fix(lmdb): tranh MDB_MAP_FULL bang auto-grow map truoc khi ghi

Cargo.toml

@@ -19,7 +19,7 @@ members = [
 ]
 
 [workspace.package]
-version = "2.2.13"
+version = "2.2.14"
 edition = "2021"
 rust-version = "1.80"
 license = "MIT"

crates/codegraph-api/src/tools.rs

@@ -269,8 +269,8 @@ pub async fn dispatch_diff(
     args: Value,
 ) -> Result<String> {
     let diff = arg_str(&args, "diff")?;
-    let parsed = codegraph_graph::diff::parse_unified_diff(diff)
-        .map_err(|e| Error::Invalid(e.to_string()))?;
+    let parsed =
+        codegraph_graph::parse_unified_diff(diff).map_err(|e| Error::Invalid(e.to_string()))?;
 
     let idx = shared.ensure_fresh().await;
     let report = idx.diff_assess(&parsed, Some(root.as_std_path())).await;
@@ -402,8 +402,8 @@ pub async fn dispatch_branch_compare(
     let base = arg_str(&args, "base")?;
     let head = args.get("head").and_then(|v| v.as_str());
     let diff = git.diff(base, head).await?;
-    let parsed = codegraph_graph::diff::parse_unified_diff(&diff)
-        .map_err(|e| Error::Invalid(e.to_string()))?;
+    let parsed =
+        codegraph_graph::parse_unified_diff(&diff).map_err(|e| Error::Invalid(e.to_string()))?;
 
     let idx = shared.ensure_fresh().await;
     let report = idx.diff_assess(&parsed, Some(root.as_std_path())).await;
@@ -461,8 +461,8 @@ pub async fn dispatch_diff_simulate(
     args: Value,
 ) -> Result<String> {
     let diff = arg_str(&args, "diff")?;
-    let parsed = codegraph_graph::diff::parse_unified_diff(diff)
-        .map_err(|e| Error::Invalid(e.to_string()))?;
+    let parsed =
+        codegraph_graph::parse_unified_diff(diff).map_err(|e| Error::Invalid(e.to_string()))?;
     let base_ref = args
         .get("base_ref")
         .and_then(|v| v.as_str())

crates/codegraph-bench/src/bin/mem.rs

@@ -16,8 +16,8 @@ use camino::Utf8Path;
 use clap::Parser;
 use codegraph_bench::{BenchOptions, Repo, extract, index_at, orchestrator, run_queries};
 use codegraph_core::{Annotation, CallRecord, EffectType, ScopeLevel, Symbol, SymbolKind};
-use codegraph_graph::meminfo::{MemTracker, fmt_bytes, rss_bytes};
-use codegraph_graph::memtrack::MemBreakdown;
+use codegraph_graph::MemBreakdown;
+use codegraph_graph::{MemTracker, fmt_bytes, rss_bytes};
 use std::sync::OnceLock;
 
 fn runtime() -> &'static tokio::runtime::Runtime {
@@ -555,7 +555,7 @@ fn main() -> anyhow::Result<()> {
         println!("\n=== breakdown: {} ===", r.repo);
         print_breakdown(&r.breakdown, &r.caches, r.rss_after_index);
     }
-    if let Some(peak) = codegraph_graph::meminfo::peak_rss_bytes() {
+    if let Some(peak) = codegraph_graph::peak_rss_bytes() {
         println!("\npeak RSS (VmHWM): {}", fmt_bytes(peak));
     } else {
         println!("\npeak RSS: không đọc được trên nền tảng này");

