Fix truncateLsn bug

libs/walproposer/src/simtest/safekeeper.rs

@@ -16,7 +16,7 @@ use safekeeper::{
     timeline::TimelineError,
     SafeKeeperConf,
 };
-use tracing::debug;
+use tracing::{debug, info_span};
 use utils::{
     id::{NodeId, TenantId, TenantTimelineId, TimelineId},
     lsn::Lsn,
@@ -147,7 +147,8 @@ impl GlobalMap {
 }
 
 pub fn run_server(os: NodeOs, disk: Arc<Disk>) -> Result<()> {
-    debug!("started server {}", os.id());
+    let _enter = info_span!("safekeeper", id = os.id()).entered();
+    debug!("started server");
     let conf = SafeKeeperConf {
         workdir: PathBuf::from("."),
         my_id: NodeId(os.id() as u64),

libs/walproposer/src/simtest/wp_sk.rs

@@ -191,7 +191,7 @@ fn test_one_schedule() -> anyhow::Result<()> {
     // test.run_schedule(&schedule)?;
     // test.world.stop_all();
 
-    let seed = 14035854184686918762;
+    let seed = 16600359418126807633;
     let test = config.start(seed);
     warn!("Running test with seed {}", seed);
 

libs/walproposer/test.c

@@ -81,6 +81,7 @@ void MyContextInit() {
             exit(1);
 
         log_min_messages = FATAL;
+        // log_min_messages = LOG;
         Log_line_prefix = "[%p] ";
 
         InitializeMaxBackends();

pgxn/neon/walproposer.c

@@ -1505,6 +1505,17 @@ DetermineEpochStartLsn(void)
 		walprop_log(LOG, "bumped epochStartLsn to the first record %X/%X", LSN_FORMAT_ARGS(propEpochStartLsn));
 	}
 
+	if (truncateLsn == InvalidXLogRecPtr && timelineStartLsn == propEpochStartLsn)
+	{
+		/*
+		 * If truncateLsn is 0 everywhere, we are bootstrapping -- nothing was
+		 * committed yet. But if timelineStartLsn is not 0, we already know
+		 * the first record location, so we can bump truncateLsn to it.
+		 */
+		truncateLsn = timelineStartLsn;
+		walprop_log(LOG, "bumped truncateLsn to timelineStartLsn %X/%X", LSN_FORMAT_ARGS(truncateLsn));
+	}
+
 	/*
 	 * If propEpochStartLsn is not 0, at least one msg with WAL was sent to
 	 * some connected safekeeper; it must have carried truncateLsn pointing to