Add test for duplicate layers

pageserver/src/tenant/layer_map.rs

@@ -56,6 +56,7 @@ use std::collections::VecDeque;
 use std::ops::Range;
 use std::sync::Arc;
 use utils::lsn::Lsn;
+use tracing::*;
 
 use historic_layer_coverage::BufferedHistoricLayerCoverage;
 pub use historic_layer_coverage::Replacement;
@@ -275,11 +276,14 @@ where
     ///
     pub(self) fn insert_historic_noflush(&mut self, layer: Arc<L>) {
         // TODO: See #3869, resulting #4088, attempted fix and repro #4094
-        self.historic.insert(
-            historic_layer_coverage::LayerKey::from(&*layer),
-            Arc::clone(&layer),
-        );
-
+        let key = historic_layer_coverage::LayerKey::from(&*layer);
+        if self.historic.contains(&key) {
+            warn!(
+                "Attempt to insert duplicate layer {} in layer map",
+                layer.short_id()
+            );
+        }
+        self.historic.insert(key, Arc::clone(&layer));
         if Self::is_l0(&layer) {
             self.l0_delta_layers.push(layer);
         }

pageserver/src/tenant/layer_map/historic_layer_coverage.rs

@@ -417,7 +417,15 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
         }
     }
 
-    pub fn insert(&mut self, layer_key: LayerKey, value: Value) {
+    pub fn contains(&self, layer_key: &LayerKey) -> bool {
+        match self.buffer.get(layer_key) {
+            Some(None) => false,                         // layer remove was buffered
+            Some(_) => true,                             // layer insert was buffered
+            None => self.layers.contains_key(layer_key), // no buffered ops for this layer
+        }
+    }
+
+	pub fn insert(&mut self, layer_key: LayerKey, value: Value) {
         self.buffer.insert(layer_key, Some(value));
     }
 

test_runner/performance/test_duplicate_layers.py

@@ -0,0 +1,42 @@
+import time
+
+import pytest
+from fixtures.neon_fixtures import NeonEnvBuilder, PgBin
+
+
+# Test duplicate layer detection
+#
+# This test sets fail point at the end of first compaction phase:
+# after flushing new L1 layers but before deletion of L0 layes
+# It should cause generation of duplicate L1 layer by compaction after restart
+@pytest.mark.timeout(600)
+def test_duplicate_layers(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin):
+    env = neon_env_builder.init_start()
+
+    # These warnings are expected, when the pageserver is restarted abruptly
+    env.pageserver.allowed_errors.append(".*found future image layer.*")
+    env.pageserver.allowed_errors.append(".*found future delta layer.*")
+    env.pageserver.allowed_errors.append(".*duplicate layer.*")
+
+    pageserver_http = env.pageserver.http_client()
+
+    # Use aggressive compaction and checkpoint settings
+    tenant_id, _ = env.neon_cli.create_tenant(
+        conf={
+            "checkpoint_distance": f"{1024 ** 2}",
+            "compaction_target_size": f"{1024 ** 2}",
+            "compaction_period": "1 s",
+            "compaction_threshold": "3",
+        }
+    )
+    endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
+    connstr = endpoint.connstr(options="-csynchronous_commit=off")
+    pg_bin.run_capture(["pgbench", "-i", "-s10", connstr])
+
+    pageserver_http.configure_failpoints(("compact-level0-phase1-finish", "exit"))
+
+    with pytest.raises(Exception):
+        pg_bin.run_capture(["pgbench", "-P1", "-N", "-c5", "-T500", "-Mprepared", connstr])
+    env.pageserver.stop()
+    env.pageserver.start()
+    time.sleep(10)  # let compaction to be performed