aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorDan Gudmundsson <[email protected]>2016-02-12 08:42:06 +0100
committerDan Gudmundsson <[email protected]>2016-02-12 08:42:06 +0100
commite84414ce53c7a1131c2356e81b5417badb1389c8 (patch)
tree545c56466a350d6e11be9ed5da923b9f58ff05b6
parent5c023eaea4b35d1f2b2ccf7809bfeb9bc6384077 (diff)
parentc1f4d4a6d5c189f7cd4b3b073fb47f8dfb8d3fff (diff)
downloadotp-e84414ce53c7a1131c2356e81b5417badb1389c8.tar.gz
otp-e84414ce53c7a1131c2356e81b5417badb1389c8.tar.bz2
otp-e84414ce53c7a1131c2356e81b5417badb1389c8.zip
Merge branch 'dgud/mnesia/del_schema_hangs/OTP-13284' into maint
* dgud/mnesia/del_schema_hangs/OTP-13284: mnesia: let loader check if tablelock is needed mnesia: Avoid deadlock possibility in mnesia:del_table_copy schema
-rw-r--r--lib/mnesia/src/mnesia_controller.erl18
-rw-r--r--lib/mnesia/src/mnesia_loader.erl36
-rw-r--r--lib/mnesia/src/mnesia_locker.erl20
-rw-r--r--lib/mnesia/src/mnesia_monitor.erl8
-rw-r--r--lib/mnesia/src/mnesia_schema.erl182
-rw-r--r--lib/mnesia/test/mnesia_isolation_test.erl71
6 files changed, 150 insertions, 185 deletions
diff --git a/lib/mnesia/src/mnesia_controller.erl b/lib/mnesia/src/mnesia_controller.erl
index 02c175760e..69ccc1d2c9 100644
--- a/lib/mnesia/src/mnesia_controller.erl
+++ b/lib/mnesia/src/mnesia_controller.erl
@@ -78,7 +78,7 @@
change_table_majority/1,
del_active_replica/2,
wait_for_tables/2,
- get_network_copy/2,
+ get_network_copy/3,
merge_schema/0,
start_remote_sender/4,
schedule_late_disc_load/2
@@ -329,14 +329,14 @@ release_schema_commit_lock() ->
unlink(whereis(?SERVER_NAME)).
%% Special for preparation of add table copy
-get_network_copy(Tab, Cs) ->
+get_network_copy(Tid, Tab, Cs) ->
% We can't let the controller queue this one
% because that may cause a deadlock between schema_operations
% and initial tableloadings which both takes schema locks.
% But we have to get copier_done msgs when the other side
% goes down.
call({add_other, self()}),
- Reason = {dumper,add_table_copy},
+ Reason = {dumper,{add_table_copy, Tid}},
Work = #net_load{table = Tab,reason = Reason,cstruct = Cs},
%% I'll need this cause it's linked trough the subscriber
%% might be solved by using monitor in subscr instead.
@@ -775,7 +775,7 @@ handle_call({net_load, Tab, Cs}, From, State) ->
true ->
Worker = #net_load{table = Tab,
opt_reply_to = From,
- reason = {dumper,add_table_copy},
+ reason = {dumper,{add_table_copy, unknown}},
cstruct = Cs
},
add_worker(Worker, State);
@@ -1180,11 +1180,11 @@ handle_info(Done = #loader_done{worker_pid=WPid, table_name=Tab}, State0) ->
Done#loader_done.needs_announce == true,
Done#loader_done.needs_reply == true ->
i_have_tab(Tab),
- %% Should be {dumper,add_table_copy} only
+ %% Should be {dumper,{add_table_copy, _}} only
reply(Done#loader_done.reply_to,
Done#loader_done.reply);
Done#loader_done.needs_reply == true ->
- %% Should be {dumper,add_table_copy} only
+ %% Should be {dumper,{add_table_copy,_}} only
reply(Done#loader_done.reply_to,
Done#loader_done.reply);
Done#loader_done.needs_announce == true, Tab == schema ->
@@ -2148,6 +2148,10 @@ load_table_fun(#net_load{cstruct=Cs, table=Tab, reason=Reason, opt_reply_to=Repl
reply_to = ReplyTo,
reply = {loaded, ok}
},
+ AddTableCopy = case Reason of
+ {dumper,{add_table_copy,_}} -> true;
+ _ -> false
+ end,
if
ReadNode == node() ->
%% Already loaded locally
@@ -2157,7 +2161,7 @@ load_table_fun(#net_load{cstruct=Cs, table=Tab, reason=Reason, opt_reply_to=Repl
Res = mnesia_loader:disc_load_table(Tab, load_local_content),
Done#loader_done{reply = Res, needs_announce = true, needs_sync = true}
end;
- AccessMode == read_only, Reason /= {dumper,add_table_copy} ->
+ AccessMode == read_only, not AddTableCopy ->
fun() -> disc_load_table(Tab, Reason, ReplyTo) end;
true ->
fun() ->
diff --git a/lib/mnesia/src/mnesia_loader.erl b/lib/mnesia/src/mnesia_loader.erl
index da8549be50..41fcd76fcb 100644
--- a/lib/mnesia/src/mnesia_loader.erl
+++ b/lib/mnesia/src/mnesia_loader.erl
@@ -180,8 +180,7 @@ do_get_disc_copy2(Tab, Reason, Storage, Type) when Storage == disc_only_copies -
-define(MAX_RAM_TRANSFERS, (?MAX_RAM_FILE_SIZE div ?MAX_TRANSFER_SIZE) + 1).
