fix(buffer_worker): check request timeout and health check interval

Fixes https://emqx.atlassian.net/browse/EMQX-9099

The default value for `request_timeout` is 15 seconds, and the default
resume interval is also 15 seconds (the health check timeout, if
`resume_interval` is not explicitly given).  This means that, in
practice, if a buffer worker ever gets into the blocked state, then
almost all requests will timeout.

Proposed improvement:

- `request_timeout` should by default be twice as much as
health_check_interval.
- Emit a alarm if `request_timeout` is not greater than
`health_check_interval`.
This commit is contained in:
Thales Macedo Garitezi 2023-03-16 13:33:52 -03:00
parent ef8f764154
commit 20414d7373
6 changed files with 178 additions and 3 deletions

View File

@ -102,8 +102,8 @@ For bridges only have ingress direction data flow, it can be set to 0 otherwise
request_timeout {
desc {
en: """Starting from the moment when the request enters the buffer, if the request remains in the buffer for the specified time or is sent but does not receive a response or acknowledgement in time, the request is considered expired."""
zh: """从请求进入缓冲区开始计时,如果请求在规定的时间内仍停留在缓冲区内或者已发送但未能及时收到响应或确认,该请求将被视为过期。"""
en: """Starting from the moment when the request enters the buffer, if the request remains in the buffer for the specified time or is sent but does not receive a response or acknowledgement in time, the request is considered expired. We recommend setting this timeout to be at least twice the health check interval, so that the buffer has the chance to recover if too many requests get enqueued."""
zh: """从请求进入缓冲区开始计时,如果请求在规定的时间内仍停留在缓冲区内或者已发送但未能及时收到响应或确认,该请求将被视为过期。我们建议将这个超时设置为健康检查间隔的至少两倍,这样,如果有太多的请求被排队,缓冲区就有机会恢复。"""
}
label {
en: """Request Expiry"""

View File

@ -91,7 +91,10 @@
-define(DEFAULT_QUEUE_SIZE, 100 * 1024 * 1024).
-define(DEFAULT_QUEUE_SIZE_RAW, <<"100MB">>).
-define(DEFAULT_REQUEST_TIMEOUT, timer:seconds(15)).
%% Note: this should be greater than the health check timeout;
%% otherwise, if the buffer worker is ever blocked, than all queued
%% requests will basically fail without being attempted.
-define(DEFAULT_REQUEST_TIMEOUT, timer:seconds(30)).
%% count
-define(DEFAULT_BATCH_SIZE, 1).

View File

@ -56,6 +56,8 @@
-export([clear_disk_queue_dir/2]).
-export([deactivate_bad_request_timeout_alarm/1]).
-elvis([{elvis_style, dont_repeat_yourself, disable}]).
-define(COLLECT_REQ_LIMIT, 1000).
@ -88,6 +90,8 @@
-type queue_query() :: ?QUERY(reply_fun(), request(), HasBeenSent :: boolean(), expire_at()).
-type request() :: term().
-type request_from() :: undefined | gen_statem:from().
-type request_timeout() :: infinity | timer:time().
-type health_check_interval() :: timer:time().
-type state() :: blocked | running.
-type inflight_key() :: integer().
-type data() :: #{
@ -199,6 +203,7 @@ init({Id, Index, Opts}) ->
RequestTimeout = maps:get(request_timeout, Opts, ?DEFAULT_REQUEST_TIMEOUT),
BatchTime0 = maps:get(batch_time, Opts, ?DEFAULT_BATCH_TIME),
BatchTime = adjust_batch_time(Id, RequestTimeout, BatchTime0),
maybe_toggle_bad_request_timeout_alarm(Id, RequestTimeout, HealthCheckInterval),
Data = #{
id => Id,
index => Index,
@ -1679,6 +1684,39 @@ adjust_batch_time(Id, RequestTimeout, BatchTime0) ->
end,
BatchTime.
%% The request timeout should be greater than the health check
%% timeout, health timeout defines how often the buffer worker tries
%% to unblock. If request timeout is <= health check timeout and the
%% buffer worker is ever blocked, than all queued requests will
%% basically fail without being attempted.
-spec maybe_toggle_bad_request_timeout_alarm(
resource_id(), request_timeout(), health_check_interval()
) -> ok.
maybe_toggle_bad_request_timeout_alarm(Id, _RequestTimeout = infinity, _HealthCheckInterval) ->
deactivate_bad_request_timeout_alarm(Id),
ok;
maybe_toggle_bad_request_timeout_alarm(Id, RequestTimeout, HealthCheckInterval) ->
case RequestTimeout > HealthCheckInterval of
true ->
deactivate_bad_request_timeout_alarm(Id),
ok;
false ->
_ = emqx_alarm:activate(
bad_request_timeout_alarm_id(Id),
#{resource_id => Id, reason => bad_request_timeout},
<<"Request timeout should be greater than health check timeout: ", Id/binary>>
),
ok
end.
-spec deactivate_bad_request_timeout_alarm(resource_id()) -> ok.
deactivate_bad_request_timeout_alarm(Id) ->
_ = emqx_alarm:ensure_deactivated(bad_request_timeout_alarm_id(Id)),
ok.
bad_request_timeout_alarm_id(Id) ->
<<"bad_request_timeout:", Id/binary>>.
-ifdef(TEST).
-include_lib("eunit/include/eunit.hrl").
adjust_batch_time_test_() ->

