diff options
author | Rickard Green <[email protected]> | 2018-03-07 01:17:21 +0100 |
---|---|---|
committer | Rickard Green <[email protected]> | 2018-03-21 10:27:03 +0100 |
commit | 4bc282d812cc2c49aa3e2d073e96c720f16aa270 (patch) | |
tree | a7b00cd079368590dc09f62a4d5402be157462ca /erts/emulator/test/bif_SUITE.erl | |
parent | 348a4e057db36fac13f1551c0a1c17f0d376da48 (diff) | |
download | otp-4bc282d812cc2c49aa3e2d073e96c720f16aa270.tar.gz otp-4bc282d812cc2c49aa3e2d073e96c720f16aa270.tar.bz2 otp-4bc282d812cc2c49aa3e2d073e96c720f16aa270.zip |
Implementation of true asynchronous signaling between processes
Communication between Erlang processes has conceptually always been
performed through asynchronous signaling. The runtime system
implementation has however previously preformed most operation
synchronously. In a system with only one true thread of execution, this
is not problematic (often the opposite). In a system with multiple threads
of execution (as current runtime system implementation with SMP support)
it becomes problematic. This since it often involves locking of structures
when updating them which in turn cause resource contention. Utilizing
true asynchronous communication often avoids these resource contention
issues.
The case that triggered this change was contention on the link lock due
to frequent updates of the monitor trees during communication with a
frequently used server. The signal order delivery guarantees of the
language makes it hard to change the implementation of only some signals
to use true asynchronous signaling. Therefore the implementations
of (almost) all signals have been changed.
Currently the following signals have been implemented as true
asynchronous signals:
- Message signals
- Exit signals
- Monitor signals
- Demonitor signals
- Monitor triggered signals (DOWN, CHANGE, etc)
- Link signals
- Unlink signals
- Group leader signals
All of the above already defined as asynchronous signals in the
language. The implementation of messages signals was quite
asynchronous to begin with, but had quite strict delivery constraints
due to the ordering guarantees of signals between a pair of processes.
The previously used message queue partitioned into two halves has been
replaced by a more general signal queue partitioned into three parts
that service all kinds of signals. More details regarding the signal
queue can be found in comments in the erl_proc_sig_queue.h file.
The monitor and link implementations have also been completely replaced
in order to fit the new asynchronous signaling implementation as good
as possible. More details regarding the new monitor and link
implementations can be found in the erl_monitor_link.h file.
Diffstat (limited to 'erts/emulator/test/bif_SUITE.erl')
-rw-r--r-- | erts/emulator/test/bif_SUITE.erl | 131 |
1 files changed, 125 insertions, 6 deletions
diff --git a/erts/emulator/test/bif_SUITE.erl b/erts/emulator/test/bif_SUITE.erl index e1b42e5d85..d16c6a320d 100644 --- a/erts/emulator/test/bif_SUITE.erl +++ b/erts/emulator/test/bif_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2005-2016. All Rights Reserved. +%% Copyright Ericsson AB 2005-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -33,7 +33,8 @@ atom_to_binary/1,min_max/1, erlang_halt/1, erl_crash_dump_bytes/1, is_builtin/1, error_stacktrace/1, - error_stacktrace_during_call_trace/1]). + error_stacktrace_during_call_trace/1, + group_leader_prio/1, group_leader_prio_dirty/1]). suite() -> [{ct_hooks,[ts_install_cth]}, @@ -46,7 +47,8 @@ all() -> display, display_string, list_to_utf8_atom, atom_to_binary, binary_to_atom, binary_to_existing_atom, erl_crash_dump_bytes, min_max, erlang_halt, is_builtin, - error_stacktrace, error_stacktrace_during_call_trace]. + error_stacktrace, error_stacktrace_during_call_trace, + group_leader_prio, group_leader_prio_dirty]. %% Uses erlang:display to test that erts_printf does not do deep recursion display(Config) when is_list(Config) -> @@ -825,7 +827,6 @@ error_stacktrace_during_call_trace(Config) when is_list(Config) -> end end, ok. - error_stacktrace_test() -> Types = [apply_const_last, apply_const, apply_last, @@ -963,9 +964,119 @@ do_error_1(call) -> erlang:error(id(oops)). +group_leader_prio(Config) when is_list(Config) -> + group_leader_prio_test(false). + +group_leader_prio_dirty(Config) when is_list(Config) -> + group_leader_prio_test(true). + +group_leader_prio_test(Dirty) -> + %% + %% Unfortunately back in the days node local group_leader/2 was not + %% implemented as sending an asynchronous signal to the process to change + %% group leader for. Instead it has always been synchronously changed, and + %% nothing in the documentation have hinted otherwise... Therefore I do not + %% dare the change this. + %% + %% In order to prevent priority inversion, the priority of the receiver of + %% the group leader signal is elevated while handling incoming signals if + %% the sender has a higher priority than the receiver. This test tests that + %% the priority elevation actually works... + %% + Tester = self(), + Init = erlang:whereis(init), + GL = erlang:group_leader(), + process_flag(priority, max), + {TestProcFun, NTestProcs} + = case Dirty of + false -> + %% These processes will handle all incoming signals + %% by them selves... + {fun () -> + Tester ! {alive, self()}, + receive after infinity -> ok end + end, + 100}; + true -> + %% These processes wont handle incoming signals by + %% them selves since they are stuck on dirty schedulers + %% when we try to change group leader. A dirty process + %% signal handler process (system process) will be notified + %% of the need to handle incoming signals for these processes, + %% and will instead handle the signal for these processes... + {fun () -> + %% The following sends the message '{alive, self()}' + %% to Tester once on a dirty io scheduler, then wait + %% there until the process terminates... + erts_debug:dirty_io(alive_waitexiting, Tester) + end, + erlang:system_info(dirty_io_schedulers)} + end, + TPs = lists:map(fun (_) -> + spawn_opt(TestProcFun, + [link, {priority, normal}]) + end, lists:seq(1, NTestProcs)), + lists:foreach(fun (TP) -> receive {alive, TP} -> ok end end, TPs), + TLs = lists:map(fun (_) -> + spawn_opt(fun () -> tok_loop() end, + [link, {priority, high}]) + end, + lists:seq(1, 2*erlang:system_info(schedulers))), + %% Wait to ensure distribution of high prio processes over schedulers... + receive after 1000 -> ok end, + %% + %% Test that we can get group-leader signals through to normal prio + %% processes from a max prio process even though all schedulers are filled + %% with executing high prio processes. + %% + lists:foreach(fun (_) -> + lists:foreach(fun (TP) -> + erlang:yield(), + %% whitebox -- Enqueue some signals on it + %% preventing us from hogging its main lock + %% and set group-leader directly.... + erlang:demonitor(erlang:monitor(process, TP)), + true = erlang:group_leader(Init, TP), + {group_leader, Init} = process_info(TP, group_leader), + erlang:demonitor(erlang:monitor(process, TP)), + true = erlang:group_leader(GL, TP), + {group_leader, GL} = process_info(TP, group_leader) + end, + TPs) + end, + lists:seq(1,100)), + %% + %% Also test when it is exiting... + %% + lists:foreach(fun (TP) -> + erlang:yield(), + M = erlang:monitor(process, TP), + unlink(TP), + exit(TP, bang), + badarg = try + true = erlang:group_leader(Init, TP) + catch + error : What -> What + end, + receive + {'DOWN', M, process, TP, Reason} -> + bang = Reason + end + end, + TPs), + lists:foreach(fun (TL) -> + M = erlang:monitor(process, TL), + unlink(TL), + exit(TL, bang), + receive + {'DOWN', M, process, TL, Reason} -> + bang = Reason + end + end, + TLs), + ok. - -%% Helpers +%% helpers id(I) -> I. @@ -1005,3 +1116,11 @@ hostname([$@ | Hostname]) -> list_to_atom(Hostname); hostname([_C | Cs]) -> hostname(Cs). + +tok_loop() -> + tok_loop(hej). + +tok_loop(hej) -> + tok_loop(hopp); +tok_loop(hopp) -> + tok_loop(hej). |