-define(MAX_NOPACKETS, 20).
-net_load_table(Tab, Reason, Ns, Cs)
- when Reason == {dumper,add_table_copy} ->
+net_load_table(Tab, {dumper,{add_table_copy, _}}=Reason, Ns, Cs) ->
try_net_load_table(Tab, Reason, Ns, Cs);
net_load_table(Tab, Reason, Ns, _Cs) ->
try_net_load_table(Tab, Reason, Ns, val({Tab, cstruct})).
@@ -233,7 +232,8 @@ do_snmpify(Tab, Us, Storage) ->
set({Tab, {index, snmp}}, Snmp).
%% Start the recieiver
-init_receiver(Node, Tab, Storage, Cs, Reas={dumper,add_table_copy}) ->
+init_receiver(Node, Tab, Storage, Cs, Reas={dumper,{add_table_copy, Tid}}) ->
+ rpc:call(Node, mnesia_lib, set, [{?MODULE, active_trans}, Tid]),
case start_remote_sender(Node, Tab, Storage) of
{SenderPid, TabSize, DetsData} ->
start_receiver(Tab,Storage,Cs,SenderPid,TabSize,DetsData,Reas);
@@ -307,7 +307,7 @@ table_init_fun(SenderPid) ->
end.
%% Add_table_copy get's it's own locks.
-start_receiver(Tab,Storage,Cs,SenderPid,TabSize,DetsData,{dumper,add_table_copy}) ->
+start_receiver(Tab,Storage,Cs,SenderPid,TabSize,DetsData,{dumper,{add_table_copy,_}}) ->
Init = table_init_fun(SenderPid),
case do_init_table(Tab,Storage,Cs,SenderPid,TabSize,DetsData,self(), Init) of
Err = {error, _} ->
@@ -658,9 +658,10 @@ send_table(Pid, Tab, RemoteS) ->
{Init, Chunk} = reader_funcs(UseDetsChunk, Tab, Storage, KeysPerTransfer),
SendIt = fun() ->
- {atomic, ok} = prepare_copy(Pid, Tab, Storage),
+ NeedLock = need_lock(Tab),
+ {atomic, ok} = prepare_copy(Pid, Tab, Storage, NeedLock),
send_more(Pid, 1, Chunk, Init(), Tab),
- finish_copy(Pid, Tab, Storage, RemoteS)
+ finish_copy(Pid, Tab, Storage, RemoteS, NeedLock)
end,
try SendIt() of
@@ -678,10 +679,10 @@ send_table(Pid, Tab, RemoteS) ->
end
end.
-prepare_copy(Pid, Tab, Storage) ->
+prepare_copy(Pid, Tab, Storage, NeedLock) ->
Trans =
fun() ->
- mnesia:lock_table(Tab, load),
+ NeedLock andalso mnesia:lock_table(Tab, load),
mnesia_subscr:subscribe(Pid, {table, Tab}),
update_where_to_write(Tab, node(Pid)),
mnesia_lib:db_fixtable(Storage, Tab, true),
@@ -689,6 +690,21 @@ prepare_copy(Pid, Tab, Storage) ->
end,
mnesia:transaction(Trans).
+
+need_lock(Tab) ->
+ case ?catch_val({?MODULE, active_trans}) of
+ #tid{} = Tid ->
+ %% move_table_copy grabs it's own table-lock
+ %% do not deadlock with it
+ mnesia_lib:unset({?MODULE, active_trans}),
+ case mnesia_locker:get_held_locks(Tab) of
+ [{write, Tid}|_] -> false;
+ _Locks -> true
+ end;
+ _ ->
+ true
+ end.
+
update_where_to_write(Tab, Node) ->
case val({Tab, access_mode}) of
read_only ->
@@ -783,12 +799,12 @@ send_packet(N, Pid, Chunk, {Recs, Cont}) when N < ?MAX_NOPACKETS ->
send_packet(_N, _Pid, _Chunk, DataState) ->
DataState.
-finish_copy(Pid, Tab, Storage, RemoteS) ->
+finish_copy(Pid, Tab, Storage, RemoteS, NeedLock) ->
RecNode = node(Pid),
DatBin = dat2bin(Tab, Storage, RemoteS),
Trans =
fun() ->
- mnesia:read_lock_table(Tab),
+ NeedLock andalso mnesia:read_lock_table(Tab),
A = val({Tab, access_mode}),
mnesia_controller:sync_and_block_table_whereabouts(Tab, RecNode, RemoteS, A),
cleanup_tab_copier(Pid, Storage, Tab),
diff --git a/lib/mnesia/src/mnesia_locker.erl b/lib/mnesia/src/mnesia_locker.erl
index 89feeba2c3..5766f22e92 100644
--- a/lib/mnesia/src/mnesia_locker.erl
+++ b/lib/mnesia/src/mnesia_locker.erl
@@ -22,7 +22,7 @@
-module(mnesia_locker).
-export([
- get_held_locks/0,
+ get_held_locks/0, get_held_locks/1,
get_lock_queue/0,
global_lock/5,
ixrlock/5,
@@ -236,6 +236,11 @@ loop(State) ->
From ! {Ref, ok},
loop(State);
+ {From, {is_locked, Oid}} ->
+ Held = ?ets_lookup(mnesia_held_locks, Oid),
+ reply(From, Held),
+ loop(State);
+
{'EXIT', Pid, _} when Pid == State#state.supervisor ->
do_stop();
@@ -1151,6 +1156,19 @@ get_held_locks() ->
Locks = receive {mnesia_held_locks, Ls} -> Ls after 5000 -> [] end,
rewrite_locks(Locks, []).
+%% Mnesia internal usage only
+get_held_locks(Tab) when is_atom(Tab) ->
+ Oid = {Tab, ?ALL},
+ ?MODULE ! {self(), {is_locked, Oid}},
+ receive
+ {?MODULE, _Node, Locks} ->
+ case Locks of
+ [] -> [];
+ [{Oid, _Prev, What}] -> What
+ end
+ end.
+
+
rewrite_locks([{Oid, _, Ls}|Locks], Acc0) ->
Acc = rewrite_locks(Ls, Oid, Acc0),
rewrite_locks(Locks, Acc);
diff --git a/lib/mnesia/src/mnesia_monitor.erl b/lib/mnesia/src/mnesia_monitor.erl
index 8313c3bda5..081d746257 100644
--- a/lib/mnesia/src/mnesia_monitor.erl
+++ b/lib/mnesia/src/mnesia_monitor.erl
@@ -82,9 +82,9 @@
going_down = [], tm_started = false, early_connects = [],
connecting, mq = [], remote_node_status = []}).
--define(current_protocol_version, {8,1}).
+-define(current_protocol_version, {8,2}).
--define(previous_protocol_version, {8,0}).
+-define(previous_protocol_version, {8,1}).
start() ->
gen_server:start_link({local, ?MODULE}, ?MODULE,
@@ -193,7 +193,7 @@ protocol_version() ->
%% A sorted list of acceptable protocols the
%% preferred protocols are first in the list
acceptable_protocol_versions() ->
- [protocol_version(), ?previous_protocol_version, {7,6}].
