is_multiple_of(N) returns true for 0, which caused debug/info
progress messages to fire at doc_num=0 (the start of every page)
rather than only at the intended 50/100 milestones. Add != 0
check to both the debug (every 50) and info (every 100) log sites.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add the ability to sync specific issues or merge requests by IID without
running a full incremental sync. This enables fast, targeted data refresh
for individual entities — useful for agent workflows, debugging, and
real-time investigation of specific issues or MRs.
Architecture:
- New CLI flags: --issue <IID> and --mr <IID> (repeatable, up to 100 total)
scoped to a single project via -p/--project
- Preflight phase validates all IIDs exist on GitLab before any DB writes,
with TOCTOU-aware soft verification at ingest time
- 6-stage pipeline: preflight -> fetch -> ingest -> dependents -> docs -> embed
- Each stage is cancellation-aware via ShutdownSignal
- Dedicated SyncRunRecorder extensions track surgical-specific counters
(issues_fetched, mrs_ingested, docs_regenerated, etc.)
New modules:
- src/ingestion/surgical.rs: Core surgical fetch/ingest/dependent logic
with preflight_fetch(), ingest_issue_by_iid(), ingest_mr_by_iid(),
and fetch_dependents_for_{issue,mr}()
- src/cli/commands/sync_surgical.rs: Full CLI orchestrator with progress
spinners, human/robot output, and cancellation handling
- src/embedding/pipeline.rs: embed_documents_by_ids() for scoped embedding
- src/documents/regenerator.rs: regenerate_dirty_documents_for_sources()
for scoped document regeneration
Database changes:
- Migration 027: Extends sync_runs with mode, phase, surgical_iids_json,
per-entity counters, and cancelled_at column
- New indexes: idx_sync_runs_mode_started, idx_sync_runs_status_phase_started
GitLab client:
- get_issue_by_iid() and get_mr_by_iid() single-entity fetch methods
Error handling:
- New SurgicalPreflightFailed error variant with entity_type, iid, project,
and reason fields. Shares exit code 6 with GitLabNotFound.
Includes comprehensive test coverage:
- 645 lines of surgical ingestion tests (wiremock-based)
- 184 lines of scoped embedding tests
- 85 lines of scoped regeneration tests
- 113 lines of GitLab client single-entity tests
- 236 lines of sync_run surgical column/counter tests
- Unit tests for SyncOptions, error codes, and CLI validation
Three micro-optimizations with zero behavioral change:
1. timeline_collect.rs: Reorder format!() before enum construction so
the owned String moves into the variant directly, eliminating
.clone() on state, label, and milestone strings in StateChanged,
LabelAdded/Removed, and MilestoneSet/Removed event paths.
2. pipeline.rs: Use Arc<str> for doc_hash shared across a document's
chunks instead of cloning the full String per chunk. Also remove
redundant embed_buf.reserve() since extend_from_slice already
handles growth and the buffer is reused across iterations.
3. rrf.rs: Pre-allocate HashMap with combined vector+fts result count
via with_capacity() to avoid rehashing during RRF score accumulation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Embedding pipeline improvements building on the concurrent batching
foundation:
- Track docs_embedded vs chunks_embedded separately. A document counts
as embedded only when ALL its chunks succeed, giving accurate
progress reporting. The sync command reads docs_embedded for its
document count.
- Reuse a single Vec<u8> buffer (embed_buf) across all store_embedding
calls instead of allocating per chunk. Eliminates ~3KB allocation per
768-dim embedding.
- Detect and record errors when Ollama silently returns fewer
embeddings than inputs (batch mismatch). Previously these dropped
chunks were invisible.
- Improve retry error messages: distinguish "retry returned unexpected
result" (wrong dims/count) from "retry request failed" (network
error) instead of generic "chunk too large" message.
- Convert all hot-path SQL from conn.execute() to prepare_cached() for
statement cache reuse (clear_document_embeddings, store_embedding,
record_embedding_error).
