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 org.slf4j.Logger;
13 import org.slf4j.LoggerFactory;
16 * Base class for tracking throughput and computing delays when processing stream of tasks.
18 * <p>The idea is to improve throughput in a typical request-response scenario.
19 * Multiple "user" threads are submitting requests to a "frontend". The frontend does some
20 * pre-processing and then sends requests to (usually one) "backend". The backend does the main work
21 * and replies to the frontend, which reports to the the corresponding user.
22 * In terms of task processing, user threads are "opening" tasks and frontend is "closing" them.
23 * Latency of the backend may fluctuate wildly. To avoid backend running out of open tasks,
24 * frontend should maintain a queue of requests for users to submit tasks to.
25 * In order to avoid excessive memory consumption, there should be a back-pressure mechanism
26 * which blocks the user (submit) threads for appropriate durations.
27 * Users can tolerate moderately delayed responses, but they only tolerate small block (submit)
30 * <p>An ideal back-pressure algorithm would keep the queue reasonably full,
31 * while fairly delaying the user threads. In other words, backend idle time should be low,
32 * as well as user 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, submit thread is supposed to block itself
38 * when asked to. The delay computation is pessimistic, it expects each participating thread
39 * to enqueue another task as soon as its delay time allows.
41 * <p>This class is to be used by single frontend. This class is not thread safe,
42 * the frontend is responsible for guarding against conflicting access.
43 * Time is measured in ticks (nanoseconds), methods never look at current time,
44 * relying on {@code now} argument where appropriate.
45 * This means the sequence of {@code now} argument values is expected to be non-decreasing.
47 * <p>Input data used for tracking is tightly coupled with TransitQueue#recordCompletion arguments.
49 * @author Vratko Polak
51 // TODO: Would bulk methods be less taxing than a loop of single task calls?
52 abstract class ProgressTracker {
53 private static final Logger LOG = LoggerFactory.getLogger(ProgressTracker.class);
56 * When no tasks has been closed yet, this will be used to estimate throughput.
58 private final long defaultTicksPerTask;
61 * Number of tasks closed so far.
63 private long tasksClosed = 0;
66 * Number of tasks so far, both open and closed.
68 private long tasksEncountered = 0;
71 * The most recent tick number when the number of open tasks has become non-positive.
73 private long lastIdle = Long.MIN_VALUE;
76 * The most recent tick number when a task has been closed.
78 private long lastClosed = Long.MIN_VALUE;
81 * Tick number when the farthest known wait time is over.
83 private long nearestAllowed = Long.MIN_VALUE;
86 * Number of ticks elapsed before lastIdle while there was at least one open task.
88 private long elapsedBeforeIdle = 0L;
93 * Construct an idle tracker with specified ticks per task value to use as default.
95 * @param ticksPerTask value to use as default
97 ProgressTracker(final long ticksPerTask) {
98 Preconditions.checkArgument(ticksPerTask >= 0);
99 defaultTicksPerTask = ticksPerTask;
103 * Construct a new tracker suitable for a new task queue related to a "reconnect".
105 * <p>When reconnecting to a new backend, tasks may need to be re-processed by the frontend,
106 * possibly resulting in a different number of tasks.
107 * Also, performance of the new backend can be different, but the perforance of the previous backend
108 * is generally still better estimate than defaults of a brand new tracker.
110 * <p>This "inheritance constructor" creates a new tracker with no open tasks (thus initially idle),
111 * but other internal values should lead to a balanced performance
112 * after tasks opened in the source tracker are "replayed" into the new tracker.
114 * <p>In particular, this impementation keeps the number of closed tasks the same,
115 * and makes it so ticksWorkedPerClosedTask is initially the same as in the old tracker.
117 * @param oldTracker the tracker used for the previously used backend
118 * @param now tick number corresponding to caller's present
120 ProgressTracker(final ProgressTracker oldTracker, final long now) {
121 this.defaultTicksPerTask = oldTracker.defaultTicksPerTask;
122 this.tasksEncountered = this.tasksClosed = oldTracker.tasksClosed;
123 this.lastClosed = oldTracker.lastClosed;
124 this.nearestAllowed = oldTracker.nearestAllowed; // Call cancelDebt explicitly if needed.