+ [protocol_version(), ?previous_protocol_version].
needs_protocol_conversion(Node) ->
case {?catch_val({protocol, Node}), protocol_version()} of
@@ -424,8 +424,6 @@ handle_call({negotiate_protocol, Mon, Version, Protocols}, From, State)
case hd(Protocols) of
?previous_protocol_version ->
accept_protocol(Mon, MyVersion, ?previous_protocol_version, From, State);
- {7,6} ->
- accept_protocol(Mon, MyVersion, {7,6}, From, State);
_ ->
verbose("Connection with ~p rejected. "
"version = ~p, protocols = ~p, "
diff --git a/lib/mnesia/src/mnesia_schema.erl b/lib/mnesia/src/mnesia_schema.erl
index 593360415d..1bd8703e1d 100644
--- a/lib/mnesia/src/mnesia_schema.erl
+++ b/lib/mnesia/src/mnesia_schema.erl
@@ -646,54 +646,18 @@ cs2list(Cs) when is_record(Cs, cstruct) ->
rec2list(Tags, Tags, 2, Cs);
cs2list(CreateList) when is_list(CreateList) ->
CreateList;
-%% 4.6
+
+%% since 4.6
cs2list(Cs) when element(1, Cs) == cstruct, tuple_size(Cs) == 19 ->
Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
load_order,access_mode,majority,index,snmp,local_content,
record_name,attributes,
user_properties,frag_properties,storage_properties,
cookie,version],
- rec2list(Tags, Tags, 2, Cs);
-%% 4.4.19
-cs2list(Cs) when element(1, Cs) == cstruct, tuple_size(Cs) == 18 ->
- Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
- load_order,access_mode,majority,index,snmp,local_content,
- record_name,attributes,user_properties,frag_properties,
- cookie,version],
- rec2list(Tags, Tags, 2, Cs);
-%% 4.4.18 and earlier
-cs2list(Cs) when element(1, Cs) == cstruct, tuple_size(Cs) == 17 ->
- Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
- load_order,access_mode,index,snmp,local_content,
- record_name,attributes,user_properties,frag_properties,
- cookie,version],
rec2list(Tags, Tags, 2, Cs).
cs2list(false, Cs) ->
- cs2list(Cs);
-cs2list(ver4_4_18, Cs) -> %% Or earlier
- Orig = record_info(fields, cstruct),
- Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
- load_order,access_mode,index,snmp,local_content,
- record_name,attributes,user_properties,frag_properties,
- cookie,version],
- rec2list(Tags, Orig, 2, Cs);
-cs2list(ver4_4_19, Cs) ->
- Orig = record_info(fields, cstruct),
- Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
- load_order,access_mode,majority,index,snmp,local_content,
- record_name,attributes,user_properties,frag_properties,
- cookie,version],
- rec2list(Tags, Orig, 2, Cs);
-cs2list(ver4_6, Cs) ->
- Orig = record_info(fields, cstruct),
- Tags = [name,type,ram_copies,disc_copies,disc_only_copies,
- load_order,access_mode,majority,index,snmp,local_content,
- record_name,attributes,
- user_properties,frag_properties,storage_properties,
- cookie,version],
- rec2list(Tags, Orig, 2, Cs).
-
+ cs2list(Cs).
rec2list([Tag | Tags], [Tag | Orig], Pos, Rec) ->
Val = element(Pos, Rec),
@@ -703,19 +667,8 @@ rec2list([], _, _Pos, _Rec) ->
rec2list(Tags, [_|Orig], Pos, Rec) ->
rec2list(Tags, Orig, Pos+1, Rec).
-normalize_cs(Cstructs, Node) ->
- %% backward-compatibility hack; normalize before returning
- case need_old_cstructs([Node]) of
- false ->
- Cstructs;
- Version ->
- %% some other format
- [convert_cs(Version, Cs) || Cs <- Cstructs]
- end.
-
-convert_cs(Version, Cs) ->
- Fields = [Value || {_, Value} <- cs2list(Version, Cs)],
- list_to_tuple([cstruct|Fields]).
+normalize_cs(Cstructs, _Node) ->
+ Cstructs.
list2cs(List) when is_list(List) ->
Name = pick(unknown, name, List, must),
@@ -1297,6 +1250,7 @@ make_del_table_copy(Tab, Node) ->
_ ->
ensure_active(Cs),
verify_cstruct(Cs2),
+ get_tid_ts_and_lock(Tab, write),
[{op, del_table_copy, Storage, Node, vsn_cs2list(Cs2)}]
end.
