Method load_vector_checkpoints

nodedb/src/data/executor/vector_checkpoint.rs:106–167

Load HNSW checkpoints from disk on startup, before WAL replay. For each checkpoint file, loads the index. WAL replay then only needs to process entries after the checkpoint LSN.

(&mut self)

    /// For each checkpoint file, loads the index. WAL replay then only
    /// needs to process entries after the checkpoint LSN.
    pub fn load_vector_checkpoints(&mut self) {
        let ckpt_dir = self.data_dir.join("vector-ckpt");
        if !ckpt_dir.exists() {
            return;
        }

        let entries = match std::fs::read_dir(&ckpt_dir) {
            Ok(e) => e,
            Err(_) => return,
        };

        let mut loaded = 0;
        for entry in entries.flatten() {
            let path = entry.path();
            if path.extension().and_then(|e| e.to_str()) != Some("ckpt") {
                continue;
            }

            // Checkpoint filenames are `"{tid}:{coll}.ckpt"` — parse back to tuple.
            let filename = path
                .file_stem()
                .and_then(|s| s.to_str())
                .unwrap_or("")
                .to_string();
            if filename.is_empty() {
                continue;
            }
            let tuple_key = parse_build_key(&filename);

            let Ok(bytes) = nodedb_wal::segment::read_checkpoint_dontneed(&path) else {
                continue;
            };
            let kek = self.vector_checkpoint_kek.as_ref();
            let load_result =
                crate::engine::vector::collection::VectorCollection::from_checkpoint(&bytes, kek);
            let collection = match load_result {
                Ok(Some(c)) => c,
                Ok(None) => continue,
                Err(e) => {
                    tracing::warn!(
                        core = self.core_id,
                        key = %filename,
                        error = %e,
                        "vector checkpoint rejected"
                    );
                    continue;
                }
            };
            tracing::info!(
                core = self.core_id,
                key = %filename,
                vectors = collection.len(),
                "loaded vector checkpoint"
            );
            self.vector_collections.insert(tuple_key, collection);
            loaded += 1;
        }