a graph is stamped v3 at
open — so the one-time sweep can never catch a legitimately-created branch.
Both checks and `branch_list` see the swept graph; correct by construction for
every write path.
Accepted residual: a read-only open of an unmigrated legacy graph still lists
`__run__*` (read-only opens must not write, so they can't sweep). Documented.
Regression test `legacy_run_branch_is_swept_on_open_and_does_not_block_schema_apply`
confirmed RED before the fix (panicked on the branch_list leak assertion) and
GREEN after. Also updates the stale schema_apply.rs comment, the writes.md
"Migration code" section, and adds the v3 row to storage.md's migration table.
* test(engine): sweep multiple legacy __run__ branches; doc nit
Strengthen the v2→v3 migration test to synthesize three `__run__*` branches
(a real legacy graph accumulates one per run) so the migration's delete loop
is exercised on a single reused dataset handle, not just a single branch.
Confirms multi-branch deletion is safe.
Also drop a stale "active runs" reference from the branch_delete doc line.
* fix(engine): force-delete in __run__ sweep for concurrency safety
`migrate_v2_to_v3` ran `Dataset::delete_branch` (= `branches().delete(.., false)`),
which errors "BranchContents not found" if the branch is already gone. Since the
sweep now runs in `Omnigraph::open(ReadWrite)`, two processes opening the same
legacy v2 graph concurrently would race: one wins each delete, the other's open
fails. The migration only claimed idempotency under *sequential* retry.
Switch to `Dataset::force_delete_branch` (= `delete(.., true)`), Lance's
documented path for cleaning up zombie branches, which tolerates an
already-absent branch. The sweep is now idempotent under concurrent runners and
robust to partial/zombie state. Found in self-review; no behavior change for the
common single-open path.
* docs(release): note MR-770 __run__ cleanup in v0.6.1
* docs(branches): reconcile branch cleanup semantics
Restores aaltshuler as an `engineering` code-owner (removed in #142), so
`crates/**` and repo-infra PRs have a second reviewer besides the sole
owner ragnorc — unblocking review of author-ragnorc PRs (e.g. #132) that
ragnorc cannot self-approve.
Edited the source of truth (.github/codeowners-roles.yml) and re-rendered
.github/CODEOWNERS + the docs/dev/codeowners.md tables via
.github/scripts/render-codeowners.py, per the documented flow.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Formalize the public contribution surface. Maintainers keep a separate internal
process and are exempt from the intake gates; everyone stays bound by review,
CODEOWNERS, and branch protection.
Model:
- Issues = problem reports only (bug form + config.yml redirects ideas to
Discussions and disables blank issues).
- Discussions = ideas + RFC incubation.
- RFCs = anyone (incl. external) authors docs/rfcs/NNNN-*.md; a maintainer
merging it is acceptance. Distinct from the maintainer-internal
docs/dev/rfc-00N-* track.
- PRs = link an `accepted` issue or accepted RFC, or use the trivial fast-lane
(typos/docs/deps). Enforced softly to start (template + review).
Adds GOVERNANCE.md, rewrites CONTRIBUTING.md, adds docs/rfcs/ (README +
template), .github issue/PR/discussion templates. Wires docs/rfcs/ into the
doc-link checker (excluded like releases; linked from docs/dev/index.md).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The CODEOWNERS required checks blocked every PR — the real root cause was a
name mismatch, compounded by a path filter:
- branch-protection.json required the contexts `CODEOWNERS / drift` and
`CODEOWNERS / noedit` (the GitHub UI "workflow / job-id" display form), but
the jobs report check-run names from their `name:` fields — "CODEOWNERS
matches source" / "CODEOWNERS not hand-edited". The required contexts
therefore never matched any reported check and sat permanently pending.
- The workflow was also path-filtered to CODEOWNERS files, so it didn't even
run for most PRs.
Net effect: with both required checks unsatisfiable, every PR could only land
via admin override (e.g. #140).
Fixes:
- A: drop the `paths:` filter so the workflow runs on every PR and both
required contexts always report.
- name fix: point branch-protection.json at the actual job names verbatim, and
add a doc note that the contexts must equal the job `name:` values.
- B: the `drift` job now re-renders and, on same-repo PRs, auto-commits the
regenerated artifacts back to the branch (mirrors the openapi.json job in
ci.yml); forks / manual runs strict-check instead. Contributors no longer
run the script by hand.
- D: render-codeowners.py also generates a "who owns what" path->owners +
roles table spliced into docs/dev/codeowners.md between markers, so the
human-readable view never drifts. Idempotent; CODEOWNERS output unchanged.
- docs: correct the stale `enforce_admins: true` line (JSON and live are
false).
NOTE: the branch-protection.json change only takes effect after an admin runs
`./scripts/apply-branch-protection.sh` (deliberate manual step, per
docs/dev/branch-protection.md). Until then `main` still requires the old
mismatched contexts, so this PR itself needs an admin-override merge — the last
one that should be necessary.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
