fix(push): run the stale DM read check on presence nodes (#3042)

This commit is contained in:
Hampus authored and GitHub committed 2026-09-29 19:51:36 +02:00
1 parent f9108f24ce
commit 071263188a
3 files changed
+180 -37

No files matched your search

@@ -166,11 +166,17 @@ flush_push_buffer(#{push_buffer := []} = State) ->
State; State;
flush_push_buffer(#{push_buffer := Buffer} = State) -> flush_push_buffer(#{push_buffer := Buffer} = State) ->
ok = push:handle_buffered_message_creates([ ok = push:handle_buffered_message_creates([
(maps:get(params, Entry))#{buffered_at => maps:get(buffered_at, Entry, undefined)} (maps:get(params, Entry))#{buffered_at => entry_buffered_at(Entry)}
|| Entry <- lists:reverse(Buffer) || Entry <- lists:reverse(Buffer)
]), ]),
State#{push_buffer := []}. State#{push_buffer := []}.
-spec entry_buffered_at(push_buffer_entry()) -> integer() | undefined.
entry_buffered_at(#{buffered_at := BufferedAt}) ->
BufferedAt;
entry_buffered_at(#{message_id := MessageId}) ->
snowflake_util:extract_timestamp(MessageId).
-spec maybe_update_push_eligibility(state()) -> state(). -spec maybe_update_push_eligibility(state()) -> state().
maybe_update_push_eligibility(State) -> maybe_update_push_eligibility(State) ->
Eligible = push_eligible(State), Eligible = push_eligible(State),
+55 -27
View File
@@ -36,7 +36,7 @@
-define(CNT_READ_STATE_SKIPPED, push_read_state_fetch_skipped). -define(CNT_READ_STATE_SKIPPED, push_read_state_fetch_skipped).
-define(MAX_FETCH_RPCS, 8). -define(MAX_FETCH_RPCS, 8).
-define(READ_STATE_FETCH_TIMEOUT_MS, 2000). -define(READ_STATE_FETCH_TIMEOUT_MS, 2000).
-define(READ_STATE_FETCH_MAX_POOL_PCT, 50). -define(READ_STATE_TABLE, push_read_state_counters).
-define(READ_STATE_FETCH_SLOTS, read_state_fetches_in_flight). -define(READ_STATE_FETCH_SLOTS, read_state_fetches_in_flight).
-define(MAX_READ_STATE_FETCHES, 64). -define(MAX_READ_STATE_FETCHES, 64).
-define(DEFAULT_FETCH_USERS, 2000). -define(DEFAULT_FETCH_USERS, 2000).
@@ -334,28 +334,26 @@ publish_unread(Channels, Stale) ->
-spec publish_unless_read(read_marks(), {read_key(), map()}) -> ok. -spec publish_unless_read(read_marks(), {read_key(), map()}) -> ok.
publish_unless_read(ReadMarks, {{UserId, ChannelId, MessageId}, Params}) -> publish_unless_read(ReadMarks, {{UserId, ChannelId, MessageId}, Params}) ->
case MessageId =< maps:get({UserId, ChannelId}, ReadMarks, 0) of case MessageId =< maps:get({UserId, ChannelId}, ReadMarks, 0) of
true -> bump_counter(?CNT_READ_STATE_SUPPRESSED); true -> bump_read_state_counter(?CNT_READ_STATE_SUPPRESSED, 1);
false -> handle_message_create(Params) false -> handle_message_create(Params)
end. end.
-spec read_state_fetch_allowed(non_neg_integer()) -> boolean(). -spec read_state_fetch_allowed(non_neg_integer()) -> boolean().
read_state_fetch_allowed(Fetches) -> read_state_fetch_allowed(Fetches) ->
case case acquire_read_state_fetch_slot() of
push_worker_pool:utilization_pct() < ?READ_STATE_FETCH_MAX_POOL_PCT andalso
acquire_read_state_fetch_slot()
of
true -> true ->
true; true;
false -> false ->
bump_counter(?CNT_READ_STATE_SKIPPED, Fetches), bump_read_state_counter(?CNT_READ_STATE_SKIPPED, Fetches),
false false
end. end.