@@ -1324,6 +1278,7 @@ remove_node_from_tabs([Tab|Rest], Node) ->
remove_node_from_tabs(Rest, Node)];
_Ns ->
verify_cstruct(Cs2),
+ get_tid_ts_and_lock(Tab, write),
[{op, del_table_copy, ram_copies, Node, vsn_cs2list(Cs2)}|
remove_node_from_tabs(Rest, Node)]
end
@@ -1355,6 +1310,11 @@ do_move_table(schema, _FromNode, _ToNode) ->
mnesia:abort({bad_type, schema});
do_move_table(Tab, FromNode, ToNode) when is_atom(FromNode), is_atom(ToNode) ->
TidTs = get_tid_ts_and_lock(schema, write),
+ AnyOld = lists:any(fun(Node) -> mnesia_monitor:needs_protocol_conversion(Node) end,
+ [ToNode|val({Tab, where_to_write})]),
+ if AnyOld -> ignore; %% Leads to deadlock on old nodes
+ true -> get_tid_ts_and_lock(Tab, write)
+ end,
insert_schema_ops(TidTs, make_move_table(Tab, FromNode, ToNode));
do_move_table(Tab, FromNode, ToNode) ->
mnesia:abort({badarg, Tab, FromNode, ToNode}).
@@ -1962,7 +1922,7 @@ prepare_op(Tid, {op, add_table_copy, Storage, Node, TabDef}, _WaitFor) ->
end,
%% Tables are created by mnesia_loader get_network code
insert_cstruct(Tid, Cs, true),
- case mnesia_controller:get_network_copy(Tab, Cs) of
+ case mnesia_controller:get_network_copy(Tid, Tab, Cs) of
{loaded, ok} ->
{true, optional};
{not_loaded, ErrReason} ->
@@ -1993,28 +1953,11 @@ prepare_op(Tid, {op, add_table_copy, Storage, Node, TabDef}, _WaitFor) ->
{true, optional}
end;
-prepare_op(Tid, {op, del_table_copy, _Storage, Node, TabDef}, _WaitFor) ->
+prepare_op(_Tid, {op, del_table_copy, _Storage, Node, TabDef}, _WaitFor) ->
Cs = list2cs(TabDef),
Tab = Cs#cstruct.name,
-
- if
- %% Schema table lock is always required to run a schema op.
- %% No need to look it.
- node(Tid#tid.pid) == node(), Tab /= schema ->
- Self = self(),
- Pid = spawn_link(fun() -> lock_del_table(Tab, Node, Cs, Self) end),
- put(mnesia_lock, Pid),
- receive
- {Pid, updated} ->
- {true, optional};
- {Pid, FailReason} ->
- mnesia:abort(FailReason);
- {'EXIT', Pid, Reason} ->
- mnesia:abort(Reason)
- end;
- true ->
- {true, optional}
- end;
+ set_where_to_read(Tab, Node, Cs),
+ {true, optional};
prepare_op(_Tid, {op, change_table_copy_type, N, FromS, ToS, TabDef}, _WaitFor)
when N == node() ->
@@ -2228,46 +2171,6 @@ receive_sync(Nodes, Pids) ->
{abort, Else}
end.
-lock_del_table(Tab, NewNode, Cs0, Father) ->
- Ns = val({schema, active_replicas}),
- process_flag(trap_exit,true),
- Lock = fun() ->
- mnesia:write_lock_table(Tab),
- %% Sigh using cs record
- Set = fun(Node) ->
- [Cs] = normalize_cs([Cs0], Node),
- rpc:call(Node, ?MODULE, set_where_to_read, [Tab, NewNode, Cs])
- end,
- Res = [Set(Node) || Node <- Ns],
- Filter = fun(ok) ->
- false;
- ({badrpc, {'EXIT', {undef, _}}}) ->
- %% This will be the case we talks with elder nodes
- %% than 3.8.2, they will set where_to_read without
- %% getting a lock.
- false;
- (_) ->
- true
- end,
- case lists:filter(Filter, Res) of
- [] ->
- Father ! {self(), updated},
- %% When transaction is commited the process dies
- %% and the lock is released.
- receive _ -> ok end;
- Err ->
- Father ! {self(), {bad_commit, Err}}
- end,
- ok
- end,
- case mnesia:transaction(Lock) of
- {atomic, ok} -> ok;
- {aborted, R} -> Father ! {self(), R}
- end,
- unlink(Father),
- unlink(whereis(mnesia_tm)),
- exit(normal).