crates/codegraph-extract/src/config.rs

@@ -470,7 +470,7 @@ pub struct ExtractConfig {
     /// Backend storage được chọn trong config (mặc định sqlite).
     pub storage: StorageConfig,
     /// Cấu hình embedding backend (semantic search) — đọc từ `[embedding]`.
-    pub embedding: codegraph_graph::embeddings::EmbeddingConfig,
+    pub embedding: codegraph_graph::EmbeddingConfig,
     /// Cấu hình Jev rerank cho `codegraph_context` — đọc từ `[jev]`.
     pub jev: codegraph_core::JevConfig,
     /// Cấu hình document graph — đọc từ `[docgraph]`.
@@ -536,7 +536,7 @@ impl ExtractConfig {
             header_language: parse_header_language(file.languages.headers.as_deref()),
             effect_classifier,
             flow_classifier,
-            embedding: codegraph_graph::embeddings::EmbeddingConfig::from_raw(
+            embedding: codegraph_graph::EmbeddingConfig::from_raw(
                 file.embedding.backend.as_deref(),
                 file.embedding.model.as_deref(),
                 file.embedding.cache_dir.as_deref(),
@@ -606,7 +606,7 @@ impl ExtractConfig {
         // Áp dụng config embedding (backend/model/cache) cho process trước khi
         // mở index — `GraphIndex::new_with_storage` đọc global này để quyết định
         // có bật vector index hay không (opt-in: chỉ khi backend = "fastembed").
-        codegraph_graph::embeddings::set_embedding_config(self.embedding.clone());
+        codegraph_graph::set_embedding_config(self.embedding.clone());
         // Cùng thời điểm: `[jev]` là config toàn cục cho `codegraph-context`
         // (module context không phụ thuộc `codegraph-extract`, nên truyền qua
         // global giống embedding).

crates/codegraph-graph/src/embeddings.rs

@@ -199,6 +199,8 @@ fn expand_tilde(path: &str) -> Option<PathBuf> {
 }
 
 /// Suffix file extension của sqlite-vss theo OS (`.dylib` / `.so` / `.dll`).
+// Chỉ `storage/sqlite.rs` dùng — tắt feature `sqlite` thì cặp hàm này dead code.
+#[cfg_attr(not(feature = "sqlite"), allow(dead_code))]
 fn vss_lib_suffix() -> &'static str {
     if cfg!(target_os = "macos") {
         "dylib"
@@ -216,6 +218,7 @@ fn vss_lib_suffix() -> &'static str {
 /// dò `<cache_dir>/vss`. Chỉ trả `Some` khi file thực sự tồn tại → caller có
 /// thể yên tâm thêm extension vào kết nối SQLite mà không làm hỏng `open`
 /// khi thiếu binary.
+#[cfg_attr(not(feature = "sqlite"), allow(dead_code))]
 pub fn resolve_vss_extensions() -> Option<(PathBuf, PathBuf)> {
     let cfg = embedding_config();
     let dir = cfg

crates/codegraph-graph/src/lib.rs

@@ -81,16 +81,29 @@ use tokio::sync::RwLock;
 
 #[cfg(feature = "bloom-search")]
 mod bloom;
-pub mod diff;
-pub mod embeddings;
+mod diff;
+mod embeddings;
 mod lru;
-pub mod meminfo;
-pub mod memtrack;
+mod meminfo;
+mod memtrack;
 mod radix;
 mod search;
 mod shared;
 mod storage;
-pub mod vector_index;
+// Chỉ `VectorIndex::knn` được dùng nội bộ; phần API còn lại (insert/set_all/
+// kmeans...) giữ cho semantic search opt-in + test, nên tắt dead-code ở đây
+// thay vì để cả module public.
+#[allow(dead_code)]
+mod vector_index;
+
+// Các module nội bộ trên chỉ mở đúng phần API mà crate khác dùng — thay cho
+// `pub mod`, để bề mặt public không phình theo từng file nội bộ.
+pub use crate::diff::{FileDiff, Hunk, ParsedDiff, parse_unified_diff};
+#[cfg(feature = "fastembed")]
+pub use crate::embeddings::warm_model_cache;
+pub use crate::embeddings::{EmbeddingConfig, set_embedding_config};
+pub use crate::meminfo::{MemTracker, fmt_bytes, peak_rss_bytes, rss_bytes};
+pub use crate::memtrack::MemBreakdown;
 
 pub use shared::SharedGraphIndex;
 

crates/codegraph-graph/src/meminfo.rs

@@ -11,7 +11,7 @@
 //! - **Platform khác**: `None` — caller phải xử lý, KHÔNG panic.
 //!
 //! ```no_run
-//! use codegraph_graph::meminfo::{MemTracker, rss_bytes};
+//! use codegraph_graph::{MemTracker, rss_bytes};
 //!
 //! let mut t = MemTracker::new();
 //! t.mark("baseline");
@@ -92,7 +92,7 @@ pub fn fmt_bytes(bytes: u64) -> String {
 /// (đỉnh), nên benchmark tự lấy mẫu tại các mốc rồi giữ max.
 ///
 /// ```no_run
-/// # use codegraph_graph::meminfo::MemTracker;
+/// # use codegraph_graph::MemTracker;
 /// let mut t = MemTracker::new();
 /// t.mark("open");
 /// t.mark("ingest");