test that reinserted collections compact

Author: rescrv

chromadb/test/property/test_add.py

@@ -21,6 +21,31 @@
 
 collection_st = st.shared(strategies.collections(with_hnsw_params=True), key="coll")
 
+@given(
+    collection=collection_st,
+    record_set=strategies.recordsets(collection_st, min_size=1, max_size=5),
+)
+@settings(
+    deadline=None,
+    parent=override_hypothesis_profile(
+        normal=hypothesis.settings(max_examples=500),
+        fast=hypothesis.settings(max_examples=200),
+    ),
+)
+def test_add_miniscule(
+    client: ClientAPI,
+    collection: strategies.Collection,
+    record_set: strategies.RecordSet,
+) -> None:
+    if (
+        client.get_settings().chroma_api_impl
+        == "chromadb.api.async_fastapi.AsyncFastAPI"
+    ):
+        pytest.skip(
+            "TODO @jai, come back and debug why CI runners fail with async + sync"
+        )
+    _test_add(client, collection, record_set, True)
+
 
 # Hypothesis tends to generate smaller values so we explicitly segregate the
 # the tests into tiers, Small, Medium. Hypothesis struggles to generate large

rust/log-service/src/lib.rs

@@ -361,18 +361,11 @@ impl RollupPerCollection {
         self.start_log_position = witness
             .map(|x| x.cursor.position)
             .unwrap_or(LogPosition::from_offset(1));
-        if self.start_log_position > self.limit_log_position {
-            tracing::error!(
-                "invariant violation; will patch: {:?} > {:?}",
-                self.start_log_position,
-                self.limit_log_position
-            );
-        }
         self.limit_log_position = self.limit_log_position.max(self.start_log_position);
     }
 
     fn is_empty(&self) -> bool {
-        self.start_log_position == self.limit_log_position
+        self.start_log_position >= self.limit_log_position
     }
 
     fn dirty_marker(&self, collection_id: CollectionUuid) -> DirtyMarker {
@@ -905,7 +898,7 @@ impl LogServer {
                 rollup.start_log_position,
                 LogPosition::from_offset(adjusted_log_offset as u64),
             );
-            if rollup.start_log_position >= rollup.limit_log_position {
+            if rollup.is_empty() {
                 entry.remove();
             }
         }
@@ -1571,7 +1564,7 @@ impl LogServer {
                     format!("max_offset={:?} < offset={:?}", max_offset, offset),
                 ));
             }
-            if offset != max_offset{
+            if offset != max_offset {
                 let mark_dirty = MarkDirty {
                     collection_id: target_collection_id,
                     dirty_log: Arc::clone(&self.dirty_log),