-
set_where_to_read(Tab, Node, Cs) ->
case mnesia_lib:val({Tab, where_to_read}) of
Node ->
@@ -2418,13 +2321,19 @@ undo_prepare_op(Tid, {op, add_table_copy, Storage, Node, TabDef}) ->
insert_cstruct(Tid, Cs2, true) % Don't care about the version
end;
-undo_prepare_op(_Tid, {op, del_table_copy, _, Node, TabDef})
- when Node == node() ->
- WriteLocker = get(mnesia_lock),
- WriteLocker =/= undefined andalso (WriteLocker ! die),
+undo_prepare_op(_Tid, {op, del_table_copy, _, Node, TabDef}) ->
Cs = list2cs(TabDef),
Tab = Cs#cstruct.name,
- mnesia_lib:set({Tab, where_to_read}, Node);
+ if node() =:= Node ->
+ mnesia_lib:set({Tab, where_to_read}, Node);
+ true ->
+ case mnesia_lib:val({Tab, where_to_read}) of
+ nowhere ->
+ mnesia_lib:set_remote_where_to_read(Tab);
+ _ ->
+ ignore
+ end
+ end;
undo_prepare_op(_Tid, {op, change_table_copy_type, N, FromS, ToS, TabDef})
when N == node() ->
@@ -2885,40 +2794,11 @@ do_merge_schema(LockTabs0) ->
end.
fetch_cstructs(Node) ->
- case need_old_cstructs([Node]) of
- false ->
- rpc:call(Node, mnesia_controller, get_remote_cstructs, []);
- _Ver ->
- case rpc:call(Node, mnesia_controller, get_cstructs, []) of
- {cstructs, Cs0, RR} ->
- {cstructs, [list2cs(cs2list(Cs)) || Cs <- Cs0], RR};
- Err -> Err
- end
- end.
+ rpc:call(Node, mnesia_controller, get_remote_cstructs, []).
-need_old_cstructs() ->
- need_old_cstructs(val({schema, where_to_write})).
-
-need_old_cstructs(Nodes) ->
- Filter = fun(Node) -> not mnesia_monitor:needs_protocol_conversion(Node) end,
- case lists:dropwhile(Filter, Nodes) of
- [] -> false;
- [Node|_] ->
- case rpc:call(Node, mnesia_lib, val, [{schema,cstruct}]) of
- #cstruct{} ->
- %% mnesia_lib:warning("Mnesia on ~p do not need to convert cstruct (~p)~n",
- %% [node(), Node]),
- false;
- {badrpc, _} ->
- need_old_cstructs(lists:delete(Node,Nodes));
- Cs when element(1, Cs) == cstruct, tuple_size(Cs) == 17 ->
- ver4_4_18; % Without majority
- Cs when element(1, Cs) == cstruct, tuple_size(Cs) == 18 ->
- ver4_4_19; % With majority
- Cs when element(1, Cs) == cstruct, tuple_size(Cs) == 19 ->
- ver4_6 % With storage_properties
- end
- end.
+need_old_cstructs() -> false.
+
+need_old_cstructs(_Nodes) -> false.
tab_to_nodes(Tab) when is_atom(Tab) ->
Cs = val({Tab, cstruct}),
diff --git a/lib/mnesia/test/mnesia_isolation_test.erl b/lib/mnesia/test/mnesia_isolation_test.erl
index b66da6e390..4dd0c01d05 100644
--- a/lib/mnesia/test/mnesia_isolation_test.erl
+++ b/lib/mnesia/test/mnesia_isolation_test.erl
@@ -39,7 +39,7 @@ groups() ->
[{locking, [],
[no_conflict, simple_queue_conflict,
advanced_queue_conflict, simple_deadlock_conflict,
- advanced_deadlock_conflict, lock_burst,
+ advanced_deadlock_conflict, schema_deadlock, lock_burst,
{group, sticky_locks}, {group, unbound_locking},
{group, admin_conflict}, nasty]},
{sticky_locks, [], [basic_sticky_functionality]},
@@ -148,20 +148,32 @@ simple_queue_conflict(Config) when is_list(Config) ->
fun_loop(Fun, AllSharedLocks, OneExclusiveLocks),
ok.
-wait_for_lock(Pid, _Nodes, 0) ->
+wait_for_lock(Pid, Nodes, Retry) ->
+ wait_for_lock(Pid, Nodes, Retry, queue).
