finishing touches

crates/re_arrow_store/src/store.rs

@@ -23,8 +23,13 @@ pub struct DataStoreConfig {
     /// to a specific timeline _and_ a specific entity.
     ///
     /// This effectively puts an upper bound on the number of rows that need to be sorted when an
-    /// indexed bucket gets out of order.
+    /// indexed bucket gets out of order (e.g. because of new insertions or a GC pass).
     /// This is a tradeoff: less rows means faster sorts at the cost of more metadata overhead.
+    /// In particular:
+    /// - Query performance scales inversely logarithmically to this number (i.e. it gets better
+    ///   the higher this number gets).
+    /// - GC performance scales quadratically with this number (i.e. it gets better the lower this
+    ///   number gets).
     ///
     /// See [`Self::DEFAULT`] for defaults.
     pub indexed_bucket_num_rows: u64,

crates/re_arrow_store/src/store_gc.rs

@@ -61,6 +61,8 @@ impl DataStore {
     //
     // TODO(#1803): The GC should be aware of latest-at semantics and make sure they are upheld
     // when purging data.
+    //
+    // TODO(#1823): Workload specific optimizations.
     pub fn gc(&mut self, target: GarbageCollectionTarget) -> (Vec<RowId>, DataStoreStats) {
         crate::profile_function!();
 

crates/re_arrow_store/src/store_write.rs

@@ -244,9 +244,10 @@ impl MetadataRegistry<TimePoint> {
             std::collections::btree_map::Entry::Occupied(mut entry) => {
                 let entry = entry.get_mut();
                 for (timeline, time) in timepoint {
-                    let overwritten = entry.insert(timeline, time).is_some();
-                    re_log::error!(%row_id, ?timeline, ?time, "detected re-used `RowId/Timeline` pair, this is illegal and will lead to undefined behavior in the datastore");
-                    debug_assert!(!overwritten);
+                    if let Some(old_time) = entry.insert(timeline, time) {
+                        re_log::error!(%row_id, ?timeline, old_time = ?old_time, new_time = ?time, "detected re-used `RowId/Timeline` pair, this is illegal and will lead to undefined behavior in the datastore");
+                        debug_assert!(false, "detected re-used `RowId/Timeline`");
+                    }
                 }
             }
         }