From f166330e01d50ccf820c6b1e1c2a2790b7dc047a Mon Sep 17 00:00:00 2001
From: Karl Nilsson <kjnilsson@gmail.com>
Date: Mon, 22 Jul 2024 11:45:58 +0100
Subject: [PATCH] Be more lenient during pre init.

Ra currently expects that any registered server in names.dets also
has a valid directory. As these two not updated in a transaction it is
possible for the two to diverge hence this change introduce more leniency
in that a registered server where there is no directory for is not considered
an error that will stop the ra system from starting.
---
 src/ra_log_pre_init.erl | 27 +++++++++++++++++++++------
 test/ra_log_2_SUITE.erl | 32 +++++++++++++++++++++++++++++---
 2 files changed, 50 insertions(+), 9 deletions(-)

diff --git a/src/ra_log_pre_init.erl b/src/ra_log_pre_init.erl
index dc04fcc2..366757d8 100644
--- a/src/ra_log_pre_init.erl
+++ b/src/ra_log_pre_init.erl
@@ -42,7 +42,7 @@ init([System]) ->
     ?INFO("ra system '~ts' running pre init for ~b registered servers",
           [System, length(Regd)]),
     _ = [begin
-             try pre_init(System, Name, UId) of
+             try pre_init(System, UId) of
                  ok -> ok
              catch _:Err ->
                        ?ERROR("pre_init failed in system ~s for UId ~ts with name ~ts"
@@ -73,7 +73,7 @@ code_change(_OldVsn, State, _Extra) ->
 %%% Internal functions
 %%%===================================================================
 
-pre_init(System, Name, UId) ->
+pre_init(System, UId) ->
     case ets:lookup(?ETSTBL, UId) of
         [{_, _}] ->
             %% already initialised
@@ -83,10 +83,25 @@ pre_init(System, Name, UId) ->
                 undefined ->
                     {error, system_not_started};
                 SysCfg ->
-                    {ok, #{log_init_args := Log}} =
-                        ra_server_sup_sup:recover_config(System, Name),
-                    ok = ra_log:pre_init(Log#{system_config => SysCfg}),
-                    ok
+                    %% check if the server dir exists, if not
+                    %% then just log and return instead of failing.
+                    Dir = ra_env:server_data_dir(System, UId),
+                    case ra_lib:is_dir(Dir) of
+                        true ->
+                            case ra_log:read_config(Dir) of
+                                {ok, #{log_init_args := Log}} ->
+                                    ok = ra_log:pre_init(Log#{system_config => SysCfg}),
+                                    ok;
+                                {error, Err} ->
+                                    ?ERROR("pre_init failed to read config file for UId '~ts', Err ~p",
+                                           [UId, Err]),
+                                    exit({pre_init_failed, Err})
+                            end;
+                        false ->
+                            ?INFO("pre_init UId '~ts' is registered but no data directory was found",
+                                  [UId]),
+                            ok
+                    end
             end
     end.
 
diff --git a/test/ra_log_2_SUITE.erl b/test/ra_log_2_SUITE.erl
index 19bf6d1e..a03ff2c4 100644
--- a/test/ra_log_2_SUITE.erl
+++ b/test/ra_log_2_SUITE.erl
@@ -31,7 +31,8 @@ all_tests() ->
      last_index_reset,
      last_index_reset_before_written,
      recovery,
-     recover_bigly,
+     recover_many,
+     recovery_with_missing_directory,
      wal_crash_recover,
      wal_down_read_availability,
      wal_down_append_throws,
@@ -74,6 +75,11 @@ end_per_suite(Config) ->
 
 init_per_group(G, Config) ->
     DataDir = filename:join(?config(priv_dir, Config), G),
+    ra_env:configure_logger(logger),
+    LogFile = filename:join(DataDir, "ra.log"),
+    logger:set_primary_config(level, debug),
+    logger:add_handler(ra_handler, logger_std_h,
+                       #{config => #{file => LogFile}}),
     [{access_pattern, G},
      {work_dir, DataDir}
      | Config].
@@ -83,7 +89,6 @@ end_per_group(_, Config) ->
 
 init_per_testcase(TestCase, Config) ->
     ok = start_ra(Config),
-    ra_env:configure_logger(logger),
     DataDir = ?config(work_dir, Config),
     UId = <<(atom_to_binary(TestCase, utf8))/binary,
             (atom_to_binary(?config(access_pattern, Config)))/binary>>,
@@ -609,7 +614,7 @@ recovery(Config) ->
 
     ok.
 
-recover_bigly(Config) ->
+recover_many(Config) ->
     Log0 = ra_log_init(Config),
     Log1 = write_n(1, 10000, 1, Log0),
     Pred = fun (L) ->
@@ -626,6 +631,27 @@ recover_bigly(Config) ->
     ra_log:close(Log),
     ok.
 
+recovery_with_missing_directory(Config) ->
+    %% checking that the ra system can be restarted even if a directory
+    %% has been deleted with a ra_directory entry still in place.
+    logger:set_primary_config(level, debug),
+    UId = ?config(uid, Config),
+    Log0 = ra_log_init(Config),
+    ra_log:close(Log0),
+
+    ServerDataDir = ra_env:server_data_dir(default, UId),
+    ok = ra_lib:recursive_delete(ServerDataDir),
+    ?assertNot(filelib:is_dir(ServerDataDir)),
+
+    application:stop(ra),
+    start_ra(Config),
+
+    Log5 = ra_log_init(Config),
+    ra_log:close(Log5),
+    ok = ra_lib:recursive_delete(ServerDataDir),
+    ?assertNot(filelib:is_dir(ServerDataDir)),
+
+    ok.
 
 resend_write(Config) ->
     % logger:set_primary_config(level, debug),