-spec acquire_read_state_fetch_slot() -> boolean(). -spec acquire_read_state_fetch_slot() -> boolean().
acquire_read_state_fetch_slot() -> acquire_read_state_fetch_slot() ->
ok = ensure_read_state_table(),
try try
ets:update_counter( ets:update_counter(
?PUSH_COUNTER_TABLE, ?READ_STATE_FETCH_SLOTS, {2, 1}, {?READ_STATE_FETCH_SLOTS, 0} ?READ_STATE_TABLE, ?READ_STATE_FETCH_SLOTS, {2, 1}, {?READ_STATE_FETCH_SLOTS, 0}
) )
of of
InFlight when InFlight > ?MAX_READ_STATE_FETCHES -> InFlight when InFlight > ?MAX_READ_STATE_FETCHES ->
@@ -369,12 +367,41 @@ acquire_read_state_fetch_slot() ->
-spec release_read_state_fetch_slot() -> ok. -spec release_read_state_fetch_slot() -> ok.
release_read_state_fetch_slot() -> release_read_state_fetch_slot() ->
try ets:update_counter(?PUSH_COUNTER_TABLE, ?READ_STATE_FETCH_SLOTS, {2, -1, 0, 0}) of try ets:update_counter(?READ_STATE_TABLE, ?READ_STATE_FETCH_SLOTS, {2, -1, 0, 0}) of
_InFlight -> ok _InFlight -> ok
catch catch
error:badarg -> ok error:badarg -> ok
end. end.
-spec bump_read_state_counter(atom(), non_neg_integer()) -> ok.
bump_read_state_counter(Key, Increment) ->
ok = ensure_read_state_table(),
try ets:update_counter(?READ_STATE_TABLE, Key, {2, Increment}, {Key, 0}) of
_Value -> ok
catch
error:badarg -> ok
end.
-spec ensure_read_state_table() -> ok.
ensure_read_state_table() ->
case ets:whereis(?READ_STATE_TABLE) of
undefined ->
guild_ets_utils:ensure_table(?READ_STATE_TABLE, [
named_table, public, set, {write_concurrency, true}
]);
_Table ->
ok
end.
-spec read_state_counter(atom()) -> non_neg_integer() | unavailable.
read_state_counter(Key) ->
try ets:lookup(?READ_STATE_TABLE, Key) of
[{Key, Value}] when is_integer(Value), Value >= 0 -> Value;
_ -> unavailable
catch
error:badarg -> unavailable
end.
-spec fetch_each_read_mark([{integer(), integer()}], read_marks()) -> read_marks(). -spec fetch_each_read_mark([{integer(), integer()}], read_marks()) -> read_marks().
fetch_each_read_mark([], ReadMarks) -> fetch_each_read_mark([], ReadMarks) ->
ReadMarks; ReadMarks;
@@ -383,7 +410,7 @@ fetch_each_read_mark([{UserId, ChannelId} = Key | Rest], ReadMarks) ->
{ok, ReadMessageId} -> {ok, ReadMessageId} ->
fetch_each_read_mark(Rest, ReadMarks#{Key => ReadMessageId}); fetch_each_read_mark(Rest, ReadMarks#{Key => ReadMessageId});
error -> error ->
bump_counter(?CNT_READ_STATE_SKIPPED, length(Rest)), bump_read_state_counter(?CNT_READ_STATE_SKIPPED, length(Rest)),
ReadMarks ReadMarks
end. end.
@@ -400,7 +427,7 @@ fetch_read_message_id(UserId, ChannelId) ->
{ok, _Data} -> {ok, _Data} ->
{ok, 0}; {ok, 0};
{error, Reason} -> {error, Reason} ->
bump_counter(?CNT_READ_STATE_FAILURES), bump_read_state_counter(?CNT_READ_STATE_FAILURES, 1),
logger:debug("Push: read state fetch failed", #{ logger:debug("Push: read state fetch failed", #{
reason => Reason, user_id => UserId, channel_id => ChannelId reason => Reason, user_id => UserId, channel_id => ChannelId
}), }),
@@ -791,10 +818,10 @@ push_loss_counters() ->
#{ #{
counters => counter_table_status(), counters => counter_table_status(),
worker_pool_dropped => read_counter(?CNT_WORKER_POOL), worker_pool_dropped => read_counter(?CNT_WORKER_POOL),
read_state_suppressed => read_counter(?CNT_READ_STATE_SUPPRESSED), read_state_suppressed => read_state_counter(?CNT_READ_STATE_SUPPRESSED),
read_state_fetch_failures => read_counter(?CNT_READ_STATE_FAILURES), read_state_fetch_failures => read_state_counter(?CNT_READ_STATE_FAILURES),
read_state_fetch_skipped => read_counter(?CNT_READ_STATE_SKIPPED), read_state_fetch_skipped => read_state_counter(?CNT_READ_STATE_SKIPPED),
read_state_fetches_in_flight => read_counter(?READ_STATE_FETCH_SLOTS) read_state_fetches_in_flight => read_state_counter(?READ_STATE_FETCH_SLOTS)
}. }.
