Produce stacktraces unless mnesia_compatible

Use latest version of emqx fork with compilation fix

.circleci/config.yml

@@ -2,15 +2,35 @@ version: 2.1
 
 executors:
   aebuilder:
+    parameters:
+      otp:
+        type: string
     docker:
-      - image: aeternity/builder
+      - image: aeternity/builder:focal-<< parameters.otp >>
         user: builder
     environment:
       ERLANG_ROCKSDB_BUILDOPTS: "-j2"
+      ERLANG_ROCKSDB_OPTS: "-DWITH_SYSTEM_ROCKSDB=ON -DWITH_SNAPPY=ON -DWITH_LZ4=ON -DWITH_BZ2=ON -DWITH_ZSTD=ON"
+
 
 jobs:
-  build:
-    executor: aebuilder
+  build_and_test:
+    parameters:
+      otp:
+        type: string
+    executor:
+      name: aebuilder
+      otp: << parameters.otp >>
     steps:
       - checkout
       - run: make test
+      - store_artifacts:
+          path: _build/test/logs
+
+workflows:
+  commit:
+    jobs:
+      - build_and_test:
+          matrix:
+            parameters:
+              otp: ["otp24", "otp25", "otp26"]

README.md

@@ -80,7 +80,7 @@ RocksDB supports a number of customization options. These can be specified
 by providing a `{Key, Value}` list named `rocksdb_opts` under `user_properties`,
 for example:
 
-```
+```erlang
 mnesia:create_table(foo, [{rocksdb_copies, [node()]},
                           ...
                           {user_properties,
@@ -93,7 +93,7 @@ for information on configuration parameters. Also see the section below on handl
 
 The default configuration for tables in `mnesia_rocksdb` is:
 
-```
+```erlang
 default_open_opts() ->
     [ {create_if_missing, true}
       , {cache_size,
@@ -131,6 +131,18 @@ depend on having an up to date size count at all times, you need to maintain
 it yourself. If you only need the size occasionally, you may traverse the
 table to count the elements.
 
+When `mrdb` transactions abort, they will return a stacktrace caught
+from within the transaction fun, giving much better debugging info.
+This is different from how mnesia does it.
+
+If behavior closer to mnesia's abort returns are needed, say, for backwards
+compatibility, this can be controlled by setting the environment variable
+`-mnesia_rocksdb mnesia_compatible_aborts true`, or by adding a transaction
+option, e.g. `mrdb:activity({tx, #{mnesia_compatible => true}}, fun() ... end)`.
+For really performance-critical transactions which may abort often, it might
+make a difference to set this option to `true`, since there is a cost involved
+in producing stacktraces.
+
 
 ### <a name="Mnesia_backend_plugins">Mnesia backend plugins</a> ###
 
@@ -195,7 +207,7 @@ our example. It returns a list of index terms.
 
 Given the following index plugin implementation:
 
-```
+```erlang
 -module(words).
 -export([words_f/3]).
 
@@ -212,7 +224,7 @@ words_(_) ->
 
 We can register the plugin and use it in table definitions:
 
-```
+```erlang
 Eshell V12.1.3  (abort with ^G)
 1> mnesia:start().
 ok
@@ -228,7 +240,7 @@ as an exported function along the node's code path.
 
 To see what happens when we insert an object, we can turn on call trace.
 
-```
+```erlang
 4> dbg:tracer().
 {ok,<0.108.0>}
 5> dbg:tp(words, x).

doc/edoc-info

@@ -2,4 +2,5 @@
 {application,mnesia_rocksdb}.
 {modules,[mnesia_rocksdb,mnesia_rocksdb_admin,mnesia_rocksdb_app,
           mnesia_rocksdb_lib,mnesia_rocksdb_params,mnesia_rocksdb_sup,
-          mnesia_rocksdb_tuning,mrdb,mrdb_index,mrdb_mutex,mrdb_select]}.
+          mnesia_rocksdb_tuning,mrdb,mrdb_index,mrdb_mutex,
+          mrdb_mutex_serializer,mrdb_select,mrdb_stats]}.

rebar.config

@@ -4,8 +4,8 @@
 {deps,
  [
   {sext, "1.8.0"},
-  {rocksdb, {git, "https://gitlab.com/seanhinde/erlang-rocksdb.git", {ref,"9ae37839"}}},
-  {hut, "1.3.0"}
+  {rocksdb, {git, "https://github.com/emqx/erlang-rocksdb.git", {ref,"d695c6e"}}},
+  {hut, "1.4.0"}
  ]}.
 
 {xref_checks, [
@@ -14,6 +14,10 @@
     deprecated_function_calls
 ]}.
 
+{dialyzer, [{plt_apps, all_deps},
+            {base_plt_apps, [erts, kernel, stdlib, mnesia ]}
+]}.
+
 {profiles,
  [
   {test,

rebar.config.script

@@ -1,11 +1,4 @@
 %% -*- erlang-mode -*-
-case os:getenv("ERLANG_ROCKSDB_OPTS") of
-    false ->
-        true = os:putenv("ERLANG_ROCKSDB_OPTS", "-DWITH_BUNDLE_LZ4=ON");
-    _ ->
-        %% If manually set, we assume it's throught through
-        skip
-end.
 case os:getenv("DEBUG") of
     "true" ->
 	Opts = proplists:get_value(erl_opts, CONFIG, []),

rebar.lock

@@ -1,15 +1,15 @@
 {"1.2.0",
-[{<<"hut">>,{pkg,<<"hut">>,<<"1.3.0">>},0},
+[{<<"hut">>,{pkg,<<"hut">>,<<"1.4.0">>},0},
  {<<"rocksdb">>,
-  {git,"https://gitlab.com/seanhinde/erlang-rocksdb.git",
-       {ref,"9ae378391ffc94200bde24efcd7a4921eba688d0"}},
+  {git,"https://github.com/emqx/erlang-rocksdb.git",
+       {ref,"d695c6ee9dd27bfe492ed4e24c72ad20ab0d770b"}},
   0},
  {<<"sext">>,{pkg,<<"sext">>,<<"1.8.0">>},0}]}.
 [
 {pkg_hash,[
- {<<"hut">>, <<"71F2F054E657C03F959CF1ACC43F436EA87580696528CA2A55C8AFB1B06C85E7">>},
+ {<<"hut">>, <<"7A1238EC00F95C9EC75412587EE11AC652ECA308A7F4B8CC9629746D579D6CF0">>},
  {<<"sext">>, <<"90A95B889F5C781B70BBCF44278B763148E313C376B60D87CE664CB1C1DD29B5">>}]},
 {pkg_hash_ext,[
- {<<"hut">>, <<"7E15D28555D8A1F2B5A3A931EC120AF0753E4853A4C66053DB354F35BF9AB563">>},
+ {<<"hut">>, <<"7AF8704B9BAE98A336F70D9560FC3C97F15665265FA603DBD05352E63D6EBB03">>},
  {<<"sext">>, <<"BC6016CB8690BAF677EACACFE6E7CADFEC8DC7E286CBBED762F6CD55B0678E73">>}]}
 ].

src/mnesia_rocksdb.erl

@@ -346,14 +346,11 @@ semantics(_Alias, index_fun) -> fun index_f/4;
 semantics(_Alias, _) -> undefined.
 
 is_index_consistent(Alias, {Tab, index, PosInfo}) ->
-    case info(Alias, Tab, {index_consistent, PosInfo}) of
-        true -> true;
-        _ -> false
-    end.
+    mnesia_rocksdb_admin:read_info(Alias, Tab, {index_consistent, PosInfo}, false).
 
-index_is_consistent(_Alias, {Tab, index, PosInfo}, Bool)
+index_is_consistent(Alias, {Tab, index, PosInfo}, Bool)
   when is_boolean(Bool) ->
-    mrdb:write_info(Tab, {index_consistent, PosInfo}, Bool).
+    mnesia_rocksdb_admin:write_info(Alias, Tab, {index_consistent, PosInfo}, Bool).
 
 
 %% PRIVATE FUN
@@ -454,8 +451,13 @@ close_table(Alias, Tab) ->
         error ->
             ok;
         _ ->
-            ok = mnesia_rocksdb_admin:prep_close(Alias, Tab),
-            close_table_(Alias, Tab)
+            case get(mnesia_dumper_dets) of
+                undefined ->
+                    ok = mnesia_rocksdb_admin:prep_close(Alias, Tab),
+                    close_table_(Alias, Tab);
+                _ ->
+                    ok
+            end
     end.
 
 close_table_(Alias, Tab) ->
@@ -798,7 +800,7 @@ handle_call({create_table, Tab, Props}, _From,
         exit:{aborted, Error} ->
             {reply, {aborted, Error}, St}
     end;
-handle_call({load_table, _LoadReason, Props}, _From,
+handle_call({load_table, _LoadReason, Props}, _,
             #st{alias = Alias, tab = Tab} = St) ->
     {ok, _Ref} = mnesia_rocksdb_admin:load_table(Alias, Tab, Props),
     {reply, ok, St#st{status = active}};
@@ -826,7 +828,7 @@ handle_call({delete, Key}, _From, St) ->
 handle_call({match_delete, Pat}, _From, #st{tab = Tab} = St) ->
     Res = mrdb:match_delete(get_ref(Tab), Pat),
     {reply, Res, St};
-handle_call(close_table, _From, #st{alias = Alias, tab = Tab} = St) ->
+handle_call(close_table, _, #st{alias = Alias, tab = Tab} = St) ->
     _ = mnesia_rocksdb_admin:close_table(Alias, Tab),
     {reply, ok, St#st{status = undefined}};
 handle_call(delete_table, _From, #st{alias = Alias, tab = Tab} = St) ->

src/mnesia_rocksdb_admin.erl

@@ -18,7 +18,8 @@
 	, clear_table/1
         ]).
 
--export([ migrate_standalone/2 ]).
+-export([ migrate_standalone/2
+        , migrate_standalone/3 ]).
 
 -export([ start_link/0
         , init/1
@@ -32,6 +33,7 @@
         , read_info/2            %% (Alias, Tab)
         , read_info/4            %% (Alias, Tab, Key, Default)
         , write_info/4           %% (Alias, Tab, Key, Value)
+        , delete_info/3          %% (Alias, Tab, Key)
         , write_table_property/3 %% (Alias, Tab, Property)
         ]).
 
@@ -66,6 +68,8 @@
 
 -type cf() :: mrdb:db_ref().
 
+-type rpt() :: undefined | map().
+
 -type req() :: {create_table, table(), properties()}
              | {delete_table, table()}
              | {load_table, table(), properties()}
@@ -74,7 +78,7 @@
              | {add_aliases, [alias()]}
              | {write_table_property, tabname(), tuple()}
              | {remove_aliases, [alias()]}
-             | {migrate, [{tabname(), map()}]}
+             | {migrate, [{tabname(), map()}], rpt()}
              | {prep_close, table()}
              | {close_table, table()}
 	     | {clear_table, table() | cf() }.
@@ -277,9 +281,20 @@ write_info(Alias, Tab, K, V) ->
     write_info_(get_ref({admin, Alias}), Tab, K, V).
 
 write_info_(Ref, Tab, K, V) ->
+    write_info_encv(Ref, Tab, K, term_to_binary(V)).
+
+write_info_encv(Ref, Tab, K, V) ->
     EncK = mnesia_rocksdb_lib:encode_key({info,Tab,K}, sext),
     maybe_write_standalone_info(Ref, K, V),
-    mrdb:rdb_put(Ref, EncK, term_to_binary(V), []).
+    mrdb:rdb_put(Ref, EncK, V, []).
+
+delete_info(Alias, Tab, K) ->
+    delete_info_(get_ref({admin, Alias}), Tab, K).
+
+delete_info_(Ref, Tab, K) ->
+    EncK = mnesia_rocksdb_lib:encode_key({info, Tab, K}, sext),
+    maybe_delete_standalone_info(Ref, K),
+    mrdb:rdb_delete(Ref, EncK, []).
 
 maybe_write_standalone_info(Ref, K, V) ->
     case Ref of
@@ -292,11 +307,28 @@ maybe_write_standalone_info(Ref, K, V) ->
             ok
     end.
 
+maybe_delete_standalone_info(Ref, K) ->
+    case Ref of
+        #{type := standalone, vsn := 1, db_ref := DbRef} ->
+            EncK = mnesia_rocksdb_lib:encode_key(K, sext),
+            Key = <<?INFO_TAG, EncK/binary>>,
+            rocksdb:delete(DbRef, Key, []);
+        _ ->
+            ok
+    end.
+
 write_table_property(Alias, Tab, Prop) when is_tuple(Prop), size(Prop) >= 1 ->
     call(Alias, {write_table_property, Tab, Prop}).
 
 migrate_standalone(Alias, Tabs) ->
-    call(Alias, {migrate, Tabs}).
+    migrate_standalone(Alias, Tabs, undefined).
+
+migrate_standalone(Alias, Tabs, Rpt0) ->
+    Rpt = case Rpt0 of
+              undefined -> undefined;
+              To -> #{to => To, tag => migrate_standalone}
+          end,
+    call(Alias, {migrate, Tabs, Rpt}).
 
 -spec call(alias() | [], req()) -> no_return() | any().
 call(Alias, Req) ->
@@ -313,13 +345,13 @@ call(Alias, Req, Timeout) ->
     end.
 
 start_link() ->
-    mrdb_mutex:ensure_tab(),
     gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
 
 init([]) ->
     _ = maybe_initial_meta(),  %% bootstrap pt
     Opts = default_opts(),
     process_flag(trap_exit, true),
+    ensure_cf_cache(),
     mnesia:subscribe({table, schema, simple}),
     {ok, recover_state(#st{default_opts = Opts})}.
 
@@ -394,12 +426,14 @@ try_load_admin_db(Alias, AliasOpts, #st{ backends = Bs
             %% We need to store the persistent ref explicitly here,
             %% since mnesia knows nothing of our admin table.
             AdminTab = {admin, Alias},
+            Stats = mrdb_stats:new(),
             CfI = update_cf_info(AdminTab, #{ status => open
                                             , name => AdminTab
                                             , vsn => ?VSN
                                             , encoding => {sext,{value,term}}
                                             , attr_pos => #{key => 1,
                                                             value => 2}
+                                            , stats => Stats
                                             , mountpoint => MP
                                             , properties =>
                                                 #{ attributes => [key, val]
@@ -499,12 +533,17 @@ intersection(A, B) ->
 
 -spec handle_req(alias(), req(), backend(), st()) -> gen_server_reply().
 handle_req(Alias, {create_table, Name, Props}, Backend, St) ->
-    case create_trec(Alias, Name, Props, Backend, St) of
-        {ok, NewCf} ->
-            St1 = update_cf(Alias, Name, NewCf, St),
-            {reply, {ok, NewCf}, St1};
-        {error, _} = Error ->
-            {reply, Error, St}
+    case find_cf(Alias, Name, Backend, St) of
+        {ok, TRec} ->
+            {reply, {ok, TRec}, St};
+        error ->
+            case create_trec(Alias, Name, Props, Backend, St) of
+                {ok, NewCf} ->
+                    St1 = update_cf(Alias, Name, NewCf, St),
+                    {reply, {ok, NewCf}, maybe_update_main(Alias, Name, create, St1)};
+                {error, _} = Error ->
+                    {reply, Error, St}
+            end
     end;
 handle_req(Alias, {load_table, Name, Props}, Backend, St) ->
     try
@@ -588,10 +627,10 @@ handle_req(Alias, {write_table_property, Tab, Prop}, Backend, St) ->
         _ ->
             {reply, {error, not_found}, St}
     end;
-handle_req(Alias, {migrate, Tabs0}, Backend, St) ->
-    case prepare_migration(Alias, Tabs0, St) of
+handle_req(Alias, {migrate, Tabs0, Rpt}, Backend, St) ->
+    case prepare_migration(Alias, Tabs0, Rpt, St) of
         {ok, Tabs} ->
-            {Res, St1} = do_migrate_tabs(Alias, Tabs, Backend, St),
+            {Res, St1} = do_migrate_tabs(Alias, Tabs, Backend, Rpt, St),
             {reply, Res, St1};
         {error, _} = Error ->
             {reply, Error, St}
@@ -617,8 +656,14 @@ maybe_update_main(Alias, {Main, index, I}, Op, St) ->
             case {Op, lists:member(I, Index)} of
                 {delete, true} ->
                     CfM1 = CfM#{properties => Props#{index => Index -- [I]}},
+                    delete_info(Alias, Main, {index_consistent, I}),
                     maybe_update_pt(Main, CfM1),
                     update_cf(Alias, Main, CfM1, St);
+                {create, _} ->
+                    %% Due to a previous bug, this marker might linger
+                    %% In any case, it mustn't be there for a newly created index
+                    delete_info(Alias, Main, {index_consistent, I}),
+                    St;
                 _ ->
                     St
             end;
@@ -629,6 +674,7 @@ maybe_update_main(Alias, {Main, index, I}, Op, St) ->
 maybe_update_main(_, _, _, St) ->
     St.
 
+
 %% The pt may not have been created yet. If so, don't do it here.
 maybe_update_pt(Name, Ref) ->
     case get_pt(Name, error) of
@@ -772,7 +818,8 @@ create_table_from_trec(Alias, Name, #{type := column_family} = TRec,
             ?log(debug, "will create ~p as standalone and migrate", [Name]),
             case create_table_as_standalone(Alias, Name, false, MP, TRec, St) of
                 {ok, OldTRec, _} ->
-                    create_cf_and_migrate(Alias, Name, OldTRec, TRec, Backend, St);
+                    ?log(info, "Migrating ~p to column family", [Name]),
+                    create_cf_and_migrate(Alias, Name, OldTRec, TRec, Backend, undefined, St);
                 _Other ->
                     create_table_as_cf(Alias, Name, TRec#{db_ref => DbRef}, St)
             end
@@ -781,11 +828,11 @@ create_table_from_trec(Alias, Name, #{type := standalone} = TRec, _, St) ->
     {Exists, MP} = table_exists_as_standalone(Name),
     create_table_as_standalone(Alias, Name, Exists, MP, TRec, St).
 
-create_cf_and_migrate(Alias, Name, OldTRec, TRec, #{db_ref := DbRef}, St) ->
+create_cf_and_migrate(Alias, Name, OldTRec, TRec, #{db_ref := DbRef}, Rpt, St) ->
     ?log(debug, "Migrate to cf (~p)", [Name]),
     {ok, NewCf, St1} = create_table_as_cf(
                          Alias, Name, TRec#{db_ref => DbRef}, St),
-    {ok, St2} = migrate_standalone_to_cf(OldTRec, NewCf, St1),
+    {ok, St2} = migrate_standalone_to_cf(OldTRec, NewCf, Rpt, St1),
     {ok, NewCf, St2}.
 
 %% Return {Migrate, MP} iff table exists standalone; just false if it doesn't
@@ -802,17 +849,33 @@ should_we_migrate_standalone(#{name := Name}) ->
             false
     end.
 
-prepare_migration(Alias, Tabs, St) ->
+prepare_migration(Alias, Tabs, Rpt, St) ->
     Res = lists:map(fun(T) ->
                             prepare_migration_(Alias, T, St)
                     end, Tabs),
     Res1 = add_related_tabs(Res, maps:get(Alias, St#st.backends), Alias, St),
     case [E || {error, _} = E <- Res1] of
-        [] -> {ok, Res1};
+        [] ->
+            rpt(Rpt, "Will migrate ~p~n", [[T || {T,_,_} <- Res1]]),
+            {ok, Res1};
         [_|_] = Errors ->
+            rpt(Rpt, "Errors encountered: ~p~n", [Errors]),
             {error, Errors}
     end.
 
+rpt(Rpt, Fmt, Args) ->
+    rpt(Rpt, erlang:system_time(millisecond), Fmt, Args).
+
+rpt(undefined, _, _, _) -> ok;
+rpt(#{to := Rpt} = R, Time, Fmt, Args) -> 
+    Rpt ! {mnesia_rocksdb, report, R#{time => Time, fmt => Fmt, args => Args}},
+    ok.
+
+maybe_progress(#{to := To}, C) when C rem 100000 =:= 0 ->
+    To ! {mnesia_rocksdb, report, progress};
+maybe_progress(_, _) ->
+    ok.
+
 add_related_tabs(Ts, Backend, Alias, St) ->
     lists:flatmap(
       fun({error,_} = E) -> [E];
@@ -839,29 +902,31 @@ prepare_migration_(Alias, T, #st{} = St) ->
             {error, {no_such_table, TName}}
     end.
 
-do_migrate_tabs(Alias, Tabs, Backend, St) ->
+do_migrate_tabs(Alias, Tabs, Backend, Rpt, St) ->
     lists:mapfoldl(fun(T, St1) ->
-                        do_migrate_table(Alias, T, Backend, St1)
+                        do_migrate_table(Alias, T, Backend, Rpt, St1)
                 end, St, Tabs).
 
-do_migrate_table(Alias, {Name, OldTRec, TRec0}, Backend, St) when is_map(TRec0) ->
+do_migrate_table(Alias, {Name, OldTRec, TRec0}, Backend, Rpt, St) when is_map(TRec0) ->
     T0 = erlang:system_time(millisecond),
+    rpt(Rpt, T0, "Migrate ~p~n", [Name]),
     TRec = maps:without([encoding, vsn], TRec0),
     maybe_write_user_props(TRec),
     {ok, CF, St1} = create_cf_and_migrate(Alias, Name, OldTRec,
-                                          TRec, Backend, St),
+                                          TRec, Backend, Rpt, St),
     put_pt(Name, CF),
     T1 = erlang:system_time(millisecond),
+    rpt(Rpt, T1, "~nDone (~p)~n", [Name]),
     Time = T1 - T0,
     io:fwrite("~p migrated, ~p ms~n", [Name, Time]),
     {{Name, {ok, Time}}, St1}.
 
 migrate_standalone_to_cf(OldTRec, #{name := T, alias := Alias} = TRec,
-                         #st{standalone = Ts} = St) ->
+                         Rpt, #st{standalone = Ts} = St) ->
     ChunkSz = chunk_size(TRec),
     KeyPos = mnesia_rocksdb_lib:keypos(T),
     migrate_to_cf(mrdb:select(OldTRec, [{'_',[],['$_']}], ChunkSz),
-                  TRec, OldTRec, KeyPos),
+                  TRec, OldTRec, KeyPos, set_count(0, Rpt)),
     case maps:is_key({Alias,T}, Ts)
          andalso table_is_empty(OldTRec) of
         true ->
@@ -871,27 +936,36 @@ migrate_standalone_to_cf(OldTRec, #{name := T, alias := Alias} = TRec,
             {ok, St}
     end.
 
-migrate_to_cf({L, Cont}, Cf, DbRec, KeyPos) ->
-    mrdb:as_batch(
-      Cf,
-      fun(New) ->
-              mrdb:as_batch(
-                DbRec,
-                fun(Old) ->
-                        lists:foreach(
-                          fun(Obj) ->
-                                  mrdb:insert(New, Obj),
-                                  mrdb:delete(Old, element(KeyPos,Obj))
-                          end, L)
-                end)
-      end),
-    migrate_to_cf(cont(Cont), Cf, DbRec, KeyPos);
-migrate_to_cf('$end_of_table', _, _, _) ->
+migrate_to_cf({L, Cont}, Cf, DbRec, KeyPos, Rpt) ->
+    Count0 = get_count(Rpt),
+    Count = mrdb:as_batch(
+              Cf,
+              fun(New) ->
+                      mrdb:as_batch(
+                        DbRec,
+                        fun(Old) ->
+                                lists:foldl(
+                                  fun(Obj, C) ->
+                                          mrdb:insert(New, Obj),
+                                          mrdb:delete(Old, element(KeyPos,Obj)),
+                                          maybe_progress(Rpt, C),
+                                          C + 1
+                                  end, Count0, L)
+                        end)
+              end),
+    migrate_to_cf(mrdb_select:select(Cont), Cf, DbRec, KeyPos, set_count(Count, Rpt));
+migrate_to_cf('$end_of_table', _, _, _, _) ->
     ok.
 
-cont('$end_of_table' = E) -> E;
-cont(F) when is_function(F,0) ->
-    F().
+get_count(undefined) ->
+    0;
+get_count(R) when is_map(R) ->
+    maps:get({count}, R, 0).
+
+set_count(_, undefined) ->
+    undefined;
+set_count(Count, R) when is_map(R) ->
+    R#{{count} => Count}.
 
 chunk_size(_) ->
     300.
@@ -1184,7 +1258,18 @@ load_info_(Res, I, ARef, Tab) ->
             DecK = mnesia_rocksdb_lib:decode_key(K),
             case read_info_(ARef, Tab, DecK, undefined) of
                 undefined ->
-                    write_info_(ARef, Tab, DecK, V);
+                    write_info_encv(ARef, Tab, DecK, V);
+                <<131,_/binary>> = Value ->
+                    %% Due to a previous bug, info values could be double-encoded with binary_to_term()
+                    try binary_to_term(Value) of
+                        _DecVal ->
+                            %% We haven't been storing erlang-term encoded data as info,
+                            %% so assume this is double-encoded and correct
+                            write_info_encv(ARef, Tab, DecK, Value)
+                    catch
+                        error:_ ->
+                            skip
+                    end;
                 _ ->
                     skip
             end,
@@ -1251,7 +1336,7 @@ rocksdb_opts_from_trec(TRec) ->
 
 create_table_as_cf(Alias, Name, #{db_ref := DbRef} = R, St) ->
     CfName = tab_to_cf_name(Name),
-    case rocksdb:create_column_family(DbRef, CfName, cfopts()) of
+    case create_column_family(DbRef, CfName, cfopts(), R) of
         {ok, CfH} ->
             R1 = check_version_and_encoding(R#{ cf_handle => CfH
                                               , type => column_family }),
@@ -1260,6 +1345,82 @@ create_table_as_cf(Alias, Name, #{db_ref := DbRef} = R, St) ->
             Error
     end.
 
+create_column_family(DbRef, CfName, CfOpts, R) ->
+    Res = case column_family_exists(CfName, R) of
+              true ->
+                  case find_active_cf_handle(DbRef, CfName) of
+                      error ->
+                          {error, {no_handle_for_existing_cf, CfName}};
+                      {ok, _} = Ok ->
+                          Ok
+                  end;
+              false ->
+                  rocksdb:create_column_family(DbRef, CfName, CfOpts)
+          end,
+    maybe_note_active_cf(Res, CfName),
+    Res.
+
+column_family_exists(CfName, #{mountpoint := MP}) ->
+    case rocksdb:list_column_families(MP, []) of
+        {ok, CFs} ->
+            lists:member(CfName, CFs);
+        _ ->
+            false
+    end;
+column_family_exists(CfName, #{alias := Alias}) ->
+    case get_ref({admin, Alias}, error) of
+        error ->
+            false;
+        Adm ->
+            column_family_exists(CfName, Adm)
+    end.
+
+%% Column family handle caching ======================================================
+%%
+%% At least as far as I can tell, there is no way to query erlang-rocksdb for currently
+%% active column family handles. This can become an issue e.g. during table migration
+%% from standalone to column families, where the meta structures aren't updated until
+%% after completed migration. A transient error during migration should be addressable
+%% by simply retrying, but if the CF has already been created, and we've lost the handle,
+%% there is no easy way to get it back. Unfortunately, if we start caching CFs, we also
+%% need to garbage collect them.
+%%
+
+-define(CFH_CACHE, mnesia_rocksdb_cf_handle_cache).
+
+ensure_cf_cache() ->
+    case ets:info(?CFH_CACHE, name) of
+        undefined ->
+            ets:new(?CFH_CACHE, [ordered_set, public, named_table]);
+        _ ->
+            true
+    end.
+
+maybe_note_active_cf({ok, CfH}, CfName) ->
+    ets:insert(?CFH_CACHE, {{CfName, CfH}});
+maybe_note_active_cf(_, _) ->
+    false.
+
+find_active_cf_handle(DbRef, CfName) ->
+    Candidates = ets:select(?CFH_CACHE, [{{{CfName,'$1'}}, [], ['$1']}]),
+    lists:foldl(fun(CfH, Acc) -> check_cfh(DbRef, CfH, CfName, Acc) end, error, Candidates).
+
+check_cfh(DbRef, CfH, CfName, Acc) ->
+    case rocksdb:iterator(DbRef, CfH, []) of
+        {ok, I} ->
+            rocksdb:iterator_close(I),
+            {ok, CfH};
+        {error, _} ->
+            ets:delete(?CFH_CACHE, {CfName, CfH}),
+            Acc
+    end.
+
+drop_cached_cf(CfName, CfH) ->
+    ets:delete(?CFH_CACHE, {CfName, CfH}).
+
+%%
+%% ===================================================================================
+
 do_prep_close(Name, Backend, St) ->
     RelTabs = get_related_resources(Name, Backend),
     erase_pt_list([Name | RelTabs]),
@@ -1302,6 +1463,7 @@ do_delete_table(Alias, Name, Backend, #st{} = St) ->
             case Where of
                 #{db_ref := DbRef, cf_handle := CfH, type := column_family} ->
                     rocksdb:drop_column_family(DbRef, CfH),
+                    drop_cached_cf(tab_to_cf_name(Name), CfH),
                     rocksdb:destroy_column_family(DbRef, CfH),
                     {ok, delete_cf(Alias, Name, St)};
                 #{type := standalone} = R ->

src/mnesia_rocksdb_sup.erl

@@ -42,4 +42,5 @@ start_link() ->
 %% ===================================================================
 
 init([]) ->
-    {ok, { {one_for_one, 5, 10}, [?CHILD(mnesia_rocksdb_params, worker)]} }.
+    {ok, { {one_for_one, 5, 10}, [ ?CHILD(mrdb_mutex_serializer, worker)
+                                 , ?CHILD(mnesia_rocksdb_params, worker)]} }.

src/mrdb.erl

@@ -105,6 +105,8 @@
 -include("mnesia_rocksdb.hrl").
 -include("mnesia_rocksdb_int.hrl").
 
+-define(BATCH_REF_DUMMY, '$mrdb_batch_ref_dummy').
+
 -type tab_name()  :: atom().
 -type alias()     :: atom().
 -type admin_tab() :: {admin, alias()}.
@@ -115,7 +117,9 @@
                | index()
                | retainer().
 
--type retries() :: non_neg_integer().
+-type inner() :: non_neg_integer().
+-type outer() :: non_neg_integer().
+-type retries() :: outer() | {inner(), outer()}.
 
 %% activity type 'ets' makes no sense in this context
 -type mnesia_activity_type() :: transaction
@@ -141,7 +145,7 @@
 
 -type tx_activity() :: #{ type    := 'tx'
                         , handle  := tx_handle()
-                        , attempt := non_neg_integer() }.
+                        , attempt := 'undefined' | retries() }.
 -type batch_activity() :: #{ type   := 'batch'
                            , handle := batch_handle() }.
 -type activity() :: tx_activity() | batch_activity().
@@ -246,7 +250,7 @@ release_snapshot(SHandle) ->
 %% <li> `{tx, TxOpts}' - A `rocksdb' transaction with sligth modifications</li>
 %% <li> `batch' - A `rocksdb' batch operation</li>
 %% </ul>
-%% 
+%%
 %% By default, transactions are combined with a snapshot with 1 retry.
 %% The snapshot ensures that writes from concurrent transactions don't leak into the transaction context.
 %% A transaction will be retried if it detects that the commit set conflicts with recent changes.
@@ -254,6 +258,14 @@ release_snapshot(SHandle) ->
 %% The re-run transaction may still fail, if new transactions, or non-transaction writes interfere with
 %% the commit set. It will then be re-run again, until the retry count is exhausted.
 %%
+%% For finer-grained retries, it's possible to set `retries => {Inner, Outer}'. Setting the retries to a
+%% single number, `Retries', is analogous to `{0, Retries}`. Each outer retry requests a 'mutex lock' by
+%% waiting in a FIFO queue. Once it receives the lock, it will try the activity once + as many retries
+%% as specified by `Inner'. If these fail, the activity again goes to the FIFO queue (ending up last
+%% in line) if there are outer retries remaining. When all retries are exhaused, the activity aborts
+%% with `retry_limit'. Note that transactions, being optimistic, do not request a lock on the first
+%% attempt, but only on outer retries (the first retry is always an outer retry).
+%%
 %% Valid `TxOpts' are `#{no_snapshot => boolean(), retries => retries()}'.
 %%
 %% To simplify code adaptation, `tx | transaction | sync_transaction' are synonyms, and
@@ -269,41 +281,92 @@ activity(Type, Alias, F) ->
                      #{ alias  => Alias
                       , db_ref => DbRef }, TxCtxt);
                batch ->
-                   Batch = get_batch_(DbRef),
+                   Batch = init_batch_ref(DbRef),
                    #{ activity => #{ type   => batch
                                    , handle => Batch }
                     , alias  => Alias
                     , db_ref => DbRef }
            end,
-    do_activity(F, Alias, Ctxt, false).
+    do_activity(F, Alias, Ctxt).
 
-do_activity(F, Alias, Ctxt, WithLock) ->
-    try run_f(F, Ctxt, WithLock, Alias) of
-        Res ->
-            try commit_and_pop(Res)
-            catch
-                throw:{?MODULE, busy} ->
-                    do_activity(F, Alias, Ctxt, true)
-            end
+do_activity(F, Alias, Ctxt) ->
+    try try_f(F, Ctxt)
     catch
+        throw:{?MODULE, busy} ->
+            retry_activity(F, Alias, Ctxt)
+    end.
+
+try_f(F, Ctxt) ->
+    try_f(mnesia_compatible_aborts(Ctxt), F, Ctxt).
+
+try_f(false, F, Ctxt) ->
+    try run_f(F, Ctxt) of
+        Res ->
+            commit_and_pop(Res)
+    catch
+        throw:Something ->
+            abort_and_pop(throw, Something);
+        Cat:Err:T ->
+            %% Without capturing the stacktract here,
+            %% debugging gets pretty difficult. Incompatible with mnesia, though.
+            abort_and_pop(Cat, {Err, T})
+    end;
+try_f(true, F, Ctxt) ->
+    try run_f(F, Ctxt) of
+        Res ->
+            commit_and_pop(Res)
+    catch
+        throw:Something ->
+            abort_and_pop(throw, Something);
         Cat:Err ->
+            %% Without capturing the stacktract here,
+            %% debugging gets pretty difficult
             abort_and_pop(Cat, Err)
     end.
 
-run_f(F, Ctxt, false, _) ->
-    push_ctxt(Ctxt),
-    F();
-run_f(F, Ctxt, true, Alias) ->
-    mrdb_mutex:do(
-      Alias,
-      fun() ->
-              push_ctxt(incr_attempt(Ctxt)),
-              F()
-      end).
 
-incr_attempt(#{ activity := #{type := tx, attempt := A} = Act, db_ref := DbRef } = C) ->
+run_f(F, Ctxt) ->
+    push_ctxt(Ctxt),
+    F().
+
+incr_attempt(0, {_,O}) when O > 0 ->
+    {outer, {0,1}};
+incr_attempt({I,O}, {Ri,Ro}) when is_integer(I), is_integer(O),
+                                  is_integer(Ri), is_integer(Ro) ->
+    if I < Ri -> {inner, {I+1, O}};
+       O < Ro -> {outer, {0, O+1}};
+       true ->
+            error
+    end;
+incr_attempt(_, _) ->
+    error.
+
+retry_activity(F, Alias, #{activity := #{ type := Type
+                                        , attempt := A
+                                        , retries := R} = Act} = Ctxt) ->
+    case incr_attempt(A, R) of
+        {RetryCtxt, A1} ->
+            Act1 = Act#{attempt := A1},
+            Ctxt1 = Ctxt#{activity := Act1},
+            try retry_activity_(RetryCtxt, F, Alias, restart_ctxt(Ctxt1))
+            catch
+                throw:{?MODULE, busy} ->
+                    retry_activity(F, Alias, Ctxt1)
+            end;
+        error ->
+            return_abort(Type, error, retry_limit)
+    end.
+
+retry_activity_(inner, F, Alias, Ctxt) ->
+    mrdb_stats:incr(Alias, inner_retries, 1),
+    try_f(F, Ctxt);
+retry_activity_(outer, F, Alias, Ctxt) ->
+    mrdb_stats:incr(Alias, outer_retries, 1),
+    mrdb_mutex:do(Alias, fun() -> try_f(F, Ctxt) end).
+
+restart_ctxt(#{ activity := #{type := tx} = Act, db_ref := DbRef } = C) ->
     {ok, TxH} = rdb_transaction(DbRef, []),
-    Act1 = Act#{attempt := A+1, handle := TxH},
+    Act1 = Act#{handle := TxH},
     C1 = C#{ activity := Act1 },
     case maps:is_key(snapshot, C) of
         true ->
@@ -368,7 +431,7 @@ apply_tx_opts(Opts0) when is_map(Opts0) ->
     check_tx_opts(maps:merge(default_tx_opts(), Opts0)).
 
 check_tx_opts(Opts) ->
-    case maps:without([no_snapshot, retries], Opts) of
+    case maps:without([no_snapshot, retries, mnesia_compatible], Opts) of
         Other when map_size(Other) > 0 ->
             abort({invalid_tx_opts, maps:keys(Other)});
         _ ->
@@ -376,10 +439,14 @@ check_tx_opts(Opts) ->
     end.
 
 check_retries(#{retries := Retries} = Opts) ->
-    if is_integer(Retries), Retries >= 0 ->
-           Opts;
-       true ->
-           error({invalid_tx_option, {retries, Retries}})
+    case Retries of
+        _ when is_integer(Retries), Retries >= 0 ->
+            Opts#{retries := {0, Retries}};
+        {Inner, Outer} when is_integer(Inner), is_integer(Outer),
+                            Inner >= 0, Outer >= 0 ->
+            Opts;
+        _ ->
+            error({invalid_tx_option, {retries, Retries}})
     end.
 
 check_nosnap(#{no_snapshot := NoSnap} = Opts) ->
@@ -394,7 +461,7 @@ create_tx(Opts, DbRef) ->
     {ok, TxH} = rdb_transaction(DbRef, []),
     Opts#{activity => maps:merge(Opts, #{ type    => tx
                                         , handle  => TxH
-                                        , attempt => 1})}.
+                                        , attempt => 0 })}.
 
 maybe_snapshot(#{no_snapshot := NoSnap} = Opts, DbRef) ->
     case NoSnap of
@@ -414,8 +481,7 @@ commit_and_pop(Res) ->
                     Res;
                 {error, {error, "Resource busy" ++ _ = Busy}} ->
                     case A of
-                        #{retries := Retries, attempt := Att}
-                          when Att =< Retries ->
+                        #{retries := {I,O}} when I > 0; O > 0 ->
                             throw({?MODULE, busy});
                         _ ->
                             error({error, Busy})
@@ -477,6 +543,11 @@ re_throw(Cat, Err) ->
 mnesia_compatible_aborts() ->
     mnesia_rocksdb_admin:get_cached_env(mnesia_compatible_aborts, false).
 
+mnesia_compatible_aborts(#{activity := #{mnesia_compatible := Bool}}) ->
+    Bool;
+mnesia_compatible_aborts(_) ->
+    mnesia_compatible_aborts().
+
 fix_error({aborted, Err}) ->
     Err;
 fix_error(Err) ->
@@ -486,7 +557,7 @@ rdb_transaction(DbRef, Opts) ->
     rocksdb:transaction(DbRef, Opts).
 
 rdb_transaction_commit_and_pop(H) ->
-    try rdb_transaction_commit(H) 
+    try rdb_transaction_commit(H)
     after
         pop_ctxt()
     end.
@@ -507,10 +578,13 @@ rdb_write_batch_and_pop(BatchRef, C) ->
         pop_ctxt()
     end.
 
+rdb_release_batch(?BATCH_REF_DUMMY) ->
+    ok;
 rdb_release_batch(H) ->
     rocksdb:release_batch(H).
 
 %% @doc Aborts an ongoing {@link activity/2}
+-spec abort(_) -> no_return().
 abort(Reason) ->
     case mnesia_compatible_aborts() of
 	true ->
@@ -529,7 +603,7 @@ new_tx(#{activity := _}, _) ->
 new_tx(Tab, Opts) ->
     #{db_ref := DbRef} = R = ensure_ref(Tab),
     {ok, TxH} = rdb_transaction(DbRef, write_opts(R, Opts)),
-    R#{activity => #{type => tx, handle => TxH, attempt => 1}}.
+    R#{activity => #{type => tx, handle => TxH, attempt => 0}}.
 
 -spec tx_ref(ref_or_tab() | db_ref() | db_ref(), tx_handle()) -> db_ref().
 tx_ref(Tab, TxH) ->
@@ -539,7 +613,7 @@ tx_ref(Tab, TxH) ->
         #{activity := #{type := tx, handle := OtherTxH}} ->
             error({tx_handle_conflict, OtherTxH});
         R ->
-            R#{activity => #{type => tx, handle => TxH, attempt => 1}}
+            R#{activity => #{type => tx, handle => TxH, attempt => 0}}
     end.
 
 -spec tx_commit(tx_handle() | db_ref()) -> ok.
@@ -798,7 +872,7 @@ update_index_do_bag(Ixs, Name, R, Key, Obj, Opts) ->
                     not_found
             end
     end.
-    
+
 update_index_do([{_Pos,ordered} = Ix|Ixs], Name, R, Key, Obj, Rest, Opts) ->
     Tab = {Name, index, Ix},
     #{ix_vals_f := IxValsF} = IRef = ensure_ref(Tab, R),
@@ -889,7 +963,7 @@ index_read_(#{name := Main, semantics := Sem} = Ref, Val, Ix) ->
             _ when is_atom(Ix) ->
                 {attr_pos(Ix, Ref), ordered};
             {_} ->
-                Ix;
+                {Ix, ordered};
             _ when is_integer(Ix) ->
                 {Ix, ordered}
         end,
@@ -970,7 +1044,7 @@ alias_of(Tab) ->
 %% and when releasing, all batches are released. This will not ensure
 %% atomicity, but there is no way in rocksdb to achieve atomicity
 %% across db instances. At least, data should end up where you expect.
-%% 
+%%
 %% @end
 -spec as_batch(ref_or_tab(), fun( (db_ref()) -> Res )) -> Res.
 as_batch(Tab, F) ->
@@ -1010,18 +1084,20 @@ get_batch(#{db_ref := DbRef, batch := BatchRef}) ->
 get_batch(_) ->
     {error, badarg}.
 
-get_batch_(DbRef) ->
+init_batch_ref(DbRef) ->
     Ref = make_ref(),
-    {ok, Batch} = rdb_batch(),
-    pdict_put({mrdb_batch, Ref}, #{DbRef => Batch}),
+    pdict_put({mrdb_batch, Ref}, #{DbRef => ?BATCH_REF_DUMMY}),
     Ref.
 
+get_batch_(DbRef) -> Ref = make_ref(), {ok, Batch} = rdb_batch(),
+    pdict_put({mrdb_batch, Ref}, #{DbRef => Batch}), Ref.
+
 get_batch_(DbRef, BatchRef) ->
     Key = batch_ref_key(BatchRef),
     case pdict_get(Key) of
         undefined ->
             error(stale_batch_ref);
-        #{DbRef := Batch} ->
+        #{DbRef := Batch} when Batch =/= ?BATCH_REF_DUMMY ->
             Batch;
         Map ->
             {ok, Batch} = rdb_batch(),
@@ -1050,7 +1126,9 @@ write_batches(BatchRef, Opts) ->
             pdict_erase(Key),
             ret_batch_write_acc(
               maps:fold(
-                fun(DbRef, Batch, Acc) ->
+                fun(_, ?BATCH_REF_DUMMY, Acc) ->
+                        Acc;
+                    (DbRef, Batch, Acc) ->
                         case rocksdb:write_batch(DbRef, Batch, Opts) of
                             ok ->
                                 Acc;
@@ -1269,7 +1347,7 @@ rdb_fold(Tab, Fun, Acc, Prefix, Limit) when is_function(Fun, 3)
     true = valid_limit(Limit),
     mrdb_select:rdb_fold(ensure_ref(Tab), Fun, Acc, Prefix, Limit).
 
-valid_limit(L) -> 
+valid_limit(L) ->
     case L of
         infinity ->
             true;
@@ -1494,8 +1572,9 @@ rdb_iterator(R) -> rdb_iterator(R, []).
 rdb_iterator(R, Opts) ->
     rdb_iterator_(R, read_opts(R, Opts)).
 
-rdb_iterator_(#{db_ref := DbRef, cf_handle := CfH, activity := #{type := tx, handle := TxH}}, ROpts) ->
-    rocksdb:transaction_iterator(DbRef, TxH, CfH, ROpts);
+rdb_iterator_(#{cf_handle := CfH, activity := #{type := tx, handle := TxH}}, ROpts) ->
+    %% Note: versions before 1.8.0 expected DbRef as first argument
+    rocksdb:transaction_iterator(TxH, CfH, ROpts);
 rdb_iterator_(#{db_ref := DbRef, cf_handle := CfH}, ROpts) ->
     rocksdb:iterator(DbRef, CfH, ROpts).
 
@@ -1506,11 +1585,11 @@ rdb_merge_(#{db_ref := DbRef, cf_handle := CfH}, K, Op, WOpts) ->
     rocksdb:merge(DbRef, CfH, K, Op, WOpts).
 
 write_opts(#{write_opts := Os}, Opts) -> Os ++ Opts;
-write_opts(_, Opts) -> 
+write_opts(_, Opts) ->
     Opts.
 
 read_opts(#{read_opts := Os}, Opts) -> Os ++ Opts;
-read_opts(_, Opts) -> 
+read_opts(_, Opts) ->
     Opts.
 
 -define(EOT, '$end_of_table').

src/mrdb_index.erl

@@ -6,11 +6,14 @@
         , iterator_move/2
         , iterator/2
         , iterator_close/1
+        , fold/4
+        , rev_fold/4
+        , index_ref/2
         ]).
 
--record(mrdb_ix_iter, { i    :: mrdb:iterator()
+-record(mrdb_ix_iter, { i    :: mrdb:mrdb_iterator()
                       , type = set :: set | bag
-                      , sub  :: mrdb:ref() | pid()
+                      , sub  :: pid() | mrdb:db_ref()
                       }).
 
 -type ix_iterator() :: #mrdb_ix_iter{}.
@@ -19,7 +22,7 @@
 
 -type object() :: tuple().
 
--record(subst, { i :: mrdb:iterator()
+-record(subst, { i :: mrdb:mrdb_iterator()
                , vals_f
                , cur
                , mref }).
@@ -62,6 +65,41 @@ iterator(Tab, IxPos) ->
 iterator_move(#mrdb_ix_iter{type = set} = IxI, Dir) -> iterator_move_set(IxI, Dir);
 iterator_move(#mrdb_ix_iter{type = bag} = IxI, Dir) -> iterator_move_bag(IxI, Dir).
 
+-spec fold(mrdb:ref_or_tab(), mrdb:index_position(),
+           fun( (index_value(), object() | [], Acc) -> Acc ), Acc) -> Acc
+          when Acc :: any().
+%% Folds over the index table corresponding to Tab and IxPos.
+%% The fold fun is called with the index key and the corresponding object,
+%% or [] if there is no such object.
+fold(Tab, IxPos, FoldFun, Acc) when is_function(FoldFun, 3) ->
+    fold_(Tab, IxPos, first, next, FoldFun, Acc).
+
+-spec rev_fold(mrdb:ref_or_tab(), mrdb:index_position(),
+               fun( (index_value(), object() | [], Acc) -> Acc ), Acc) -> Acc
+              when Acc :: any().
+%% Like fold/4 above, but folds over the index in the reverse order.
+rev_fold(Tab, IxPos, FoldFun, Acc) when is_function(FoldFun, 3) ->
+    fold_(Tab, IxPos, last, prev, FoldFun, Acc).
+
+fold_(Tab, IxPos, Start, Dir, FoldFun, Acc) ->
+    with_iterator(
+      Tab, IxPos,
+      fun(I) ->
+              iter_fold(I, Start, Dir, FoldFun, Acc)
+      end).
+
+iter_fold(I, Start, Dir, Fun, Acc) ->
+    iter_fold_(iterator_move(I, Start), I, Dir, Fun, Acc).
+
+iter_fold_({ok, IxVal, Obj}, I, Dir, Fun, Acc) ->
+    iter_fold_(iterator_move(I, Dir), I, Dir, Fun, Fun(IxVal, Obj, Acc));
+iter_fold_({error, _}, _, _, _, Acc) ->
+    Acc.
+
+index_ref(Tab, Pos) ->
+    TRef = mrdb:ensure_ref(Tab),
+    ensure_index_ref(Pos, TRef).
+
 iterator_move_set(#mrdb_ix_iter{i = I, sub = Sub}, Dir) ->
     case mrdb:iterator_move(I, Dir) of
         {ok, {{FKey, PKey}}} ->
@@ -83,6 +121,7 @@ iterator_move_bag(#mrdb_ix_iter{i = I, sub = Sub}, Dir) ->
             Other
     end.
 
+-spec opt_read(mrdb:ref_or_tab(), Key :: any()) -> any().
 opt_read(R, Key) ->
     case mrdb:read(R, Key, []) of
         [Obj] ->

src/mrdb_mutex.erl

@@ -3,79 +3,109 @@
 
 -export([ do/2 ]).
 
--export([ ensure_tab/0 ]).
+-ifdef(TEST).
+-include_lib("eunit/include/eunit.hrl").
+-endif.
 
--define(LOCK_TAB, ?MODULE).
-
-%% We use a wrapping ets counter (default: 0) as a form of semaphor.
-%% The claim operation is done using an atomic list of two updates:
-%% first, incrementing with 0 - this returns the previous value
-%% then, incrementing with 1, but wrapping at 1, ensuring that we get 1 back,
-%% regardless of previous value. This means that if [0,1] is returned, the resource
-%% was not locked previously; if [1,1] is returned, it was.
+%% We use a gen_server-based FIFO queue (one queue per alias) to manage the
+%% critical section.
 %%
-%% Releasing the resource is done by deleting the resource. If we just decrement,
-%% we will end up with lingering unlocked resources, so we might as well delete.
-%% Either operation is atomic, and the claim op creates the object if it's missing.
+%% Releasing the resource is done by notifying the server.
+%%
+%% Previous implementations tested:
+%% * A counter (in ets), incremented atomically first with 0, then 1. This lets
+%%   the caller know if the 'semaphore' was 1 or zero before. If it was already
+%%   1, busy-loop over the counter until it reads as a transition from 0 to 1.
+%%   This is a pretty simple implementation, but it lacks ordering, and appeared
+%%   to sometimes lead to starvation.
+%% * A duplicate_bag ets table: These are defined such that insertion order is
+%%   preserved. Thus, they can serve as a mutex with FIFO characteristics.
+%%   Unfortunately, performance plummeted when tested with > 25 concurrent
+%%   processes. While this is likely a very high number, given that we're talking
+%%   about contention in a system using optimistic concurrency, it's never good
+%%   if performance falls off a cliff.
 
 do(Rsrc, F) when is_function(F, 0) ->
-    true = claim(Rsrc),
+    {ok, Ref} = mrdb_mutex_serializer:wait(Rsrc),
     try F()
     after
-        release(Rsrc)
+        mrdb_mutex_serializer:done(Rsrc, Ref)
     end.
 
-claim(Rsrc) ->
-    case claim_(Rsrc) of
-        true -> true;
-        false -> busy_wait(Rsrc, 1000)
+-ifdef(TEST).
+
+mutex_test_() ->
+    {foreach,
+     fun setup/0,
+     fun cleanup/1,
+     [
+      {"Check that all operations complete", fun swarm_do/0}
+     ]}.
+
+setup() ->
+    case whereis(mrdb_mutex_serializer) of
+        undefined ->
+            {ok, Pid} = mrdb_mutex_serializer:start_link(),
+            Pid;
+        Pid ->
+            Pid
     end.
 
-claim_(Rsrc) ->
-    case ets:update_counter(?LOCK_TAB, Rsrc, [{2, 0},
-                                              {2, 1, 1, 1}], {Rsrc, 0}) of
-        [0, 1] ->
-            %% have lock
-            true;
-        [1, 1] ->
-            false
-    end.
+cleanup(Pid) ->
+    unlink(Pid),
+    exit(Pid, kill).
 
-%% The busy-wait function makes use of the fact that we can read a timer to find out
-%% if it still has time remaining. This reduces the need for selective receive, looking
-%% for a timeout message. We yield, then retry the claim op. Yielding at least used to
-%% also be necessary for the `read_timer/1` value to refresh.
-%%
-busy_wait(Rsrc, Timeout) ->
-    Ref = erlang:send_after(Timeout, self(), {claim, Rsrc}),
-    do_wait(Rsrc, Ref).
-
-do_wait(Rsrc, Ref) ->
-    erlang:yield(),
-    case erlang:read_timer(Ref) of
-        false ->
-            erlang:cancel_timer(Ref),
-            error(lock_wait_timeout);
-        _ ->
-            case claim_(Rsrc) of
-                true ->
-                    erlang:cancel_timer(Ref),
-                    ok;
-                false ->
-                    do_wait(Rsrc, Ref)
-            end
-    end.
-
-release(Rsrc) ->
-    ets:delete(?LOCK_TAB, Rsrc),
+swarm_do() ->
+    Rsrc = ?LINE,
+    Pid = spawn(fun() -> collect([]) end),
+    L = lists:seq(1, 1000),
+    Evens = [X || X <- L, is_even(X)],
+    Pids = [spawn_monitor(fun() ->
+                                  send_even(Rsrc, N, Pid)
+                          end) || N <- lists:seq(1,1000)],
+    await_pids(Pids),
+    Results = fetch(Pid),
+    {incorrect_results, []} = {incorrect_results, Results -- Evens},
+    {missing_correct_results, []} = {missing_correct_results, Evens -- Results},
     ok.
 
-
-%% Called by the process holding the ets table.
-ensure_tab() ->
-    case ets:info(?LOCK_TAB, name) of
-        undefined ->
-            ets:new(?LOCK_TAB, [set, public, named_table, {write_concurrency, true}]);
-        _ ->
-            true
+collect(Acc) ->
+    receive
+        {_, result, N} ->
+            collect([N|Acc]);
+        {From, fetch} ->
+            From ! {fetch_reply, Acc},
+            done
     end.
+
+fetch(Pid) ->
+    Pid ! {self(), fetch},
+    receive
+        {fetch_reply, Result} ->
+            Result
+    end.
+
+is_even(N) ->
+    (N rem 2) =:= 0.
+
+await_pids([{_, MRef}|Pids]) ->
+    receive
+        {'DOWN', MRef, _, _, _} ->
+            await_pids(Pids)
+    after 10000 ->
+            error(timeout)
+    end;
+await_pids([]) ->
+    ok.
+
+send_even(Rsrc, N, Pid) ->
+    do(Rsrc, fun() ->
+                     case is_even(N) of
+                         true ->
+                             Pid ! {self(), result, N};
+                         false ->
+                             exit(not_even)
+                     end
+             end).
+
+-endif.

src/mrdb_mutex_serializer.erl

@@ -0,0 +1,98 @@
+-module(mrdb_mutex_serializer).
+
+-behavior(gen_server).
+
+-export([wait/1,
+         done/2]).
+
+-export([start_link/0]).
+
+-export([init/1,
+         handle_call/3,
+         handle_cast/2,
+         handle_info/2,
+         terminate/2,
+         code_change/3]).
+
+-record(st, { queues = #{}
+            , empty_q = queue:new() }).  %% perhaps silly optimization
+
+wait(Rsrc) ->
+    gen_server:call(?MODULE, {wait, Rsrc}, infinity).
+
+done(Rsrc, Ref) ->
+    gen_server:cast(?MODULE, {done, Rsrc, Ref}).
+
+start_link() ->
+    gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
+
+init([]) ->
+    {ok, #st{}}.
+
+handle_call({wait, Rsrc}, {Pid, _} = From, #st{ queues = Queues
+                                              , empty_q = NewQ } = St) ->
+    MRef = erlang:monitor(process, Pid),
+    Q0 = maps:get(Rsrc, Queues, NewQ),
+    WasEmpty = queue:is_empty(Q0),
+    Q1 = queue:in({From, MRef}, Q0),
+    St1 = St#st{ queues = Queues#{Rsrc => Q1} },
+    case WasEmpty of
+        true ->
+            {reply, {ok, MRef}, St1};
+        false ->
+            {noreply, St1}
+    end.
+
+handle_cast({done, Rsrc, MRef}, #st{ queues = Queues } = St) ->
+    case maps:find(Rsrc, Queues) of
+        {ok, Q} ->
+            case queue:out(Q) of
+                {{value, {_From, MRef}}, Q1} ->
+                    erlang:demonitor(MRef),
+                    ok = maybe_dispatch_one(Q1),
+                    {noreply, St#st{ queues = Queues#{Rsrc := Q1} }};
+                {_, _} ->
+                    %% Not the lock holder
+                    {noreply, St}
+            end;
+        error ->
+            {noreply, St}
+    end.
+
+handle_info({'DOWN', MRef, process, _, _}, #st{queues = Queues} = St) ->
+    Queues1 =
+        maps:map(
+          fun(_Rsrc, Q) ->
+                  drop_or_filter(Q, MRef)
+          end, Queues),
+    {noreply, St#st{ queues = Queues1 }};
+handle_info(_, St) ->
+    {noreply, St}.
+
+terminate(_, _) ->
+    ok.
+
+code_change(_FromVsn, St, _Extra) ->
+    {ok, St}.
+
+%% In this function, we don't actually pop
+maybe_dispatch_one(Q) ->
+    case queue:peek(Q) of
+        empty ->
+            ok;
+        {value, {From, MRef}} ->
+            gen_server:reply(From, {ok, MRef}),
+            ok
+    end.
+        
+drop_or_filter(Q, MRef) ->
+    case queue:peek(Q) of
+        {value, {_, MRef}} ->
+            Q1 = queue:drop(Q),
+            ok = maybe_dispatch_one(Q1),
+            Q1;
+        {value, _Other} ->
+            queue:filter(fun({_,M}) -> M =/= MRef end, Q);
+        empty ->
+            Q
+    end.

src/mrdb_stats.erl

@@ -0,0 +1,74 @@
+%% -*- mode: erlang; erlang-indent-level: 4; indent-tabs-mode: nil -*-
+%% @doc Statistics API for the mnesia_rocksdb plugin
+%%
+%% Some counters are maintained for each active alias. Currently, the following
+%% counters are supported:
+%% * inner_retries
+%% * outer_retries
+%%
+-module(mrdb_stats).
+
+-export([new/0]).
+
+-export([incr/3,
+         get/1,
+         get/2]).
+
+-type alias() :: mnesia_rocksdb:alias().
+-type db_ref() :: mrdb:db_ref().
+-type counter() :: atom().
+-type increment() :: integer().
+
+-type counters() :: #{ counter() := integer() }.
+
+new() ->
+    #{ ref => counters:new(map_size(ctr_meta()), [write_concurrency])
+     , meta => ctr_meta()}.
+
+ctr_meta() ->
+    #{ inner_retries => 1
+     , outer_retries => 2 }.
+
+-spec incr(alias() | db_ref(), counter(), increment()) -> ok.
+%% @doc Increment `Ctr' counter for `Alias` with increment `N'.
+%%
+%% Note that the first argument may also be a `db_ref()' map,
+%% corresponding to `mrdb:get_ref({admin, Alias})'.
+%% @end
+incr(Alias, Ctr, N) when is_atom(Alias) ->
+    #{stats := #{ref := Ref, meta := Meta}} = mrdb:get_ref({admin, Alias}),
+    incr_(Ref, Meta, Ctr, N);
+incr(#{stats := #{ref := Ref, meta := Meta}}, Ctr, N) ->
+    incr_(Ref, Meta, Ctr, N).
+
+-spec get(alias() | db_ref(), counter()) -> integer().
+%% @doc Fetches the integer value of the known counter `Ctr' for `Alias'.
+%% @end
+get(Alias, Ctr) when is_atom(Alias) ->
+    #{stats := #{ref := Ref, meta := Meta}} = mrdb:get_ref({admin, Alias}),
+    get_(Ref, Meta, Ctr);
+get(#{stats := #{ref := Ref, meta := Meta}}, Ctr) ->
+    get_(Ref, Meta, Ctr).
+
+-spec get(alias() | db_ref()) -> counters().
+%% @doc Fetches all known counters for `Alias', in the form of a map,
+%% `#{Counter => Value}'.
+%% @end
+get(Alias) when is_atom(Alias) ->
+    get_(mrdb:get_ref({admin, Alias}));
+get(Ref) when is_map(Ref) ->
+    get_(Ref).
+
+get_(#{stats := #{ref := Ref, meta := Meta}}) ->
+    lists:foldl(
+      fun({K, P}, M) ->
+              M#{K := counters:get(Ref, P)}
+      end, Meta, maps:to_list(Meta)).
+
+get_(Ref, Meta, Attr) ->
+    Ix = maps:get(Attr, Meta),
+    counters:get(Ref, Ix).
+
+incr_(Ref, Meta, Attr, N) ->
+    Ix = maps:get(Attr, Meta),
+    counters:add(Ref, Ix, N).

test/mnesia_rocksdb_SUITE.erl

@@ -24,6 +24,7 @@
         , mrdb_abort/1
         , mrdb_two_procs/1
         , mrdb_two_procs_tx_restart/1
+        , mrdb_two_procs_tx_inner_restart/1
         , mrdb_two_procs_snap/1
         , mrdb_three_procs/1
         ]).
@@ -53,6 +54,7 @@ groups() ->
                          , mrdb_abort
                          , mrdb_two_procs
                          , mrdb_two_procs_tx_restart
+                         , mrdb_two_procs_tx_inner_restart
                          , mrdb_two_procs_snap
                          , mrdb_three_procs ]}
     ].
@@ -267,7 +269,7 @@ mrdb_abort(Config) ->
     Pre = mrdb:read(tx_abort, a),
     D0 = get_dict(),
     TRes = try mrdb:activity(
-                 tx, rdb,
+                 {tx, #{mnesia_compatible => true}}, rdb,
                  fun() ->
                          [{tx_abort, a, N}] = mrdb:read(tx_abort, a),
                          error(abort_here),
@@ -287,10 +289,17 @@ mrdb_abort(Config) ->
 mrdb_two_procs(Config) ->
     tr_ct:with_trace(fun mrdb_two_procs_/1, Config,
                      tr_flags(
-                       {self(), [call, sos, p]},
+                       {self(), [call, sos, p, 'receive']},
                        tr_patterns(
                          mrdb, [ {mrdb, insert, 2, x}
                                , {mrdb, read, 2, x}
+                               , {mrdb, retry_activity, 3, x}
+                               , {mrdb, try_f, 2, x}
+                               , {mrdb, incr_attempt, 2, x}
+                               , {mrdb_mutex, do, 2, x}
+                               , {mrdb_mutex_serializer, do, 2, x}
+                               , {?MODULE, wait_for_other, 2, x}
+                               , {?MODULE, go_ahead_other, 1, x}
                                , {mrdb, activity, x}], tr_opts()))).
 
 mrdb_two_procs_(Config) ->
@@ -314,11 +323,16 @@ mrdb_two_procs_(Config) ->
                  Pre = mrdb:read(R, a),
                  go_ahead_other(POther),
                  await_other_down(POther, MRef, ?LINE),
+                 %% The test proc is still inside the transaction, and POther
+                 %% has terminated/committed. If we now read 'a', we should see
+                 %% the value that POther wrote (no isolation).
                  [{R, a, 17}] = mrdb:read(R, a),
                  ok = mrdb:insert(R, {R, a, 18})
          end,
-    go_ahead_other(1, POther),
+    go_ahead_other(0, POther),
     Do0 = get_dict(),
+    %% Our transaction should fail due to the resource conflict, and because
+    %% we're not using retries.
     try mrdb:activity({tx, #{no_snapshot => true,
                              retries => 0}}, rdb, F1) of
         ok -> error(unexpected)
@@ -339,6 +353,7 @@ mrdb_two_procs_tx_restart_(Config) ->
     R = ?FUNCTION_NAME,
     Parent = self(),
     Created = create_tabs([{R, []}], Config),
+    check_stats(rdb),
     mrdb:insert(R, {R, a, 1}),
     Pre = mrdb:read(R, a),
     F0 = fun() ->
@@ -354,7 +369,7 @@ mrdb_two_procs_tx_restart_(Config) ->
                  OtherWrite = [{R, a, 17}],
                  Att = get_attempt(),
                  Expected = case Att of
-                                1 -> Pre;
+                                0 -> Pre;
                                 _ -> OtherWrite
                             end,
                  Expected = mrdb:read(R, a),
@@ -363,14 +378,104 @@ mrdb_two_procs_tx_restart_(Config) ->
                  OtherWrite = mrdb:read(R, a),
                  ok = mrdb:insert(R, {R, a, 18})
          end,
-    go_ahead_other(1, POther),
+    go_ahead_other(0, POther),
     Do0 = get_dict(),
     mrdb:activity({tx, #{no_snapshot => true}}, rdb, F1),
     dictionary_unchanged(Do0),
+    check_stats(rdb),
     [{R, a, 18}] = mrdb:read(R, a),
     delete_tabs(Created),
     ok.
 
+mrdb_two_procs_tx_inner_restart(Config) ->
+    tr_ct:with_trace(fun mrdb_two_procs_tx_inner_restart_/1, Config,
+                     dbg_tr_opts()).
+
+mrdb_two_procs_tx_inner_restart_(Config) ->
+    R = ?FUNCTION_NAME,
+    Parent = self(),
+    Created = create_tabs([{R, []}], Config),
+    mrdb:insert(R, {R, a, 1}),
+    mrdb:insert(R, {R, b, 1}),
+    PreA = mrdb:read(R, a),
+    PreB = mrdb:read(R, b),
+    #{inner_retries := Ri0, outer_retries := Ro0} = check_stats(rdb),
+    F0 = fun() ->
+                 wait_for_other(Parent, ?LINE),
+                 ok = mrdb:insert(R, {R, a, 17}),
+                 wait_for_other(Parent, ?LINE)
+         end,
+    F1 = fun() ->
+                 wait_for_other(Parent, ?LINE),
+                 ok = mrdb:insert(R, {R, b, 147}),
+                 wait_for_other(Parent, ?LINE)
+         end,
+
+    Spawn = fun(F) ->
+                    Res = spawn_opt(
+                            fun() ->
+                                    ok = mrdb:activity(tx, rdb, F)
+                            end, [monitor]),
+                    Res
+            end,
+    {P1, MRef1} = Spawn(F0),
+    {P2, MRef2} = Spawn(F1),
+    F2 = fun() ->
+                 PostA = [{R, a, 17}],  % once F0 (P1) has committed
+                 PostB = [{R, b, 147}], % once F1 (P2) has committed
+                 ARes = mrdb:read(R, a),  % We need to read first to establish a pre-image
+                 BRes = mrdb:read(R, b),
+                 Att = get_attempt(),
+                 ct:log("Att = ~p", [Att]),
+                 case Att of
+                     0 ->                         % This is the first run (no retry yet)
+                         ARes = PreA,
+                         BRes = PreB,
+                         go_ahead_other(0, P1),   % Having our pre-image, now let P1 write
+                         go_ahead_other(0, P1),   % Let P1 commit, then await DOWN (normal)
+                         await_other_down(P1, MRef1, ?LINE),
+                         PostA = mrdb:read(R, a); % now, P1's write should leak through here
+                     {0, 1} ->                    % This is the first (outer) retry
+                         go_ahead_other(0, P2),   % Let P2 write
+                         go_ahead_other(0, P2),   % Let P2 commit, then await DOWN (normal)
+                         await_other_down(P2, MRef2, ?LINE),
+                         PostA = mrdb:read(R, a), % now we should see writes from both P1
+                         %% On OTP 23, the following step might fail due to timing, even
+                         %% though the trace output looks as expected. Possibly some quirk
+                         %% with leakage propagation time in rocksdb. Let's retry to be sure.
+                         ok = try_until(PostB, fun() -> mrdb:read(R, b) end); % ... and P2
+                     {1, 1} ->
+                         PostA = mrdb:read(R, a),
+                         PostB = mrdb:read(R, b),
+                         ok
+                 end,
+                 mrdb:insert(R, {R, a, 18}),
+                 mrdb:insert(R, {R, b, 18})
+         end,
+    Do0 = get_dict(),
+    mrdb:activity({tx, #{no_snapshot => true, retries => {1,1}}}, rdb, F2),
+    check_stats(rdb),
+    dictionary_unchanged(Do0),
+    [{R, a, 18}] = mrdb:read(R, a),
+    [{R, b, 18}] = mrdb:read(R, b),
+    #{inner_retries := Ri1, outer_retries := Ro1} = check_stats(rdb),
+    {restarts, {1, 1}} = {restarts, {Ri1 - Ri0, Ro1 - Ro0}},
+    delete_tabs(Created),
+    ok.
+
+try_until(Result, F) ->
+    try_until(Result, F, 10).
+
+try_until(Result, F, N) when N > 0 ->
+    case F() of
+        Result ->
+            ok;
+        _ ->
+            receive after 100 -> ok end,
+            try_until(Result, F, N-1)
+    end;
+try_until(Result, F, _) ->
+    error({badmatch, {Result, F}}).
 
 %
 %% For testing purposes, we use side-effects inside the transactions
@@ -383,7 +488,7 @@ mrdb_two_procs_tx_restart_(Config) ->
 %% attempt, and ignore the sync ops on retries.
 %%
 -define(IF_FIRST(N, Expr),
-        if N == 1 ->
+        if N == 0 ->
                 Expr;
            true ->
                 ok
@@ -413,8 +518,8 @@ mrdb_two_procs_snap(Config) ->
                  go_ahead_other(Att, POther),
                  ARes = mrdb:read(R, a),
                  ARes = case Att of
-                            1 -> Pre;
-                            2 -> [{R, a, 17}]
+                            0 -> Pre;
+                            _ -> [{R, a, 17}]
                         end,
                  await_other_down(POther, MRef, ?LINE),
                  PreB = mrdb:read(R, b),
@@ -434,7 +539,7 @@ mrdb_two_procs_snap(Config) ->
 %% We make sure that P2 commits before finishing the other two, and P3 and the 
 %% main thread sync, so as to maximize the contention for the retry lock.
 mrdb_three_procs(Config) ->
-    tr_ct:with_trace(fun mrdb_three_procs_/1, Config, light_tr_opts()).
+    tr_ct:with_trace(fun mrdb_three_procs_/1, Config, dbg_tr_opts()).
 
 mrdb_three_procs_(Config) ->
     R = ?FUNCTION_NAME,
@@ -452,7 +557,7 @@ mrdb_three_procs_(Config) ->
         spawn_opt(fun() ->
                           D0 = get_dict(),
                           do_when_p_allows(
-                            1, Parent, ?LINE,
+                            0, Parent, ?LINE,
                             fun() ->
                                     ok = mrdb:activity({tx,#{retries => 0}}, rdb, F1)
                             end),
@@ -488,8 +593,8 @@ mrdb_three_procs_(Config) ->
                        fun() ->
                                Att = get_attempt(),
                                ARes = case Att of
-                                          1 -> [A0];
-                                          2 -> [A1]
+                                          0 -> [A0];
+                                          _ -> [A1]
                                       end,
                                %% First, ensure that P2 tx is running
                                go_ahead_other(Att, P2),
@@ -519,6 +624,7 @@ tr_opts() ->
                   , {?MODULE, await_other_down, 3, x}
                   , {?MODULE, do_when_p_allows, 4, x}
                   , {?MODULE, allow_p, 3, x}
+                  , {?MODULE, check_stats, 1, x}
                   ]}.
 
 light_tr_opts() ->
@@ -529,6 +635,23 @@ light_tr_opts() ->
               , {mrdb, read, 2, x}
               , {mrdb, activity, x} ], tr_opts())).
 
+dbg_tr_opts() ->
+    tr_flags(
+      {self(), [call, sos, p, 'receive']},
+      tr_patterns(
+        mrdb, [ {mrdb, insert, 2, x}
+              , {mrdb, read, 2, x}
+              , {mrdb, retry_activity, 3, x}
+              , {mrdb, try_f, 2, x}
+              , {mrdb, incr_attempt, 2, x}
+              , {mrdb, current_context, 0, x}
+              , {mrdb_mutex, do, 2, x}
+              , {mrdb_mutex_serializer, do, 2, x}
+              , {?MODULE, wait_for_other, 2, x}
+              , {?MODULE, go_ahead_other, 1, x}
+              , {?MODULE, try_until, 3, x}
+              , {mrdb, activity, x} ], tr_opts())).
+
 tr_patterns(Mod, Ps, #{patterns := Pats} = Opts) ->
     Pats1 = [P || P <- Pats, element(1,P) =/= Mod],
     Opts#{patterns => Ps ++ Pats1}.
@@ -542,12 +665,13 @@ wait_for_other(Parent, L) ->
 wait_for_other(Att, Parent, L) ->
     wait_for_other(Att, Parent, 1000, L).
 
-wait_for_other(1, Parent, Timeout, L) ->
+wait_for_other(0, Parent, Timeout, L) ->
     MRef = monitor(process, Parent),
     Parent ! {self(), ready},
     receive
         {Parent, cont} ->
             demonitor(MRef),
+            Parent ! {self(), cont_ack},
             ok;
         {'DOWN', MRef, _, _, Reason} ->
             ct:log("Parent died, Reason = ~p", [Reason]),
@@ -586,7 +710,13 @@ go_ahead_other(Att, POther, Timeout) ->
 go_ahead_other_(POther, Timeout) ->
     receive
         {POther, ready} ->
-            POther ! {self(), cont}
+            POther ! {self(), cont},
+            receive
+                {POther, cont_ack} ->
+                    ok
+            after Timeout ->
+                    error(cont_ack_timeout)
+            end
     after Timeout ->
             error(go_ahead_timeout)
     end.
@@ -645,3 +775,8 @@ dictionary_unchanged(Old) ->
      , added   := [] } = #{ deleted => Old -- New
                           , added   => New -- Old },
     ok.
+
+check_stats(Alias) ->
+    Stats = mrdb_stats:get(Alias),
+    ct:log("Stats: ~p", [Stats]),
+    Stats.

test/mnesia_rocksdb_indexes_SUITE.erl

@@ -33,6 +33,7 @@
 -export([
           index_plugin_mgmt/1
         , add_indexes/1
+        , delete_indexes/1
         , create_bag_index/1
         , create_ordered_index/1
         , test_1_ram_copies/1
@@ -40,10 +41,12 @@
         , fail_1_disc_only/1
         , plugin_ram_copies1/1
         , plugin_ram_copies2/1
+        , plugin_ram_copies3/1
         , plugin_disc_copies/1
         , fail_plugin_disc_only/1
         , plugin_disc_copies_bag/1
         , plugin_rdb_ordered/1
+        , plugin_rdb_none/1
         , index_iterator/1
         ]).
 
@@ -73,10 +76,12 @@ run(Config) ->
                     {pfx},mnesia_rocksdb, ix_prefixes),
     test_index_plugin(cfg([pr1, ram_copies, ordered], Config)),
     test_index_plugin(cfg([pr2, ram_copies, bag], Config)),
+    test_index_plugin(cfg([pr3, ram_copies, none], Config)),
     test_index_plugin(cfg([pd1, disc_copies, ordered], Config)),
     fail(test_index_plugin, [cfg([pd2, disc_only_copies, ordered], Config)]),
     test_index_plugin(cfg([pd2, disc_copies, bag], Config)),
     test_index_plugin(cfg([pl2, rdb, ordered], Config)),
+    test_index_plugin(cfg([pl3, rdb, none], Config)),
     index_plugin_mgmt(Config),
     ok.
 
@@ -94,6 +99,7 @@ groups() ->
                          , create_ordered_index
                          , index_plugin_mgmt
                          , add_indexes
+                         , delete_indexes
                          ]}
     , {access, [sequence], [
                              test_1_ram_copies
@@ -104,10 +110,12 @@ groups() ->
     , {plugin, [sequence], [
                              plugin_ram_copies1
                            , plugin_ram_copies2
+                           , plugin_ram_copies3
                            , plugin_disc_copies
                            , fail_plugin_disc_only
                            , plugin_disc_copies_bag
                            , plugin_rdb_ordered
+                           , plugin_rdb_none
                            ]}
     ].
 
@@ -177,11 +185,14 @@ fail_1_disc_only(  _Cfg) -> fail(test, [1, disc_only_copies, do1]).
 
 plugin_ram_copies1(Cfg) -> test_index_plugin(cfg([pr1, ram_copies, ordered], Cfg)).
 plugin_ram_copies2(Cfg) -> test_index_plugin(cfg([pr2, ram_copies, bag], Cfg)).
+plugin_ram_copies3(Cfg) -> test_index_plugin(cfg([pr3, ram_copies, none], Cfg)).
 plugin_disc_copies(Cfg) -> test_index_plugin(cfg([pd1, disc_copies, ordered], Cfg)).
 fail_plugin_disc_only(Cfg) -> fail(test_index_plugin, [cfg([pd2, disc_only_copies, ordered], Cfg)]).
 plugin_disc_copies_bag(Cfg) -> test_index_plugin(cfg([pd2, disc_copies, bag], Cfg)).
 plugin_rdb_ordered(Cfg) -> test_index_plugin(cfg([pl2, rdb, ordered], Cfg)).
 
+plugin_rdb_none(Cfg) -> test_index_plugin(cfg([pl3, rdb, none], Cfg)).
+
 test(N, Type, T) ->
     {atomic, ok} = mnesia:create_table(T, [{Type,[node()]},
                                            {attributes,[k,a,b,c]},
@@ -204,7 +215,7 @@ add_del_indexes() ->
 test_index_plugin(Config) ->
     #{tab := Tab, type := Type, ixtype := IxType} = cfg(Config),
     {atomic, ok} = mnesia:create_table(Tab, [{Type, [node()]},
-                                             {index, [{{pfx}, IxType}]}]),
+                                             {index, [ixtype(IxType)]}]),
     mnesia:dirty_write({Tab, "foobar", "sentence"}),
     mnesia:dirty_write({Tab, "yellow", "sensor"}),
     mnesia:dirty_write({Tab, "truth", "white"}),
@@ -216,13 +227,27 @@ test_index_plugin(Config) ->
             Res2 = lists:sort(mnesia:dirty_index_read(Tab,<<"whi">>, {pfx})),
             [{Tab,"foobar","sentence"}] = mnesia:dirty_index_read(
                                             Tab, <<"foo">>, {pfx});
-       IxType == ordered ->
+       IxType == ordered; IxType == none ->
             Res1 = lists:sort(mnesia:dirty_index_read(Tab,<<"sen">>, {pfx})),
             Res2 = lists:sort(mnesia:dirty_index_read(Tab,<<"whi">>, {pfx})),
             [{Tab,"foobar","sentence"}] = mnesia:dirty_index_read(
                                             Tab, <<"foo">>, {pfx})
+    end,
+    if Type == rdb ->
+            Res1 = lists:sort(mrdb:index_read(Tab,<<"sen">>, {pfx})),
+            Res2 = lists:sort(mrdb:index_read(Tab,<<"whi">>, {pfx})),
+            [{Tab,"foobar","sentence"}] = mrdb:index_read(
+                                            Tab, <<"foo">>, {pfx});
+       true ->
+            ok
     end.
 
+ixtype(T) when T==bag;
+               T==ordered ->
+    {{pfx}, T};
+ixtype(none) ->
+    {pfx}.
+
 create_bag_index(_Config) ->
     {aborted, {combine_error, _, _}} =
         mnesia:create_table(bi, [{rdb, [node()]}, {index, [{val, bag}]}]),
@@ -239,6 +264,25 @@ add_indexes(_Config) ->
     {atomic, ok} = mnesia:add_table_index(T, a),
     ok.
 
+delete_indexes(_Config) ->
+    T = ?TAB(t1),
+    {atomic, ok} = mnesia:create_table(T, [{rdb, [node()]}, {attributes, [k, a, b]}]),
+    ok = mrdb:insert(T, {T, a, 1, 1}),
+    ok = mrdb:insert(T, {T, b, 1, 2}),
+    {atomic, ok} = mnesia:add_table_index(T, a),
+    [{1, {T,a,1,1}}, {1, {T,b,1,2}}] =
+        mrdb_index:rev_fold(T, a, fun ix_fold_acc/3, []),
+    {atomic, ok} = mnesia:del_table_index(T, a),
+    ok = mrdb:delete(T, b),
+    ok = mrdb:insert(T, {T, c, 2, 3}),
+    {atomic, ok} = mnesia:add_table_index(T, a),
+    [{1, {T,a,1,1}}, {2, {T,c,2,3}}] =
+        mrdb_index:rev_fold(T, a, fun ix_fold_acc/3, []),
+    ok.
+
+ix_fold_acc(K, V, Acc) ->
+    [{K, V} | Acc].
+
 index_plugin_mgmt(_Config) ->
     {aborted,_} = mnesia:create_table(x, [{index,[{unknown}]}]),
     {aborted,_} = mnesia:create_table(x, [{index,[{{unknown},bag}]}]),

test/mnesia_rocksdb_migration_SUITE.erl

@@ -149,8 +149,9 @@ create_migrateable_db(Config) ->
     Config.
 
 fill_tabs(Tabs) ->
+    %% Fill with more than 300, since that's the currently hard-coded chunk size
     lists:foreach(fun(Tab) ->
-                          [mrdb:insert(Tab, {Tab, X, a}) || X <- lists:seq(1,3)]
+                          [mrdb:insert(Tab, {Tab, X, a}) || X <- lists:seq(1,500)]
                   end, Tabs).
 
 create_tabs(Tabs, Config) ->

test/mrdb_bench.erl

@@ -1,6 +1,6 @@
 -module(mrdb_bench).
 
--compile(export_all).
+-compile([export_all, nowarn_export_all]).
 
 init() ->
     mnesia:delete_schema([node()]),