Merge pull request #11090 from neondatabase/vlad/release-gate-previous-heatmap

Storage release 2025-03-05

libs/pageserver_api/src/config.rs

@@ -123,6 +123,10 @@ pub struct ConfigToml {
     pub enable_read_path_debugging: Option<bool>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub validate_wal_contiguity: Option<bool>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub load_previous_heatmap: Option<bool>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub generate_unarchival_heatmap: Option<bool>,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -523,6 +527,8 @@ impl Default for ConfigToml {
                 None
             },
             validate_wal_contiguity: None,
+            load_previous_heatmap: None,
+            generate_unarchival_heatmap: None,
         }
     }
 }

pageserver/src/config.rs

@@ -194,6 +194,13 @@ pub struct PageServerConf {
     /// Interpreted protocol feature: if enabled, validate that the logical WAL received from
     /// safekeepers does not have gaps.
     pub validate_wal_contiguity: bool,
+
+    /// When set, the previously written to disk heatmap is loaded on tenant attach and used
+    /// to avoid clobbering the heatmap from new, cold, attached locations.
+    pub load_previous_heatmap: bool,
+
+    /// When set, include visible layers in the next uploaded heatmaps of an unarchived timeline.
+    pub generate_unarchival_heatmap: bool,
 }
 
 /// Token for authentication to safekeepers
@@ -358,6 +365,8 @@ impl PageServerConf {
             get_vectored_concurrent_io,
             enable_read_path_debugging,
             validate_wal_contiguity,
+            load_previous_heatmap,
+            generate_unarchival_heatmap,
         } = config_toml;
 
         let mut conf = PageServerConf {
@@ -447,6 +456,8 @@ impl PageServerConf {
             no_sync: no_sync.unwrap_or(false),
             enable_read_path_debugging: enable_read_path_debugging.unwrap_or(false),
             validate_wal_contiguity: validate_wal_contiguity.unwrap_or(false),
+            load_previous_heatmap: load_previous_heatmap.unwrap_or(false),
+            generate_unarchival_heatmap: generate_unarchival_heatmap.unwrap_or(false),
         };
 
         // ------------------------------------------------------------
@@ -493,6 +504,8 @@ impl PageServerConf {
             metric_collection_interval: Duration::from_secs(60),
             synthetic_size_calculation_interval: Duration::from_secs(60),
             background_task_maximum_delay: Duration::ZERO,
+            load_previous_heatmap: Some(true),
+            generate_unarchival_heatmap: Some(true),
             ..Default::default()
         };
         PageServerConf::parse_and_validate(NodeId(0), config_toml, &repo_dir).unwrap()

pageserver/src/tenant.rs

@@ -1149,7 +1149,7 @@ impl Tenant {
         // a previous heatmap which contains all visible layers in the layer map.
         // This previous heatmap will be used whenever a fresh heatmap is generated
         // for the timeline.
-        if matches!(cause, LoadTimelineCause::Unoffload) {
+        if self.conf.generate_unarchival_heatmap && matches!(cause, LoadTimelineCause::Unoffload) {
             let mut tline_ending_at = Some((&timeline, timeline.get_last_record_lsn()));
             while let Some((tline, end_lsn)) = tline_ending_at {
                 let unarchival_heatmap = tline.generate_unarchival_heatmap(end_lsn).await;
@@ -1578,6 +1578,10 @@ impl Tenant {
     }
 
     async fn read_on_disk_heatmap(&self) -> Option<(HeatMapTenant, std::time::Instant)> {
+        if !self.conf.load_previous_heatmap {
+            return None;
+        }
+
         let on_disk_heatmap_path = self.conf.tenant_heatmap_path(&self.tenant_shard_id);
         match tokio::fs::read_to_string(on_disk_heatmap_path).await {
             Ok(heatmap) => match serde_json::from_str::<HeatMapTenant>(&heatmap) {

test_runner/fixtures/neon_fixtures.py

@@ -1164,6 +1164,8 @@ class NeonEnv:
                 # Disable pageserver disk syncs in tests: when running tests concurrently, this avoids
                 # the pageserver taking a long time to start up due to syncfs flushing other tests' data
                 "no_sync": True,
+                # Look for gaps in WAL received from safekeepeers
+                "validate_wal_contiguity": True,
             }
 
             # Batching (https://github.com/neondatabase/neon/issues/9377):
@@ -1176,11 +1178,12 @@ class NeonEnv:
 
             if config.test_may_use_compatibility_snapshot_binaries:
                 log.info(
-                    "Skipping WAL contiguity validation to avoid forward-compatibility related test failures"
+                    "Skipping prev heatmap settings to avoid forward-compatibility related test failures"
                 )
             else:
                 # Look for gaps in WAL received from safekeepeers
-                ps_cfg["validate_wal_contiguity"] = True
+                ps_cfg["load_previous_heatmap"] = True
+                ps_cfg["generate_unarchival_heatmap"] = True
 
             get_vectored_concurrent_io = self.pageserver_get_vectored_concurrent_io
             if get_vectored_concurrent_io is not None: