rabbit_db_queue: Transactionally delete transient queues from Khepri

The prior code skirted transactions because the filter function might
cause Khepri to call itself. We want to use the same idea as the old
code - get all queues, filter them, then delete them - but we want to
perform the deletion in a transaction and fail the transaction if any
queues changed since we read them.

This fixes a bug - that the call to `delete_in_khepri/2` could return
an error tuple that would be improperly recognized as `Deletions` -
but should also make deleting transient queues atomic and fast.
Each call to `delete_in_khepri/2` needed to wait on Ra to replicate
because the deletion is an individual command sent from one process.
Performing all deletions at once means we only need to wait for one
command to be replicated across the cluster.

We also bubble up any errors to delete now rather than storing them as
deletions. This fixes a crash that occurs on node down when Khepri is
in a minority.

(cherry picked from commit 0dd26f0)
(cherry picked from commit 0f90906)

# Conflicts:
#	deps/rabbit_common/src/rabbit_misc.erl

Author: the-mikedavis

deps/rabbit/src/rabbit_db_queue.erl

@@ -1027,7 +1027,8 @@ set_many_in_khepri(Qs) ->
       Queue :: amqqueue:amqqueue(),
       FilterFun :: fun((Queue) -> boolean()),
       QName :: rabbit_amqqueue:name(),
-      Ret :: {[QName], [Deletions :: rabbit_binding:deletions()]}.
+      Ret :: {[QName], [Deletions :: rabbit_binding:deletions()]}
+             | rabbit_khepri:timeout_error().
 %% @doc Deletes all transient queues that match `FilterFun'.
 %%
 %% @private
@@ -1088,26 +1089,59 @@ delete_transient_in_khepri(FilterFun) ->
     %% process might call itself. Instead we can fetch all of the transient
     %% queues with `get_many' and then filter and fold the results outside of
     %% Khepri's Ra server process.
-    case rabbit_khepri:get_many(PathPattern) of
-        {ok, Qs} ->
-            Items = maps:fold(
-                     fun(Path, Queue, Acc) when ?is_amqqueue(Queue) ->
-                             case FilterFun(Queue) of
-                                 true ->
-                                     QueueName = khepri_queue_path_to_name(
-                                                   Path),
-                                     case delete_in_khepri(QueueName, false) of
-                                         ok ->
-                                             Acc;
-                                         Deletions ->
-                                             [{QueueName, Deletions} | Acc]
-                                     end;
-                                 false ->
-                                     Acc
-                             end
-                     end, [], Qs),
-            {QueueNames, Deletions} = lists:unzip(Items),
-            {QueueNames, lists:flatten(Deletions)};
+    case rabbit_khepri:adv_get_many(PathPattern) of
+        {ok, Props} ->
+            Qs = maps:fold(
+                   fun(Path0, #{data := Q, payload_version := Vsn}, Acc)
+                       when ?is_amqqueue(Q) ->
+                           case FilterFun(Q) of
+                               true ->
+                                   Path = khepri_path:combine_with_conditions(
+                                            Path0,
+                                            [#if_payload_version{version = Vsn}]),
+                                   QName = amqqueue:get_name(Q),
+                                   [{Path, QName} | Acc];
+                               false ->
+                                   Acc
+                           end
+                   end, [], Props),
+            do_delete_transient_queues_in_khepri(Qs, FilterFun);
+        {error, _} = Error ->
+            Error
+    end.
+
+do_delete_transient_queues_in_khepri([], _FilterFun) ->
+    %% If there are no changes to make, avoid performing a transaction. When
+    %% Khepri is in a minority this avoids a long timeout waiting for the
+    %% transaction command to be processed. Otherwise it avoids appending a
+    %% somewhat large transaction command to Khepri's log.
+    {[], []};
+do_delete_transient_queues_in_khepri(Qs, FilterFun) ->
+    Res = rabbit_khepri:transaction(
+            fun() ->
+                    rabbit_misc:fold_while_ok(
+                      fun({Path, QName}, Acc) ->
+                              %% Also see `delete_in_khepri/2'.
+                              case khepri_tx_adv:delete(Path) of
+                                  {ok, #{data := _}} ->
+                                      Deletions = rabbit_db_binding:delete_for_destination_in_khepri(
+                                                    QName, false),
+                                      {ok, [{QName, Deletions} | Acc]};
+                                  {ok, _} ->
+                                      {ok, Acc};
+                                  {error, _} = Error ->
+                                      Error
+                              end
+                      end, [], Qs)
+            end),
+    case Res of
+        {ok, Items} ->
+            {QNames, Deletions} = lists:unzip(Items),
+            {QNames, lists:flatten(Deletions)};
+        {error, {khepri, mismatching_node, _}} ->
+            %% One of the queues changed while attempting to update all
+            %% queues. Retry the operation.
+            delete_transient_in_khepri(FilterFun);
         {error, _} = Error ->
             Error
     end.
@@ -1382,6 +1416,3 @@ khepri_queues_path() ->
 
