Add TTL-based snapshot caching for DuckLake catalog

.github/workflows/ci.yml

@@ -31,7 +31,7 @@ jobs:
     strategy:
       matrix:
         os:
-          - ubuntu-latest
+          - ubuntu-latest-m
           - macos-latest
         rust: [stable]
     steps:

CLAUDE.md

@@ -44,10 +44,11 @@ The codebase follows a layered architecture with clear separation of concerns:
 
 2. **DataFusion Integration Layer** (`src/catalog.rs`, `src/schema.rs`, `src/table.rs`)
    - Bridges DuckLake concepts to DataFusion's catalog system
-   - `DuckLakeCatalog`: Implements `CatalogProvider`, uses dynamic metadata lookup (queries on every call to `schema()` and `schema_names()`)
+   - `DuckLakeCatalog`: Implements `CatalogProvider`, uses dynamic metadata lookup with configurable snapshot resolution
    - `DuckLakeSchema`: Implements `SchemaProvider`, uses dynamic metadata lookup (queries on every call to `table()` and `table_names()`)
    - `DuckLakeTable`: Implements `TableProvider`, caches table structure and file lists at creation time
    - **No HashMaps**: Catalog and schema providers query metadata on-demand rather than caching
+   - **Snapshot Resolution**: Configurable TTL (time-to-live) for balancing freshness and performance
 
 3. **Path Resolution** (`src/path_resolver.rs`)
    - Centralized utilities for parsing object store URLs and resolving hierarchical paths
@@ -77,7 +78,8 @@ The catalog uses a **pure dynamic lookup** approach with no caching at the catal
 - **DuckLakeCatalog** (`catalog.rs`):
   - `schema_names()`: Queries `list_schemas()` on every call
   - `schema()`: Queries `get_schema_by_name()` on every call
-  - `new()`: O(1) - only fetches snapshot ID and data_path
+  - `new()`: O(1) - only fetches data_path
+  - **Snapshot Resolution**: Configurable via `SnapshotConfig`
 
 - **DuckLakeSchema** (`schema.rs`):
   - `table_names()`: Queries `list_tables()` on every call
@@ -92,12 +94,12 @@ The catalog uses a **pure dynamic lookup** approach with no caching at the catal
 **Benefits**:
 - O(1) memory usage regardless of catalog size
 - Fast catalog startup (no upfront schema/table listing)
-- Always fresh metadata (no stale cache issues)
-- Simple implementation (no cache invalidation logic)
+- Configurable freshness vs performance trade-off
+- Simple implementation (no complex cache invalidation logic)
 
 **Trade-offs**:
 - Small query overhead per metadata lookup (acceptable for read-only DuckDB connections)
-- Future optimization: Add optional caching layer via wrapper implementation
+- Snapshot resolution adds one SQL query per catalog operation (configurable via TTL)
 
 ### Data Flow
 

examples/basic_query.rs

@@ -2,8 +2,9 @@
 //!
 //! This example demonstrates how to:
 //! 1. Create a DuckLake catalog from a DuckDB catalog file
-//! 2. Register it with DataFusion
-//! 3. Execute a simple SELECT query
+//! 2. Configure snapshot resolution with TTL (time-to-live)
+//! 3. Register it with DataFusion
+//! 4. Execute a simple SELECT query
 //!
 //! To run this example, you need:
 //! - A DuckDB database file with DuckLake tables
@@ -14,6 +15,8 @@
 use datafusion::execution::runtime_env::RuntimeEnv;
 use datafusion::prelude::*;
 use datafusion_ducklake::{DuckLakeCatalog, DuckdbMetadataProvider};
+// Uncomment when using custom snapshot config:
+// use datafusion_ducklake::SnapshotConfig;
 use object_store::ObjectStore;
 use object_store::aws::AmazonS3Builder;
 use std::env;
@@ -56,9 +59,20 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
     );
     runtime.register_object_store(&Url::parse("s3://ducklake-data/")?, s3);
 
-    // Create the DuckLake catalog
+    // Configure snapshot resolution behavior
+    //
+    // Option 1: Default configuration (TTL=0) - Always fresh, queries snapshot on every access
     let ducklake_catalog = DuckLakeCatalog::new(provider)?;
 
+    // Option 2: Custom TTL - Balance freshness and performance
+    // Caches snapshot for 5 seconds, then refreshes
+    // let config = SnapshotConfig { ttl_seconds: Some(5) };
+    // let ducklake_catalog = DuckLakeCatalog::new_with_config(provider, config)?;
+
+    // Option 3: Cache forever - Maximum performance, snapshot frozen at catalog creation
+    // let config = SnapshotConfig { ttl_seconds: None };
+    // let ducklake_catalog = DuckLakeCatalog::new_with_config(provider, config)?;
+
     println!("✓ Connected to DuckLake catalog");
 
     let config = SessionConfig::new().with_default_catalog_and_schema("ducklake", "main");