View File

@ -506,6 +506,7 @@ handle_remove_event(From, ClearMetrics, Data) ->
true -> ok = emqx_metrics_worker:clear_metrics(?RES_METRICS, Data#data.id);
false -> ok
end,
emqx_resource_buffer_worker:deactivate_bad_request_timeout_alarm(Data#data.id),
{stop_and_reply, {shutdown, removed}, [{reply, From, ok}]}.
start_resource(Data, From) ->

View File

@ -2554,6 +2554,132 @@ do_t_recursive_flush() ->
),
ok.
%% Check that we raise an alarm if a bad request timeout config is
%% issued. Request timeout should be greater than health check
%% timeout.
t_bad_request_timeout_alarm(_Config) ->
emqx_connector_demo:set_callback_mode(async_if_possible),
%% 1) Same values.
{ok, _} = emqx_resource:create(
?ID,
?DEFAULT_RESOURCE_GROUP,
?TEST_RESOURCE,
#{name => test_resource},
#{
query_mode => async,
request_timeout => 1_000,
health_check_interval => 1_000,
worker_pool_size => 2
}
),
ExpectedMessage =
<<"Request timeout should be greater than health check timeout: ", ?ID/binary>>,
?assertMatch(
[
#{
message := ExpectedMessage,
details := #{reason := bad_request_timeout, resource_id := ?ID},
deactivate_at := infinity
}
],
emqx_alarm:get_alarms(activated)
),
%% The unexpected termination of one of the buffer workers should
%% not turn the alarm off.
[Pid, _ | _] = emqx_resource_buffer_worker_sup:worker_pids(?ID),
MRef = monitor(process, Pid),
exit(Pid, kill),
receive
{'DOWN', MRef, process, Pid, _} ->
ok
after 300 ->
ct:fail("buffer worker didn't die")
end,
?assertMatch(
[
#{
message := ExpectedMessage,
details := #{reason := bad_request_timeout, resource_id := ?ID},
deactivate_at := infinity
}
],
emqx_alarm:get_alarms(activated)
),
ok = emqx_resource:remove_local(?ID),
?assertEqual([], emqx_alarm:get_alarms(activated)),
%% 2) Request timeout < health check interval.
{ok, _} = emqx_resource:create(
?ID,
?DEFAULT_RESOURCE_GROUP,
?TEST_RESOURCE,
#{name => test_resource},
#{
query_mode => async,
request_timeout => 999,
health_check_interval => 1_000,
worker_pool_size => 2
}
),
?assertMatch(
[
#{
message := ExpectedMessage,
details := #{reason := bad_request_timeout, resource_id := ?ID},
deactivate_at := infinity
}
],
emqx_alarm:get_alarms(activated)
),
ok = emqx_resource:remove_local(?ID),
?assertEqual([], emqx_alarm:get_alarms(activated)),
%% 2) Request timeout < health check interval.
{ok, _} = emqx_resource:create(
?ID,
?DEFAULT_RESOURCE_GROUP,
?TEST_RESOURCE,
#{name => test_resource},
#{
query_mode => async,
request_timeout => 999,
health_check_interval => 1_000,
worker_pool_size => 2
}
),
?assertMatch(
[
#{
message := ExpectedMessage,
details := #{reason := bad_request_timeout, resource_id := ?ID},
deactivate_at := infinity
}
],
emqx_alarm:get_alarms(activated)
),
ok = emqx_resource:remove_local(?ID),
?assertEqual([], emqx_alarm:get_alarms(activated)),
%% 3) Request timeout > health check interval.
{ok, _} = emqx_resource:create(
?ID,
?DEFAULT_RESOURCE_GROUP,
?TEST_RESOURCE,
#{name => test_resource},
#{
query_mode => async,
request_timeout => 1_001,
health_check_interval => 1_000,
worker_pool_size => 2
}
),
?assertEqual([], emqx_alarm:get_alarms(activated)),
ok = emqx_resource:remove_local(?ID),
?assertEqual([], emqx_alarm:get_alarms(activated)),
ok.
%%------------------------------------------------------------------------------
%% Helpers
%%------------------------------------------------------------------------------

View File

@ -0,0 +1,7 @@
Change the default `request_timeout` for bridges and connectors to be
twice the default `health_check_interval`.
Before this change, the default values for those two options meant
that, if a buffer ever got blocked due to resource errors or high
message volumes, then, by the time the buffer would try to resume its
normal operations, almost all requests would have timed out.