- Record embedding_metadata errors for empty documents so they don't
appear as perpetually pending on subsequent runs.
- Accept concurrency parameter (configurable via config.embedding.concurrency)
instead of hardcoded EMBED_CONCURRENCY=2.
- Add schema version pre-flight check in embed command to fail fast
with actionable error instead of cryptic SQL errors.
- Fix --retry-failed to use DELETE instead of UPDATE. UPDATE clears
last_error but the row still matches config params in the LEFT JOIN,
making the doc permanently invisible to find_pending_documents.
DELETE removes the row entirely so the LEFT JOIN returns NULL.
Regression test added (old_update_approach_leaves_doc_invisible).
- Add chunking forward-progress guard: after floor_char_boundary()
rounds backward, ensure start advances by at least one full
character to prevent infinite loops on multi-byte sequences
(box-drawing chars, smart quotes). Test cases cover the exact
patterns that caused production hangs on document 18526.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three fixes to the embedding pipeline:
1. Concurrent HTTP batching: fire EMBED_CONCURRENCY (2) Ollama requests
in parallel via join_all, then write results serially to SQLite.
~2x throughput improvement on GPU-bound workloads.
2. UTF-8 boundary safety: all computed byte offsets in split_into_chunks
(paragraph/sentence/word break finders + overlap advance) now use
floor_char_boundary() to prevent panics on multi-byte characters
like smart quotes and non-breaking spaces.
3. CHUNK_MAX_BYTES reduced from 6000 to 1500 to fit nomic-embed-text's
actual 2048-token context window, eliminating context-length retry
storms that were causing 10x slowdowns.
Also threads ShutdownSignal through embed pipeline for graceful Ctrl+C.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Change OllamaClient::embed_batch to accept &[&str] instead of
Vec<String>. The EmbedRequest struct now borrows both model name and
input texts, eliminating per-batch cloning of chunk text (up to 32KB
per chunk x 32 chunks per batch). Serialization output is identical
since serde serializes &str and String to the same JSON.
In hybrid search, defer the RrfResult->HybridResult mapping until
after filter+take, so only `limit` items (typically 20) are
constructed instead of up to 1,500 at RECALL_CAP. Also switch
filtered_ids to into_iter() to avoid an extra .copied() pass.
Switch FTS search_fts from prepare() to prepare_cached() for statement
reuse across repeated searches. Benchmarked at ~1.6x faster.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Change detection queries (embedding/change_detector.rs):
- Replace triple-EXISTS subquery pattern with LEFT JOIN + NULL check
- SQLite now scans embedding_metadata once instead of three times
- Semantically identical: returns docs needing embedding when no
embedding exists, hash changed, or config mismatch
Count queries (cli/commands/count.rs):
- Consolidate 3 separate COUNT queries for issues into single query
using conditional aggregation (CASE WHEN state = 'x' THEN 1)
- Same optimization for MRs: 5 queries reduced to 1
Search filter queries (search/filters.rs):
- Replace N separate EXISTS clauses for label filtering with single
IN() clause with COUNT/GROUP BY HAVING pattern
- For multi-label AND queries, this reduces N subqueries to 1
FTS tokenization (search/fts.rs):
- Replace collect-into-Vec-then-join pattern with direct String building
- Pre-allocate capacity hint for result string
Discussion truncation (documents/truncation.rs):
- Calculate total length without allocating concatenated string first
- Only allocate full string when we know it fits within limit
Embedding pipeline (embedding/pipeline.rs):
- Add Vec::with_capacity hints for chunk work and cleared_docs hashset
- Reduces reallocations during embedding batch processing
Backoff calculation (core/backoff.rs):
- Replace unchecked addition with saturating_add to prevent overflow
- Add test case verifying overflow protection
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Removes module-level doc comments (//! lines) and excessive inline doc
comments that were duplicating information already evident from:
- Function/struct names (self-documenting code)
- Type signatures (the what is clear from types)
- Implementation context (the how is clear from code)
Affected modules:
- cli/* - Removed command descriptions duplicating clap help text
- core/* - Removed module headers and obvious function docs
- documents/* - Removed extractor/regenerator/truncation docs
- embedding/* - Removed pipeline and chunking docs
- gitlab/* - Removed client and transformer docs (kept type definitions)
- ingestion/* - Removed orchestrator and ingestion docs
- search/* - Removed FTS and vector search docs
Philosophy: Code should be self-documenting. Comments should explain
"why" (business decisions, non-obvious constraints) not "what" (which
the code itself shows). This change reduces noise and maintenance burden
while keeping the codebase just as understandable.