+
+wait_for_lock(Pid, _Nodes, 0, queue) ->
Queue = mnesia:system_info(lock_queue),
?error("Timeout while waiting for lock on Pid ~p in queue ~p~n", [Pid, Queue]);
-wait_for_lock(Pid, Nodes, N) ->
- rpc:multicall(Nodes, sys, get_status, [mnesia_locker]),
- List = [rpc:call(Node, mnesia, system_info, [lock_queue]) || Node <- Nodes],
+wait_for_lock(Pid, _Nodes, 0, held) ->
+ Held = mnesia:system_info(held_locks),
+ ?error("Timeout while waiting for lock on Pid ~p (held) ~p~n", [Pid, Held]);
+wait_for_lock(Pid, Nodes, N, Where) ->
+ rpc:multicall(Nodes, sys, get_status, [mnesia_locker]),
+ List = case Where of
+ queue ->
+ [rpc:call(Node, mnesia, system_info, [lock_queue]) || Node <- Nodes];
+ held ->
+ [rpc:call(Node, mnesia, system_info, [held_locks]) || Node <- Nodes]
+ end,
Q = lists:append(List),
- check_q(Pid, Q, Nodes, N).
+ check_q(Pid, Q, Nodes, N, Where).
-check_q(Pid, [{_Oid, _Op, Pid, _Tid, _WFT} | _Tail], _N, _Count) -> ok;
-check_q(Pid, [_ | Tail], N, Count) -> check_q(Pid, Tail, N, Count);
-check_q(Pid, [], N, Count) ->
- timer:sleep(500),
- wait_for_lock(Pid, N, Count - 1).
+check_q(Pid, [{_Oid, _Op, Pid, _Tid, _WFT} | _Tail], _N, _Count, _Where) -> ok;
+check_q(Pid, [{_Oid, _Op, {tid,_,Pid}} | _Tail], _N, _Count, _Where) -> ok;
+check_q(Pid, [_ | Tail], N, Count, Where) -> check_q(Pid, Tail, N, Count, Where);
+check_q(Pid, [], N, Count, Where) ->
+ timer:sleep(200),
+ wait_for_lock(Pid, N, Count - 1, Where).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@@ -270,6 +282,43 @@ advanced_deadlock_conflict(Config) when is_list(Config) ->
?match([], mnesia:system_info(lock_queue)),
ok.
+%% Verify (and regression test) deadlock in del_table_copy(schema, Node)
+schema_deadlock(Config) when is_list(Config) ->
+ Ns = [Node1, Node2] = ?acquire_nodes(2, Config),
+ ?match({atomic, ok}, mnesia:create_table(a, [{disc_copies, Ns}])),
+ ?match({atomic, ok}, mnesia:create_table(b, [{disc_copies, Ns}])),
+
+ Tester = self(),
+
+ Deadlocker = fun() ->
+ mnesia:write({a,1,1}), %% grab write lock on A
+ receive
+ continue ->
+ mnesia:write({b,1,1}), %% grab write lock on B
+ end_trans
+ end
+ end,
+
+ ?match(stopped, rpc:call(Node2, mnesia, stop, [])),
+ timer:sleep(500), %% Let Node1 reconfigure
+ sys:get_status(mnesia_monitor),
+
+ DoingTrans = spawn_link(fun() -> Tester ! {self(),mnesia:transaction(Deadlocker)} end),
+ wait_for_lock(DoingTrans, [Node1], 10, held),
+ %% Will grab write locks on schema, a, and b
+ DoingSchema = spawn_link(fun() -> Tester ! {self(), mnesia:del_table_copy(schema, Node2)} end),
+ timer:sleep(500), %% Let schema trans start, and try to grab locks
+ DoingTrans ! continue,
+
+ ?match(ok, receive {DoingTrans, {atomic, end_trans}} -> ok after 5000 -> timeout end),
+ ?match(ok, receive {DoingSchema, {atomic, ok}} -> ok after 5000 -> timeout end),
+
+ sys:get_status(whereis(mnesia_locker)), % Explicit sync, release locks is async
+ ?match([], mnesia:system_info(held_locks)),
+ ?match([], mnesia:system_info(lock_queue)),
+ ok.
+
+
one_oid(Tab) -> {Tab, 1}.
other_oid(Tab) -> {Tab, 2}.