 khepri_queue_path(#resource{virtual_host = VHost, name = Name}) ->
     [?MODULE, queues, VHost, Name].
-
-khepri_queue_path_to_name([?MODULE, queues, VHost, Name]) ->
-    rabbit_misc:r(VHost, queue, Name).

deps/rabbit_common/src/rabbit_misc.erl

@@ -88,6 +88,10 @@
          maps_put_falsy/3
         ]).
 -export([remote_sup_child/2]).
+<<<<<<< HEAD
+=======
+-export([for_each_while_ok/2, fold_while_ok/3]).
+>>>>>>> 0f90906254 (rabbit_db_queue: Transactionally delete transient queues from Khepri)
 
 %% Horrible macro to use in guards
 -define(IS_BENIGN_EXIT(R),
@@ -1621,3 +1625,49 @@ remote_sup_child(Node, Sup) ->
         []                              -> {error, no_child};
         {badrpc, {'EXIT', {noproc, _}}} -> {error, no_sup}
     end.
+<<<<<<< HEAD
+=======
+
+-spec for_each_while_ok(ForEachFun, List) -> Ret when
+      ForEachFun :: fun((Element) -> ok | {error, ErrReason}),
+      ErrReason :: any(),
+      Element :: any(),
+      List :: [Element],
+      Ret :: ok | {error, ErrReason}.
+%% @doc Calls the given `ForEachFun' for each element in the given `List',
+%% short-circuiting if the function returns `{error,_}'.
+%%
+%% @returns the first `{error,_}' returned by `ForEachFun' or `ok' if
+%% `ForEachFun' never returns an error tuple.
+
+for_each_while_ok(Fun, [Elem | Rest]) ->
+    case Fun(Elem) of
+        ok ->
+            for_each_while_ok(Fun, Rest);
+        {error, _} = Error ->
+            Error
+    end;
+for_each_while_ok(_, []) ->
+    ok.
+
+-spec fold_while_ok(FoldFun, Acc, List) -> Ret when
+      FoldFun :: fun((Element, Acc) -> {ok, Acc} | {error, ErrReason}),
+      Element :: any(),
+      List :: Element,
+      Ret :: {ok, Acc} | {error, ErrReason}.
+%% @doc Calls the given `FoldFun' on each element of the given `List' and the
+%% accumulator value, short-circuiting if the function returns `{error,_}'.
+%%
+%% @returns the first `{error,_}' returned by `FoldFun' or `{ok,Acc}' if
+%% `FoldFun' never returns an error tuple.
+
+fold_while_ok(Fun, Acc0, [Elem | Rest]) ->
+    case Fun(Elem, Acc0) of
+        {ok, Acc} ->
+            fold_while_ok(Fun, Acc, Rest);
+        {error, _} = Error ->
+            Error
+    end;
+fold_while_ok(_Fun, Acc, []) ->
+    {ok, Acc}.
+>>>>>>> 0f90906254 (rabbit_db_queue: Transactionally delete transient queues from Khepri)