Retains comments for:
- Non-obvious business logic
- Important safety invariants
- Complex algorithm explanations
- Public API boundaries where generated docs matter
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Three correctness fixes found during peer code review:
Embedding pipeline savepoint leak (HIGH severity):
The SAVEPOINT embed_page / RELEASE embed_page pattern had ~10 `?`
propagation points between them. Any error from record_embedding_error,
clear_document_embeddings, or store_embedding would exit the function
without rolling back, leaving the SQLite connection in a broken
transactional state and causing cascading failures for the rest of the
session. Fixed by extracting page processing into `embed_page()` and
wrapping with explicit rollback-on-error handling.
Dependent queue fail_job race (MEDIUM severity):
fail_job performed a SELECT followed by a separate UPDATE on the
attempts counter without a transaction. Under concurrent lock
reclamation, the attempts value could be read stale. Replaced with a
single atomic UPDATE that increments attempts and computes exponential
backoff entirely in SQL, also halving DB round-trips. Added explicit
error when the job no longer exists.
RRF duplicate document score inflation (MEDIUM severity):
If a retriever returned the same document_id multiple times, the RRF
score accumulated multiple rank contributions while the rank only
recorded the first occurrence. Moved the score accumulation inside the
`if is_none` guard so only the first occurrence per list contributes.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add end-to-end observability to the sync and ingest pipelines:
Sync command:
- Generate UUID-based run_id for each sync invocation, propagated through
all child spans for log correlation across stages
- Accept MetricsLayer reference to extract hierarchical StageTiming data
after pipeline completion for robot-mode performance output
- Record sync runs in DB via SyncRunRecorder (start/succeed/fail lifecycle)
- Wrap entire sync execution in a root tracing span with run_id field
Ingest command:
- Wrap run_ingest in an instrumented root span with run_id and resource_type
- Add project path prefix to discussion progress bars for multi-project clarity
- Reset resource_events_synced_for_updated_at on --full re-sync
Sync status:
- Expand from single last_run to configurable recent runs list (default 10)
- Parse and expose StageTiming metrics from stored metrics_json
- Add run_id, total_items_processed, total_errors to SyncRunInfo
- Add mr_count to DataSummary for complete entity coverage
Orchestrator:
- Add #[instrument] with structured fields to issue and MR ingestion functions
- Record items_processed, items_skipped, errors on span close for MetricsLayer
- Emit granular progress events (IssuesFetchStarted, IssuesFetchComplete)
- Pass project_id through to drain_resource_events for scoped job claiming
Document regenerator and embedding pipeline:
- Add #[instrument] spans with items_processed, items_skipped, errors fields
- Record final counts on span close for metrics extraction
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
11 isomorphic performance fixes from deep audit (no behavior changes):
- Eliminate double serialization: store_payload now accepts pre-serialized
bytes (&[u8]) instead of re-serializing from serde_json::Value. Uses
Cow<[u8]> for zero-copy when compression is disabled.
