adapter: fix panic in EXPLAIN FILTER PUSHDOWN FOR MATERIALIZED VIEW

src/adapter/src/coord/sequencer/inner/create_materialized_view.rs

@@ -1008,8 +1008,16 @@ impl Coordinator {
         // We don't have any way to "duplicate" the read hold of the actual collection, which we
         // obtain below... but the current implementation of read holds guarantees that the storage
         // holds we obtain here will not be any greater than the hold we actually want.
-        let read_holds =
-            Some(self.acquire_read_holds(&dataflow_import_id_bundle(&plan, mview.cluster_id)));
+        //
+        // We hold only the plan's storage imports, whose persist part stats the pushdown
+        // explanation reads. We must not acquire holds on the plan's compute imports: an index
+        // that the materialized view was planned against can be dropped while the view keeps
+        // running, and a dropped index no longer has a compute collection to hold.
+        let id_bundle = CollectionIdBundle {
+            storage_ids: plan.source_imports.keys().copied().collect(),
+            compute_ids: BTreeMap::new(),
+        };
+        let read_holds = Some(self.acquire_read_holds(&id_bundle));
 
         let frontiers = self
             .controller

test/sqllogictest/explain/pushdown.slt

@@ -136,6 +136,27 @@ EXPLAIN FILTER PUSHDOWN FOR MATERIALIZED VIEW big_numbers
 ----
 materialize.public.numbers  1313  0  1  0
 
+# Regression test for a panic (SQL-519): EXPLAIN FILTER PUSHDOWN FOR
+# MATERIALIZED VIEW panicked with "missing compute collection" when the view's
+# plan imported an index that was dropped after the view was created. The
+# dropped index no longer has a compute collection, but the view's dataflow
+# keeps running and reading from it.
+
+statement ok
+CREATE INDEX numbers_value_idx ON numbers (value);
+
+statement ok
+CREATE MATERIALIZED VIEW indexed_numbers AS SELECT * FROM numbers WHERE value > 10000;
+
+statement ok
+DROP INDEX numbers_value_idx;
+
+# The view's plan reads entirely from the (dropped) index, so there are no
+# storage imports to report stats for.
+query TIIII
+EXPLAIN FILTER PUSHDOWN FOR MATERIALIZED VIEW indexed_numbers
+----
+
 # EXPLAIN FILTER PUSHDOWN should work even if there are no replicas.
 statement ok
 CREATE CLUSTER no_replicas (SIZE 'scale=1,workers=1', REPLICATION FACTOR 0);