Merge branch 'main' into problame/async-walredo/better-benchmark

pageserver/src/tenant/secondary/downloader.rs

@@ -534,7 +534,11 @@ impl<'a> TenantDownloader<'a> {
             .await
             .maybe_fatal_err(&context_msg)?;
 
-        tracing::debug!("Wrote local heatmap to {}", heatmap_path);
+        tracing::debug!(
+            "Wrote local heatmap to {}, with {} timelines",
+            heatmap_path,
+            heatmap.timelines.len()
+        );
 
         // Clean up any local layers that aren't in the heatmap.  We do this first for all timelines, on the general
         // principle that deletions should be done before writes wherever possible, and so that we can use this
@@ -547,6 +551,10 @@ impl<'a> TenantDownloader<'a> {
         // Download the layers in the heatmap
         for timeline in heatmap.timelines {
             if self.secondary_state.cancel.is_cancelled() {
+                tracing::debug!(
+                    "Cancelled before downloading timeline {}",
+                    timeline.timeline_id
+                );
                 return Ok(());
             }
 
@@ -764,10 +772,13 @@ impl<'a> TenantDownloader<'a> {
             }
         };
 
+        tracing::debug!(timeline_id=%timeline.timeline_id, "Downloading layers, {} in heatmap", timeline.layers.len());
+
         // Download heatmap layers that are not present on local disk, or update their
         // access time if they are already present.
         for layer in timeline.layers {
             if self.secondary_state.cancel.is_cancelled() {
+                tracing::debug!("Cancelled -- dropping out of layer loop");
                 return Ok(());
             }
 

pageserver/src/tenant/secondary/scheduler.rs

@@ -300,6 +300,7 @@ where
 
         let tenant_shard_id = job.get_tenant_shard_id();
         let barrier = if let Some(barrier) = self.get_running(tenant_shard_id) {
+            tracing::info!("Command already running, waiting for it");
             barrier
         } else {
             let running = self.spawn_now(job);

test_runner/regress/test_pageserver_secondary.py

@@ -432,6 +432,10 @@ def test_secondary_downloads(neon_env_builder: NeonEnvBuilder):
      - Eviction of layers on the attached location results in deletion
        on the secondary location as well.
     """
+
+    # For debug of https://github.com/neondatabase/neon/issues/6966
+    neon_env_builder.rust_log_override = "DEBUG"
+
     neon_env_builder.num_pageservers = 2
     neon_env_builder.enable_pageserver_remote_storage(
         remote_storage_kind=RemoteStorageKind.MOCK_S3,