- Add SQLite cache_size (64MB) and mmap_size (256MB) pragmas
- Replace SELECT-then-INSERT label upserts with INSERT...ON CONFLICT
RETURNING in both issues.rs and merge_requests.rs
- Replace INSERT + SELECT milestone upsert with RETURNING
- Use prepare_cached for 5 hot-path queries in extractor.rs
- Optimize compute_list_hash: index-sort + incremental SHA-256 instead
of clone+sort+join+hash
- Pre-allocate embedding float-to-bytes buffer with Vec::with_capacity
- Replace RandomState::new() in rand_jitter with atomic counter XOR nanos
- Remove redundant per-note payload storage (discussion payload contains
all notes already)
- Change transform_issue to accept &GitLabIssue (avoids full struct clone)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Automated formatting and lint corrections from parallel agent work:
- cargo fmt: import reordering (alphabetical), line wrapping to respect
max width, trailing comma normalization, destructuring alignment,
function signature reformatting, match arm formatting
- clippy (pedantic): Range::contains() instead of manual comparisons,
i64::from() instead of `as i64` casts, .clamp() instead of
.max().min() chains, let-chain refactors (if-let with &&),
#[allow(clippy::too_many_arguments)] and
#[allow(clippy::field_reassign_with_default)] where warranted
- Removed trailing blank lines and extra whitespace
No behavioral changes. All existing tests pass unmodified.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Reduces CHUNK_MAX_BYTES from 32KB to 6KB and CHUNK_OVERLAP_CHARS from
500 to 200 to stay within nomic-embed-text's 8,192-token context
window. This commit addresses all downstream consequences of that
reduction:
- Config drift detection: find_pending_documents and
count_pending_documents now take model_name and compare
chunk_max_bytes, model, and dims against stored metadata. Documents
embedded with stale config are automatically re-queued.
- Overflow guard: documents producing >= CHUNK_ROWID_MULTIPLIER chunks
are skipped with a sentinel error recorded in embedding_metadata,
preventing both rowid collision and infinite re-processing loops.
- Deferred clearing: old embeddings are no longer cleared before
attempting new ones. clear_document_embeddings is deferred until the
first successful chunk embedding, so if all chunks fail the document
retains its previous embeddings rather than losing all data.
- Savepoints: each page of DB writes is wrapped in a SQLite savepoint
so a crash mid-page rolls back atomically instead of leaving partial
state (cleared embeddings with no replacements).
- Per-chunk retry on context overflow: when a batch fails with a
context-length error, each chunk is retried individually so one
oversized chunk doesn't poison the entire batch.
- Adaptive dedup in vector search: replaces the static 3x over-fetch
multiplier with a dynamic one based on actual max chunks per document
(using the new chunk_count column with a fallback COUNT query for
pre-migration data). Also replaces partial_cmp with total_cmp for
f64 distance sorting.
- Stores chunk_max_bytes and chunk_count (on sentinel rows) in
embedding_metadata to support config drift detection and adaptive
dedup without runtime queries.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements the embedding module that generates vector representations
of documents using a local Ollama instance with the nomic-embed-text
model. These embeddings enable semantic (vector) search and the hybrid
search mode that fuses lexical and semantic results via RRF.
Key components:
- embedding::ollama: HTTP client for the Ollama /api/embeddings
endpoint. Handles connection errors with actionable error messages
(OllamaUnavailable, OllamaModelNotFound) and validates response
dimensions.
- embedding::chunking: Splits long documents into overlapping
paragraph-aware chunks for embedding. Uses a configurable max token
estimate (8192 default for nomic-embed-text) with 10% overlap to
preserve cross-chunk context.
- embedding::chunk_ids: Encodes chunk identity as
doc_id * 1000 + chunk_index for the embeddings table rowid. This
allows vector search to map results back to documents and
deduplicate by doc_id efficiently.
- embedding::change_detector: Compares document content_hash against
stored embedding hashes to skip re-embedding unchanged documents,
making incremental embedding runs fast.
- embedding::pipeline: Orchestrates the full embedding flow: detect
changed documents, chunk them, call Ollama in configurable
concurrency (default 4), store results. Supports --retry-failed
to re-attempt previously failed embeddings.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
