Avoid loading files when constructing tables

This provides a performance benefit and this setting is _largely_ used
by `EagerSnapshot` to do some redundant work when opening tables.
`EagerSnapshot` is being whittled down to nothing with everything being
a pass-through onto the underlying kernel in future releases.

This has been tested with development workloads for oxbow and sqs-ingest
with neutral or positive performance changes

Author: rtyler

crates/oxbow/src/lib.rs

@@ -85,7 +85,7 @@ pub async fn convert(
         Err(e) => {
             info!("No Delta table at {}: {:?}", location, e);
             let store = logstore_for(
-                &location,
+                location,
                 StorageConfig::parse_options(storage_options.unwrap_or_default())?,
             )?;
             let files = discover_parquet_files(store.object_store(None).clone()).await?;

crates/oxbow/src/lock.rs

@@ -1,3 +1,4 @@
+use deltalake::DeltaTableBuilder;
 ///
 /// THe lock module contains some simple helpers for handling table locks in DynamoDB. This is
 /// something required of deltalake 0.16.x and earlier.
@@ -12,11 +13,12 @@ use std::collections::HashMap;
 ///Wrapper aroudn [deltalake::open_table] which will open the table with the appropriate storage
 ///options needed for locking
 pub async fn open_table(table_uri: &str) -> deltalake::DeltaResult<deltalake::DeltaTable> {
-    deltalake::open_table_with_storage_options(
-        Url::parse(table_uri).expect("Fail"),
-        storage_options(table_uri),
-    )
-    .await
+    let table_url = Url::parse(table_uri).expect("Fatal error trying to parse a table URL");
+    DeltaTableBuilder::from_url(table_url)?
+        .without_files()
+        .with_storage_options(storage_options(table_uri))
+        .load()
+        .await
 }
 
 /// Default storage options for using with `deltalake` calls
@@ -83,3 +85,18 @@ pub async fn release(
     }
     true
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use deltalake::DeltaResult;
+
+    #[tokio::test]
+    async fn test_open_table() -> DeltaResult<()> {
+        let _table = open_table("memory://")
+            .await
+            .expect_err("Can't possibly load a table that doesn't exist!");
+        Ok(())
+    }
+}