From 75e91684d434b9fea12d822032146e523beb62d0 Mon Sep 17 00:00:00 2001
From: Amaury Chamayou <amchamay@microsoft.com>
Date: Tue, 10 Sep 2024 13:42:50 +0000
Subject: [PATCH] Test and fix recovery with snapshot without ledger

---
 src/host/ledger.h | 10 +++++++++-
 tests/recovery.py | 37 ++++++++++++++++++++++++++++++++++---
 2 files changed, 43 insertions(+), 4 deletions(-)

diff --git a/src/host/ledger.h b/src/host/ledger.h
index 37ecd504f3e8..f5fb4645c0f1 100644
--- a/src/host/ledger.h
+++ b/src/host/ledger.h
@@ -1392,8 +1392,16 @@ namespace asynchost
 
       LOG_DEBUG_FMT("Ledger truncate: {}/{}", idx, last_idx);
 
-      if (idx >= last_idx || idx < committed_idx)
+      // Conservative check to avoid truncating to future indices, or dropping
+      // committed entries. If the ledger is being initialised from a snapshot
+      // alone, the first truncation effectively sets the last index.
+      if (last_idx != 0 && (idx >= last_idx || idx < committed_idx))
       {
+        LOG_DEBUG_FMT(
+          "Ignoring truncate to {} - last_idx: {}, committed_idx: {}",
+          idx,
+          last_idx,
+          committed_idx);
         return;
       }
 
diff --git a/tests/recovery.py b/tests/recovery.py
index 72040d543b8f..2f170262b493 100644
--- a/tests/recovery.py
+++ b/tests/recovery.py
@@ -45,7 +45,7 @@ def get_and_verify_historical_receipt(network, ref_msg):
 
 @reqs.description("Recover a service")
 @reqs.recover(number_txs=2)
-def test_recover_service(network, args, from_snapshot=True):
+def test_recover_service(network, args, from_snapshot=True, no_ledger=False):
     network.save_service_identity(args)
     old_primary, _ = network.find_primary()
 
@@ -71,7 +71,11 @@ def test_recover_service(network, args, from_snapshot=True):
 
     watcher.wait_for_recovery()
 
-    current_ledger_dir, committed_ledger_dirs = old_primary.get_ledger()
+    if no_ledger:
+        current_ledger_dir = None
+        committed_ledger_dirs = None
+    else:
+        current_ledger_dir, committed_ledger_dirs = old_primary.get_ledger()
 
     with tempfile.NamedTemporaryFile(mode="w+") as node_data_tf:
         start_node_data = {"this is a": "recovery node"}
@@ -116,7 +120,7 @@ def test_recover_service(network, args, from_snapshot=True):
         args.initial_service_cert_validity_days
     )
 
-    new_nodes = recovered_network.find_primary_and_any_backup()
+    new_nodes = recovered_network.get_joined_nodes()
     for n in new_nodes:
         with n.client() as c:
             r = c.get("/node/service/previous_identity")
@@ -811,6 +815,26 @@ def run(args):
     )
 
 
+def run_recover_snapshot_alone(args):
+    """
+    Recover a service from a snapshot alone, without any ledger files from a previous service.
+    """
+    txs = app.LoggingTxs("user0")
+    with infra.network.network(
+        args.nodes,
+        args.binary_dir,
+        args.debug_nodes,
+        args.perf_nodes,
+        pdb=args.pdb,
+        txs=txs,
+    ) as network:
+        network.start_and_open(args)
+        primary, _ = network.find_primary()
+        # Recover node solely from snapshot
+        test_recover_service(network, args, from_snapshot=True, no_ledger=True)
+        return network
+
+
 if __name__ == "__main__":
 
     def add(parser):
@@ -860,4 +884,11 @@ def add(parser):
         snapshot_tx_interval=1000000,
     )
 
+    cr.add(
+        "recovery_snapshot_alone",
+        run_recover_snapshot_alone,
+        package="samples/apps/logging/liblogging",
+        nodes=infra.e2e_args.min_nodes(cr.args, f=0),  # 1 node suffices for recovery
+    )
+
     cr.run()