-spec counter_table_status() -> live | unavailable. -spec counter_table_status() -> live | unavailable.
@@ -1037,19 +1064,21 @@ a_buffered_dm_without_a_buffer_time_is_published_without_a_read_state_fetch_test
), ),
?assertEqual({1, 0}, Published). ?assertEqual({1, 0}, Published).
a_dm_buffered_past_the_outbox_window_is_published_without_a_fetch_when_the_pool_is_busy_test() -> a_dm_buffered_past_the_outbox_window_is_published_without_a_fetch_when_all_slots_are_taken_test() ->
Skipped = read_counter_value(?CNT_READ_STATE_SKIPPED), Skipped = read_counter_value(?CNT_READ_STATE_SKIPPED),
ok = meck:new(push_worker_pool, [passthrough, no_link]), ok = ensure_read_state_table(),
true = ets:insert(?READ_STATE_TABLE, {?READ_STATE_FETCH_SLOTS, ?MAX_READ_STATE_FETCHES}),
try try
ok = meck:expect(push_worker_pool, utilization_pct, fun() -> Published = with_buffered_dm_push(read_through(#{123 => <<"456">>}), fun() ->
?READ_STATE_FETCH_MAX_POOL_PCT ok = handle_buffered_message_creates([
buffered_dm_params(stale_buffered_at(), 123, 456)
]),
length(published_message_ids(0))
end), end),
Published = with_buffered_dms( ?assertEqual({1, 0}, Published),
[{stale_buffered_at(), 123, 456}], read_through(#{123 => <<"456">>}) ?assertEqual(?MAX_READ_STATE_FETCHES, read_state_counter(?READ_STATE_FETCH_SLOTS))
),
?assertEqual({1, 0}, Published)
after after
meck:unload(push_worker_pool) true = ets:insert(?READ_STATE_TABLE, {?READ_STATE_FETCH_SLOTS, 0})
end, end,
?assertEqual(Skipped + 1, read_counter_value(?CNT_READ_STATE_SKIPPED)). ?assertEqual(Skipped + 1, read_counter_value(?CNT_READ_STATE_SKIPPED)).
@@ -1105,8 +1134,7 @@ read_through(ReadMessageIds) ->
fun(ChannelId) -> {ok, #{<<"last_message_id">> => maps:get(ChannelId, ReadMessageIds)}} end. fun(ChannelId) -> {ok, #{<<"last_message_id">> => maps:get(ChannelId, ReadMessageIds)}} end.
read_counter_value(Key) -> read_counter_value(Key) ->
ok = push_worker_pool:init_counter(), case read_state_counter(Key) of
case read_counter(Key) of
Value when is_integer(Value) -> Value; Value when is_integer(Value) -> Value;
unavailable -> 0 unavailable -> 0
end. end.
@@ -1126,7 +1154,7 @@ with_buffered_dms(Entries, ReadStateReply) ->
wait_for_read_state_fetches(0) -> wait_for_read_state_fetches(0) ->
erlang:error(read_state_fetch_still_in_flight); erlang:error(read_state_fetch_still_in_flight);
wait_for_read_state_fetches(Attempts) -> wait_for_read_state_fetches(Attempts) ->
case read_counter(?READ_STATE_FETCH_SLOTS) of case read_state_counter(?READ_STATE_FETCH_SLOTS) of
InFlight when is_integer(InFlight), InFlight > 0 -> InFlight when is_integer(InFlight), InFlight > 0 ->
timer:sleep(20), timer:sleep(20),
wait_for_read_state_fetches(Attempts - 1); wait_for_read_state_fetches(Attempts - 1);
+118 -9
View File
@@ -85,15 +85,16 @@ a_dm_older_than_the_reply_arriving_after_it_is_not_pushed_test() ->
end), end),
?assertEqual([?REPLY_MSG + 1], Pushed). ?assertEqual([?REPLY_MSG + 1], Pushed).
