2 * Copyright (c) 2016, 2017 Cisco Systems, Inc. and others. All rights reserved.
4 * This program and the accompanying materials are made available under the
5 * terms of the Eclipse Public License v1.0 which accompanies this distribution,
6 * and is available at http://www.eclipse.org/legal/epl-v10.html
9 package org.opendaylight.controller.cluster.access.client;
11 import com.google.common.base.Preconditions;
12 import javax.annotation.concurrent.NotThreadSafe;
13 import org.slf4j.Logger;
14 import org.slf4j.LoggerFactory;
17 * Base class for tracking throughput and computing delays when processing stream of tasks.
19 * <p>The idea is to improve throughput in a typical request-response scenario.
20 * A "frontend" is sending requests towards "backend", backend is sending responses back to fronted.
21 * Both frontend and backend may be realized by multiple Java threads,
22 * so there may be multiple requests not yet responded to.
23 * In terms of taks processing, frontend is "opening" tasks and backend is "closing" them.
24 * Latency of the backend may fluctuate wildly. To avoid backend running out of open tasks,
25 * there should be a queue of requests frontend can add to.
26 * In order to avoid excessive memory consumption, there should be a back-pressure mechanism
27 * which blocks the frontend threads for appropriate durations.
28 * Frontend can tolerate moderately delayed responses, but it only tolerates small block times.
30 * <p>An ideal back-pressure algorithm would keep the queue reasonably full,
31 * while fairly delaying the frontend threads. In other words, backend idle time should be low,
32 * as well as frontend block time dispersion
33 * (as opposed to block time average, which is dictated by overall performance).
35 * <p>In order for an algorithm to compute reasonable wait times,
36 * various inputs can be useful, mostly related to timing of various stages of task processing.
37 * Methods of this class assume "enqueue and wait" usage.
38 * The delay computation is pessimistic, it expects each participating thread to enqueue another task
39 * as soon as its delay time allows.
41 * <p>This class is not thread safe, the callers are responsible for guarding against conflicting access.
42 * Time is measured in ticks (nanos), methods never look at current time, relying on {@code now} argument instead.
43 * This means the sequence of {$code now} argument values is expected to be non-decreasing.
45 * <p>Input data used for tracking is tightly coupled with TransitQueue#recordCompletion arguments.
47 * @author Vratko Polak
49 // TODO: Would bulk methods be less taxing than a loop of single task calls?
51 abstract class ProgressTracker {
52 private static final Logger LOG = LoggerFactory.getLogger(ProgressTracker.class);
55 * When no tasks has been closed yet, this will be used to estimate throughput.
57 private final long defaultTicksPerTask;
60 * Number of tasks closed so far.
62 private long tasksClosed = 0;
65 * Number of tasks so far, both open and closed.
67 private long tasksEncountered = 0;
70 * The most recent tick number when the number of open tasks has become non-positive.
72 private long lastIdle = Long.MIN_VALUE;
75 * The most recent tick number when a task has been closed.
77 private long lastClosed = Long.MIN_VALUE;
80 * Tick number when the farthest known wait time is over.
82 private long nearestAllowed = Long.MIN_VALUE;
85 * Number of ticks elapsed before lastIdle while there was at least one open task.
87 private long elapsedBeforeIdle = 0L;
92 * Construct an idle tracker with specified ticks per task value to use as default.
94 * @param ticksPerTask value to use as default
96 ProgressTracker(final long ticksPerTask) {
97 Preconditions.checkArgument(ticksPerTask >= 0);
98 defaultTicksPerTask = ticksPerTask;
102 * Construct a new tracker suitable for a new task queue related to a "reconnect".
104 * <p>When reconnecting to a new backend, tasks may need to be re-processed by the frontend,
105 * possibly resulting in a different number of tasks.
106 * Also, performance of the new backend can be different, but the perforance of the previous backend
107 * is generally still better estimate than defaults of a brand new tracker.
109 * <p>This "inheritance constructor" creates a new tracker with no open tasks (thus initially idle),
110 * but other internal values should lead to a balanced performance
111 * after tasks opened in the source tracker are "replayed" into the new tracker.
113 * <p>In particular, this impementation keeps the number of closed tasks the same,
114 * and makes it so ticksWorkedPerClosedTask is initially the same as in the old tracker.
116 * @param oldTracker the tracker used for the previously used backend
117 * @param now tick number corresponding to caller's present
119 ProgressTracker(final ProgressTracker oldTracker, final long now) {
120 this.defaultTicksPerTask = oldTracker.defaultTicksPerTask;
121 this.tasksEncountered = this.tasksClosed = oldTracker.tasksClosed;
122 this.lastClosed = oldTracker.lastClosed;
123 this.nearestAllowed = oldTracker.nearestAllowed; // Call cancelDebt explicitly if needed.
124 this.lastIdle = oldTracker.lastIdle;
125 this.elapsedBeforeIdle = oldTracker.elapsedBeforeIdle;
126 if (!oldTracker.isIdle()) {
131 // "Public" shared access (read-only) accessor-like methods
134 * Get number of tasks closed so far.
136 * @return number of tasks known to be finished already; the value never decreases
138 final long tasksClosed() {
143 * Get umber of tasks so far, both open and closed.
145 * @return number of tasks encountered so far, open or finished; the value never decreases
147 final long tasksEncountered() {
148 return tasksEncountered;
152 * Get number of tasks currently open.
154 * @return number of tasks started but not finished yet
156 final long tasksOpen() { // TODO: Should we return int?
157 // TODO: Should we check the return value is non-negative?
158 return tasksEncountered - tasksClosed;
162 * When idle, there are no open tasks so no progress is made.
164 * @return {@code true} if every encountered task is already closed, {@code false} otherwise
166 final boolean isIdle() {
167 return tasksClosed >= tasksEncountered;
171 * Number of ticks elapsed (before now) since the last closed task while there was at least one open task.
173 * @param now tick number corresponding to caller's present
174 * @return number of ticks backend is neither idle nor responding
176 final long ticksStalling(final long now) {
177 return isIdle() ? 0 : Math.max(now, lastClosed) - lastClosed;
180 // Read only protected methods.
183 * Get the value of default ticks per task this instance was created to use.
185 * @return default ticks per task value
187 protected final long defaultTicksPerTask() {
188 return defaultTicksPerTask;
192 * One task is roughly estimated to take this long to close.
194 * @param now tick number corresponding to caller's present
195 * @return total ticks worked divided by closed tasks, or the default value if no closed tasks
197 protected final double ticksWorkedPerClosedTask(final long now) {
198 if (tasksClosed < 1) {
199 return defaultTicksPerTask;
201 return (double) ticksWorked(now) / tasksClosed;
204 // Read only private methods.
207 * Number of ticks elapsed (before now) while there was at least one open task.
209 * @param now tick number corresponding to caller's present
210 * @return number of ticks there was at least one task open
212 private long ticksWorked(final long now) {
213 return isIdle() ? elapsedBeforeIdle : Math.max(now, lastIdle) - lastIdle + elapsedBeforeIdle;
217 * Give an estimate of a tick number when there will be no accumulated delays.
219 * <p>The delays accumulated include one more open task.
220 * Basically, the return value corresponds to openTask() return value,
221 * but this gives an absolute time, instead of delay relative to now.
223 * @param now tick number corresponding to caller's present
224 * @return estimated tick number when all threads with opened tasks are done waiting
226 private long estimateAllowed(final long now) {
227 return Math.max(now, nearestAllowed + estimateIsolatedDelay(now));
230 // State-altering "public" methods.
233 * Track a task is being closed.
235 * @param now tick number corresponding to caller's present
236 * @param enqueuedTicks see TransitQueue#recordCompletion
237 * @param transmitTicks see TransitQueue#recordCompletion
238 * @param execNanos see TransitQueue#recordCompletion
240 final void closeTask(final long now, final long enqueuedTicks, final long transmitTicks, final long execNanos) {
242 LOG.info("Attempted to close a task while no tasks are open");
244 unsafeCloseTask(now, enqueuedTicks, transmitTicks, execNanos);
249 * Track a task that is being opened.
251 * @param now tick number corresponding to caller's present
252 * @return number of ticks (nanos) the caller thread should wait before opening another task
254 final long openTask(final long now) {
255 openTaskWithoutThrottle(now);
256 return reserveDelay(now);
260 * Set nearestAllowed value to now.
262 * <p>This is useful when new a backend has just connected,
263 * after a period of no working backend present.
264 * The accumulated delays should not limit future tasks.
265 * The queue fullness and the averaged backend performance are kept,
266 * as they should result in good enough estimations for new tasks.
268 * @param now tick number corresponding to caller's present
270 final void cancelDebt(final long now) {
271 nearestAllowed = now;
274 // Private state-altering methods.
277 * Compute the next delay and update nearestAllowed value accordingly.
279 * @param now tick number corresponding to caller's present
280 * @return number of ticks (nanos) the caller thread should wait before opening another task
282 private long reserveDelay(final long now) {
283 nearestAllowed = estimateAllowed(now);
284 return nearestAllowed - now;
288 * Track a task is being closed.
290 * <p>This method does not verify there was any task open.
291 * This call can empty the collection of open tasks, that special case should be handled.
293 * @param now tick number corresponding to caller's present
294 * @param enqueuedTicks see TransmitQueue#recordCompletion
295 * @param transmitTicks see TransmitQueue#recordCompletion
296 * @param execNanos see TransmitQueue#recordCompletion
298 private void unsafeCloseTask(final long now, final long enqueuedTicks, final long transmitTicks,
299 final long execNanos) {
308 * Track a task is being opened.
310 * <p>This method does not aggregate delays, allowing the caller to sidestep the throttling.
311 * This call can make the collection of open tasks non-empty, that special case should be handled.
313 * @param now tick number corresponding to caller's present
315 private void openTaskWithoutThrottle(final long now) {
317 transitFromIdle(now);
323 * Update lastIdle as a new "last" just hapened.
325 private void transitFromIdle(final long now) {
326 lastIdle = Math.max(now, lastIdle);
330 * Update elapsedBeforeIdle as the "before" has jast moved.
332 private void transitToIdle(final long now) {
333 elapsedBeforeIdle += Math.max(0, now - lastIdle);
336 // Protected abstract read-only methods.
339 * Give an estimate of a fair delay, assuming delays caused by other opened tasks are ignored.
341 * @param now tick number corresponding to caller's present
342 * @return delay (in ticks) after which another openTask() would be fair to be called by the same thread again
344 protected abstract long estimateIsolatedDelay(long now);