Copilot commented on code in PR #13821:
URL: https://github.com/apache/trafficserver/pull/13821#discussion_r4298375522
##########
src/iocore/net/NetHandler.cc:
##########
@@ -113,9 +126,92 @@ NetHandler::stopCop(NetEvent *ne)
ink_release_assert(ne->nh == this);
open_list.remove(ne);
- cop_list.remove(ne);
remove_from_keep_alive_queue(ne);
remove_from_active_queue(ne);
+ timer_wheel.cancel(ne);
+}
+
+ink_hrtime
+NetHandler::_earliest_deadline(NetEvent *ne) const
+{
+ ink_hrtime const inactivity = ne->next_inactivity_timeout_at;
+ ink_hrtime const activity = ne->next_activity_timeout_at;
Review Comment:
These deadline fields are non-atomic, but foreign threads can update them
through `set_enabled()` and the timeout setters. `Fire::deadline_of()` now
reads them before acquiring the connection mutex, and `process_handoffs()` also
reads them through this helper without that lock. Owner-only wheel updates
therefore do not prevent a C++ data race on the deadlines. Synchronize deadline
reads and writes too, using nonblocking lock-and-retry handling or synchronized
deadline storage; deferring only the wheel linkage is insufficient.
##########
src/iocore/net/NetEvent.cc:
##########
@@ -0,0 +1,61 @@
+/** @file
+
+ NetEvent timer helpers: owner-thread-safe timer rearm and default inactivity
deadline rebasing.
+
+ @section license License
+
+ Licensed to the Apache Software Foundation (ASF) under one
+ or more contributor license agreements. See the NOTICE file
+ distributed with this work for additional information
+ regarding copyright ownership. The ASF licenses this file
+ to you under the Apache License, Version 2.0 (the
+ "License"); you may not use this file except in compliance
+ with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+ */
+
+#include "iocore/net/NetEvent.h"
+#include "iocore/net/NetHandler.h"
+
+void
+NetEvent::rearm_timer()
+{
+ NetHandler *const h = nh;
+
+ if (h == nullptr) {
+ return;
+ }
+ // The owning thread touches its wheel without holding the NetHandler mutex,
+ // so the mutex does not protect it and a foreign thread must not take it.
+ if (h->thread == this_ethread()) {
+ h->rearm_timer(this);
+ } else {
+ h->defer_rearm(this);
+ }
+}
+
+void
+NetEvent::rebase_default_inactivity_deadline(ink_hrtime previous_in)
+{
+ if (!use_default_inactivity_timeout || inactivity_timeout_in != 0 ||
next_inactivity_timeout_at == 0) {
+ return; // the deadline did not come from the default
+ }
Review Comment:
Installing a positive default on an enabled connection with no explicit
timeout and no existing deadline returns here without scheduling anything. For
example, this can happen when a per-remap override replaces a zero default. If
no more traffic arrives, neither `netActivity()` nor `set_enabled()` repairs
it, so the connection never gets the fallback timeout that the old sweep would
arm. Handle this unarmed state as well as rebasing existing default-derived
deadlines.
##########
src/iocore/net/UnixNetVConnection.cc:
##########
@@ -1292,13 +1339,14 @@ UnixNetVConnection::set_inactivity_timeout(ink_hrtime
timeout_in)
Dbg(dbg_ctl_socket, "Set inactive timeout=%" PRId64 ", for NetVC=%p",
timeout_in, this);
inactivity_timeout_in = timeout_in;
next_inactivity_timeout_at = (timeout_in > 0) ? ink_get_hrtime() +
inactivity_timeout_in : 0;
+ rearm_timer();
Review Comment:
With I/O already enabled, `set_inactivity_timeout(0)` clears the deadline
and cancels the wheel entry even when a positive default timeout exists. An
idle connection gets no further `set_enabled()` or `netActivity()` call, so the
default may never fire. This is reachable through zero-valued transaction
inactivity timeouts in `HttpSM`. Apply the default when clearing the explicit
timeout on an enabled connection, as the old sweep did.
##########
src/iocore/net/UnixNetVConnection.cc:
##########
@@ -1052,7 +1093,8 @@ UnixNetVConnection::mainEvent(int event, Event * /* e
ATS_UNUSED */)
}
*signal_timeout_at = 0;
- writer_cont = write.vio.cont;
+ rearm_timer();
Review Comment:
If either VIO mutex is contended, `mainEvent()` returns `EVENT_CONT` at
lines 1065–1068 before reaching this rearm. The wheel has already removed the
connection, and `InactivityCop::Fire` ignores the callback result, so both
remaining deadlines can stay unscheduled indefinitely. Add `rearm_timer()`
before that lock-failure return so both timeout types are retried, as they were
by the old sweep.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
