From 036d3287b3235f391acde55be2dd0d3193979bc8 Mon Sep 17 00:00:00 2001
From: Ferenc Szili <ferenc.szili@scylladb.com>
Date: Tue, 26 Nov 2024 16:23:19 +0100
Subject: [PATCH 1/2] database: correctly save replay position for truncate

This commit fixes a problem with way truncate saves commit log
replay positions. On shards without mutations, truncate would save the
replay position into system.truncated with shard number 0 regardless of
the actual shard number that the replay position was saved for.
---
 replica/database.cc | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/replica/database.cc b/replica/database.cc
index ff01efa3d7..71bc03a0f8 100644
--- a/replica/database.cc
+++ b/replica/database.cc
@@ -2512,7 +2512,18 @@ future<> database::truncate(db::system_keyspace& sys_ks, column_family& cf, cons
     // smaller than low_mark.
     SCYLLA_ASSERT(!st.did_flush || rp == db::replay_position() || (truncated_at <= st.low_mark_at ? rp <= st.low_mark : st.low_mark <= rp));
     if (rp == db::replay_position()) {
-        rp = st.low_mark;
+        // If this shard had no mutations, st.low_mark will be an empty, default constructed
+        // replay_position. This is a problem because an empty replay_position has the shard_id
+        // part of segment_id set to 0, even though we may be running on a shard other than
+        // shard 0. In that case, we will save the empty low_mark as a replay position into
+        // system.truncated with an incorrect shard number, which could overwrite the replay
+        // position of the actual shard 0. So, we fix the problem by creating a replay position
+        // with the correct shard_id and 0 for base_id and position
+        if (st.low_mark == db::replay_position()) {
+            rp = db::replay_position(this_shard_id(), 0, 0);
+        } else {
+            rp = st.low_mark;
+        }
     }
     co_await coroutine::parallel_for_each(cf.views(), [this, &sys_ks, truncated_at] (view_ptr v) -> future<> {
         auto& vcf = find_column_family(v);

From e54c07ba7580ccfd7cfdd61010c1b659b9ba6ac7 Mon Sep 17 00:00:00 2001
From: Ferenc Szili <ferenc.szili@scylladb.com>
Date: Thu, 28 Nov 2024 17:20:50 +0100
Subject: [PATCH 2/2] test: add test for truncate saving replay positions

This change adds a test for truncate correctly saving commit log replay
positions.
---
 test/boost/database_test.cc | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/test/boost/database_test.cc b/test/boost/database_test.cc
index 15c1b007f1..e30c5ed2d7 100644
--- a/test/boost/database_test.cc
+++ b/test/boost/database_test.cc
@@ -196,6 +196,28 @@ SEASTAR_TEST_CASE(test_truncate_without_snapshot_during_writes) {
     }, cfg);
 }
 
+// Reproducer for:
+//   https://github.com/scylladb/scylla/issues/21719
+SEASTAR_TEST_CASE(test_truncate_saves_replay_position) {
+    auto cfg = make_shared<db::config>();
+    cfg->auto_snapshot.set(false);
+    return do_with_cql_env_and_compaction_groups([] (cql_test_env& e) {
+        BOOST_REQUIRE_GT(smp::count, 1);
+        const sstring ks_name = "ks";
+        const sstring cf_name = "cf";
+        e.execute_cql(fmt::format("CREATE TABLE {}.{} (k TEXT PRIMARY KEY, v INT);", ks_name, cf_name)).get();
+        const table_id uuid = e.local_db().find_uuid(ks_name, cf_name);
+
+        replica::database::truncate_table_on_all_shards(e.db(), e.get_system_keyspace(), ks_name, cf_name, db_clock::now(), false /* with_snapshot */).get();
+
+        auto res = e.execute_cql(fmt::format("SELECT * FROM system.truncated WHERE table_uuid = {}", uuid)).get();
+        auto rows = dynamic_pointer_cast<cql_transport::messages::result_message::rows>(res);
+        BOOST_REQUIRE(rows);
+        auto row_count = rows->rs().result_set().size();
+        BOOST_REQUIRE_EQUAL(row_count, smp::count);
+    }, cfg);
+}
+
 SEASTAR_TEST_CASE(test_querying_with_limits) {
     return do_with_cql_env_and_compaction_groups([](cql_test_env& e) {
             // FIXME: restore indent.