Commit 260f12ae authored by James Sedgwick's avatar James Sedgwick Committed by Anton Likhtarov

move Codel to wangle

Summary: under folly::wangle::

Test Plan: ran unit for experimental/wangle, thrift/lib/cpp/concurrency, and a user of FbThreadManager

Reviewed By: hans@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, fugalh, alandau, bmatheny, everstore-dev@, njormrod

FB internal diff: D1566128
parent 69a28bbf
/*
* Copyright 2014 Facebook, Inc.
*
* Licensed 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 <folly/experimental/wangle/concurrent/Codel.h>
#include <algorithm>
#include <math.h>
#ifndef NO_LIB_GFLAGS
#include <gflags/gflags.h>
DEFINE_int32(codel_interval, 100,
"Codel default interval time in ms");
DEFINE_int32(codel_target_delay, 5,
"Target codel queueing delay in ms");
#endif
namespace folly { namespace wangle {
#ifdef NO_LIB_GFLAGS
int32_t FLAGS_codel_interval = 100;
int32_t FLAGS_codel_target_delay = 5;
#endif
Codel::Codel()
: codelMinDelay_(0),
codelIntervalTime_(std::chrono::steady_clock::now()),
codelResetDelay_(true),
overloaded_(false) {}
bool Codel::overloaded(std::chrono::microseconds delay) {
bool ret = false;
auto now = std::chrono::steady_clock::now();
// Avoid another thread updating the value at the same time we are using it
// to calculate the overloaded state
auto minDelay = codelMinDelay_;
if (now > codelIntervalTime_ &&
(!codelResetDelay_.load(std::memory_order_acquire)
&& !codelResetDelay_.exchange(true))) {
codelIntervalTime_ = now + std::chrono::milliseconds(FLAGS_codel_interval);
if (minDelay > std::chrono::milliseconds(FLAGS_codel_target_delay)) {
overloaded_ = true;
} else {
overloaded_ = false;
}
}
// Care must be taken that only a single thread resets codelMinDelay_,
// and that it happens after the interval reset above
if (codelResetDelay_.load(std::memory_order_acquire) &&
codelResetDelay_.exchange(false)) {
codelMinDelay_ = delay;
// More than one request must come in during an interval before codel
// starts dropping requests
return false;
} else if(delay < codelMinDelay_) {
codelMinDelay_ = delay;
}
if (overloaded_ &&
delay > std::chrono::milliseconds(FLAGS_codel_target_delay * 2)) {
ret = true;
}
return ret;
}
int Codel::getLoad() {
return std::min(100, (int)codelMinDelay_.count() / FLAGS_codel_interval);
}
}} //namespace
/*
* Copyright 2014 Facebook, Inc.
*
* Licensed 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.
*/
#pragma once
#include <atomic>
#include <chrono>
namespace folly { namespace wangle {
/* Codel algorithm implementation:
* http://en.wikipedia.org/wiki/CoDel
*
* Algorithm modified slightly: Instead of changing the interval time
* based on the average min delay, instead we use an alternate timeout
* for each task if the min delay during the interval period is too
* high.
*
* This was found to have better latency metrics than changing the
* window size, since we can communicate with the sender via thrift
* instead of only via the tcp window size congestion control, as in TCP.
*/
class Codel {
public:
Codel();
// Given a delay, returns wether the codel algorithm would
// reject a queued request with this delay.
//
// Internally, it also keeps track of the interval
bool overloaded(std::chrono::microseconds delay);
// Get the queue load, as seen by the codel algorithm
// Gives a rough guess at how bad the queue delay is.
//
// Return: 0 = no delay, 100 = At the queueing limit
int getLoad();
private:
std::chrono::microseconds codelMinDelay_;
std::chrono::time_point<std::chrono::steady_clock> codelIntervalTime_;
// flag to make overloaded() thread-safe, since we only want
// to reset the delay once per time period
std::atomic<bool> codelResetDelay_;
bool overloaded_;
};
}} // Namespace
/*
* Copyright 2014 Facebook, Inc.
*
* Licensed 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 <chrono>
#include <folly/experimental/wangle/concurrent/Codel.h>
#include <gtest/gtest.h>
#include <thread>
TEST(CodelTest, Basic) {
using std::chrono::milliseconds;
folly::wangle::Codel c;
std::this_thread::sleep_for(milliseconds(110));
// This interval is overloaded
EXPECT_FALSE(c.overloaded(milliseconds(100)));
std::this_thread::sleep_for(milliseconds(90));
// At least two requests must happen in an interval before they will fail
EXPECT_FALSE(c.overloaded(milliseconds(50)));
EXPECT_TRUE(c.overloaded(milliseconds(50)));
std::this_thread::sleep_for(milliseconds(110));
// Previous interval is overloaded, but 2ms isn't enough to fail
EXPECT_FALSE(c.overloaded(milliseconds(2)));
std::this_thread::sleep_for(milliseconds(90));
// 20 ms > target interval * 2
EXPECT_TRUE(c.overloaded(milliseconds(20)));
}
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment