Clarify in documentation that cache is only for async decoder

parquet/src/arrow/arrow_reader/mod.rs

@@ -344,14 +344,20 @@ impl<T> ArrowReaderBuilder<T> {
         Self { metrics, ..self }
     }
 
-    /// Set the maximum size (per row group) of the predicate cache in bytes.
+    /// Set the maximum size (per row group) of the predicate cache in bytes for
+    /// the async decoder.
     ///
-    /// Defaults to 100MB
+    /// Defaults to 100MB (across all columns). Set to `usize::MAX` to use
+    /// unlimited cache size.
     ///
     /// This cache is used to store decoded arrays that are used in
     /// predicate evaluation ([`Self::with_row_filter`]).
     ///
-    /// Set to `usize::MAX` to use unlimited cache size.
+    /// This cache is only used for the "async" decoder, [`ParquetRecordBatchStream`]. See
+    /// [this ticket] for more details and alternatives.
+    ///
+    /// [`ParquetRecordBatchStream`]: https://docs.rs/parquet/latest/parquet/arrow/async_reader/struct.ParquetRecordBatchStream.html
+    /// [this ticket]: https://github.com/apache/arrow-rs/issues/8000
     pub fn with_max_predicate_cache_size(self, max_predicate_cache_size: usize) -> Self {
         Self {
             max_predicate_cache_size,

parquet/tests/arrow_reader/predicate_cache.rs

@@ -38,8 +38,6 @@ use std::ops::Range;
 use std::sync::Arc;
 use std::sync::LazyLock;
 
-// TODO file a ticket about the duplication here
-
 #[tokio::test]
 async fn test_default_read() {
     // The cache is not used without predicates, so we expect 0 records read from cache
@@ -51,18 +49,24 @@ async fn test_default_read() {
 }
 
 #[tokio::test]
-async fn test_cache_with_filters() {
+async fn test_async_cache_with_filters() {
     let test = ParquetPredicateCacheTest::new().with_expected_records_read_from_cache(49);
-    // TODO The sync reader does not use the cache yet....
-    // let sync_builder = test.sync_builder();
-    // let sync_builder = test.add_project_ab_and_filter_b(sync_builder);
-    // test.run_sync(sync_builder);
-
     let async_builder = test.async_builder().await;
     let async_builder = test.add_project_ab_and_filter_b(async_builder);
     test.run_async(async_builder).await;
 }
 
+#[tokio::test]
+async fn test_sync_cache_with_filters() {
+    let test = ParquetPredicateCacheTest::new()
+        // The sync reader does not use the cache. See https://github.com/apache/arrow-rs/issues/8000
+        .with_expected_records_read_from_cache(0);
+
+    let sync_builder = test.sync_builder();
+    let sync_builder = test.add_project_ab_and_filter_b(sync_builder);
+    test.run_sync(sync_builder);
+}
+
 #[tokio::test]
 async fn test_cache_disabled_with_filters() {
     // expect no records to be read from cache, because the cache is disabled