a_buffered_dm_without_a_buffer_time_is_still_pushed_test() -> a_recent_buffered_dm_without_a_buffer_time_is_pushed_test() ->
Recent = recent_message_id(),
Pushed = with_desktop_presence(fun(PresencePid, _SessionPid) -> Pushed = with_desktop_presence(fun(PresencePid, _SessionPid) ->
sys:replace_state(PresencePid, fun(State) -> sys:replace_state(PresencePid, fun(State) ->
State#{ State#{
push_buffer := [ push_buffer := [
#{ #{
channel_id => ?DM, channel_id => ?DM,
message_id => ?PARTNER_MSG, message_id => Recent,
params => #{message_data => dm_message(?PARTNER_MSG, ?PARTNER)} params => #{message_data => dm_message(Recent, ?PARTNER)}
} }
] ]
} }
@@ -102,7 +103,7 @@ a_buffered_dm_without_a_buffer_time_is_still_pushed_test() ->
sync(PresencePid), sync(PresencePid),
pushed_message_ids([]) pushed_message_ids([])
end), end),
?assertEqual([?PARTNER_MSG], Pushed). ?assertEqual([Recent], Pushed).
a_message_ack_drops_the_buffered_dm_push_test() -> a_message_ack_drops_the_buffered_dm_push_test() ->
Pushed = with_desktop_presence(fun(PresencePid, _SessionPid) -> Pushed = with_desktop_presence(fun(PresencePid, _SessionPid) ->
@@ -198,12 +199,113 @@ stale_read_state_fetches_are_capped_per_node_and_fail_open_past_the_cap_test() -
?assertEqual(64, InFlight), ?assertEqual(64, InFlight),
?assertEqual(Flushes - 64, Published), ?assertEqual(Flushes - 64, Published),
?assertEqual(0, Late), ?assertEqual(0, Late),
?assertEqual(0, read_state_fetches_in_flight()). ?assertEqual(0, read_state_fetches_in_flight()),
?assertEqual(undefined, ets:whereis(push_worker_counter)).
a_read_dm_buffered_past_the_outbox_window_is_not_pushed_by_a_presence_node_test() ->
Self = self(),
{Fetches, Published, Suppressed} = with_presence_node(fun() ->
with_desktop_presence(fun(PresencePid, _SessionPid) ->
ok = meck:expect(push, handle_buffered_message_creates, fun(ParamsList) ->
meck:passthrough([ParamsList])
end),
ok = meck:expect(rpc_client, call, fun(Request, _Timeout) ->
Self ! {fetching, maps:get(<<"channel_id">>, Request)},
{ok, #{<<"last_message_id">> => integer_to_binary(?PARTNER_MSG)}}
end),
dispatch(PresencePid, message_create, dm_message(?PARTNER_MSG, ?PARTNER)),
sys:replace_state(PresencePid, fun(#{push_buffer := Buffer} = State) ->
State#{
push_buffer := [
Entry#{buffered_at => erlang:system_time(millisecond) - 600000}
|| Entry <- Buffer
]
}
end),
gen_server:cast(PresencePid, {presence_update, desktop_request(true)}),
sync(PresencePid),
{Blocked, Immediate} = collect_read_state_traffic(1, [], 0),
Late = collect_published(0),
{Blocked, Immediate + Late, read_state_counters(push_read_state_suppressed)}
end)
end),
?assertEqual([integer_to_binary(?DM)], Fetches),
?assertEqual(0, Published),
?assertEqual([{push_read_state_suppressed, 1}], Suppressed),
?assertEqual(undefined, ets:whereis(push_worker_counter)).
a_read_dm_buffered_before_buffer_times_existed_is_aged_from_its_id_test() ->
Self = self(),
{Fetches, Published, Suppressed} = with_presence_node(fun() ->
with_desktop_presence(fun(PresencePid, _SessionPid) ->
ok = meck:expect(push, handle_buffered_message_creates, fun(ParamsList) ->
meck:passthrough([ParamsList])
end),
ok = meck:expect(rpc_client, call, fun(Request, _Timeout) ->
Self ! {fetching, maps:get(<<"channel_id">>, Request)},
{ok, #{<<"last_message_id">> => integer_to_binary(?PARTNER_MSG)}}
end),
dispatch(PresencePid, message_create, dm_message(?PARTNER_MSG, ?PARTNER)),
sys:replace_state(PresencePid, fun(#{push_buffer := Buffer} = State) ->
State#{push_buffer := [maps:remove(buffered_at, Entry) || Entry <- Buffer]}
end),
gen_server:cast(PresencePid, {presence_update, desktop_request(true)}),
sync(PresencePid),
{Blocked, Immediate} = collect_read_state_traffic(1, [], 0),
Late = collect_published(0),
{Blocked, Immediate + Late, read_state_counters(push_read_state_suppressed)}
end)
end),
?assertEqual([integer_to_binary(?DM)], Fetches),
?assertEqual(0, Published),
?assertEqual([{push_read_state_suppressed, 1}], Suppressed).
recent_message_id() ->
(erlang:system_time(millisecond) - 1420070400000) bsl 22.
with_presence_node(Fun) ->
Self = self(),
ok = delete_table(push_worker_counter),
ok = delete_table(push_read_state_counters),
ok = meck:new(fluxer_gateway_env, [passthrough, no_link]),
ok = meck:new(gateway_node_router, [passthrough, no_link]),
ok = meck:new(rpc_client, [passthrough, no_link]),
try
ok = meck:expect(fluxer_gateway_env, get, fun
(push_enabled) -> true;
(Key) -> meck:passthrough([Key])
end),
ok = meck:expect(gateway_node_router, owner_node_result, fun(_Key, push) ->
Self ! published,
{error, test}
end),
with_ets_owner(Fun)
after
meck:unload(rpc_client),
meck:unload(gateway_node_router),
meck:unload(fluxer_gateway_env)
end.
with_ets_owner(Fun) ->
{ok, Owner} = guild_ets_owner:start_link(),
unlink(Owner),
try
Fun()
after
stop_quietly(Owner)
end.
delete_table(Table) ->
try ets:delete(Table) of
true -> ok
catch
error:badarg -> ok
end.
with_stale_read_state_mocks(Fun) -> with_stale_read_state_mocks(Fun) ->
Self = self(), Self = self(),
ok = push_worker_pool:init_counter(), ok = delete_table(push_worker_counter),
true = ets:delete(push_worker_counter, read_state_fetches_in_flight), ok = delete_table(push_read_state_counters),
ok = meck:new(fluxer_gateway_env, [passthrough, no_link]), ok = meck:new(fluxer_gateway_env, [passthrough, no_link]),
ok = meck:new(gateway_node_router, [passthrough, no_link]), ok = meck:new(gateway_node_router, [passthrough, no_link]),
ok = meck:new(rpc_client, [passthrough, no_link]), ok = meck:new(rpc_client, [passthrough, no_link]),
@@ -224,7 +326,7 @@ with_stale_read_state_mocks(Fun) ->
{error, timeout} {error, timeout}
end end
end), end),
Fun() with_ets_owner(Fun)
after after
meck:unload(rpc_client), meck:unload(rpc_client),
meck:unload(gateway_node_router), meck:unload(gateway_node_router),
@@ -271,11 +373,18 @@ wait_for_released_fetch_slots(Attempts) ->
end. end.
read_state_fetches_in_flight() -> read_state_fetches_in_flight() ->
case ets:lookup(push_worker_counter, read_state_fetches_in_flight) of case read_state_counters(read_state_fetches_in_flight) of
[{_, InFlight}] -> InFlight; [{_, InFlight}] -> InFlight;
[] -> 0 [] -> 0
end. end.
read_state_counters(Key) ->
try
ets:lookup(push_read_state_counters, Key)
catch
error:badarg -> []
end.
an_outbox_read_watermark_filters_a_later_job_for_the_read_message_test() -> an_outbox_read_watermark_filters_a_later_job_for_the_read_message_test() ->
{Sent, Stats} = with_outbox(60000, fun() -> {Sent, Stats} = with_outbox(60000, fun() ->
ok = push_outbox:truncate_read(?USER, ?DM, ?PARTNER_MSG), ok = push_outbox:truncate_read(?USER, ?DM, ?PARTNER_MSG),