125 this.lastIdle = oldTracker.lastIdle;
126 this.elapsedBeforeIdle = oldTracker.elapsedBeforeIdle;
127 if (!oldTracker.isIdle()) {
132 // "Public" shared access (read-only) accessor-like methods
135 * Get number of tasks closed so far.
137 * @return number of tasks known to be finished already; the value never decreases
139 final long tasksClosed() {
144 * Get umber of tasks so far, both open and closed.
146 * @return number of tasks encountered so far, open or finished; the value never decreases
148 final long tasksEncountered() {
149 return tasksEncountered;
153 * Get number of tasks currently open.
155 * @return number of tasks started but not finished yet
157 final long tasksOpen() { // TODO: Should we return int?
158 // TODO: Should we check the return value is non-negative?
159 return tasksEncountered - tasksClosed;
163 * When idle, there are no open tasks so no progress is made.
165 * @return {@code true} if every encountered task is already closed, {@code false} otherwise
167 final boolean isIdle() {
168 return tasksClosed >= tasksEncountered;
172 * Number of ticks elapsed (before now) since the last closed task while there was at least one open task.
174 * @param now tick number corresponding to caller's present
175 * @return number of ticks backend is neither idle nor responding
177 final long ticksStalling(final long now) {
178 return isIdle() ? 0 : Math.max(now, lastClosed) - lastClosed;
181 // Read only protected methods.
184 * Get the value of default ticks per task this instance was created to use.
186 * @return default ticks per task value
188 protected final long defaultTicksPerTask() {
189 return defaultTicksPerTask;
193 * One task is roughly estimated to take this long to close.
195 * @param now tick number corresponding to caller's present
196 * @return total ticks worked divided by closed tasks, or the default value if no closed tasks
198 protected final double ticksWorkedPerClosedTask(final long now) {
199 if (tasksClosed < 1) {
200 return defaultTicksPerTask;
202 return (double) ticksWorked(now) / tasksClosed;
205 // Read only private methods.
208 * Number of ticks elapsed (before now) while there was at least one open task.
210 * @param now tick number corresponding to caller's present
211 * @return number of ticks there was at least one task open
213 private long ticksWorked(final long now) {
214 return isIdle() ? elapsedBeforeIdle : Math.max(now, lastIdle) - lastIdle + elapsedBeforeIdle;
218 * Give an estimate of a tick number when there will be no accumulated delays.
220 * <p>The delays accumulated include one more open task.
221 * Basically, the return value corresponds to openTask() return value,
222 * but this gives an absolute time, instead of delay relative to now.
224 * @param now tick number corresponding to caller's present
225 * @return estimated tick number when all threads with opened tasks are done waiting
227 private long estimateAllowed(final long now) {
228 return Math.max(now, nearestAllowed + estimateIsolatedDelay(now));
231 // State-altering "public" methods.
234 * Track a task is being closed.
236 * @param now tick number corresponding to caller's present
237 * @param enqueuedTicks see TransitQueue#recordCompletion
238 * @param transmitTicks see TransitQueue#recordCompletion
239 * @param execNanos see TransitQueue#recordCompletion
241 final void closeTask(final long now, final long enqueuedTicks, final long transmitTicks, final long execNanos) {
243 LOG.info("Attempted to close a task while no tasks are open");
245 unsafeCloseTask(now, enqueuedTicks, transmitTicks, execNanos);
250 * Track a task that is being opened.
252 * @param now tick number corresponding to caller's present
253 * @return number of ticks (nanos) the caller thread should wait before opening another task
255 final long openTask(final long now) {
256 openTaskWithoutThrottle(now);
257 return reserveDelay(now);
261 * Set nearestAllowed value to now.
263 * <p>This is useful when new a backend has just connected,
264 * after a period of no working backend present.
265 * The accumulated delays should not limit future tasks.
266 * The queue fullness and the averaged backend performance are kept,
267 * as they should result in good enough estimations for new tasks.
269 * @param now tick number corresponding to caller's present
271 final void cancelDebt(final long now) {
272 nearestAllowed = now;
275 // Private state-altering methods.
278 * Compute the next delay and update nearestAllowed value accordingly.
280 * @param now tick number corresponding to caller's present
281 * @return number of ticks (nanos) the caller thread should wait before opening another task
283 private long reserveDelay(final long now) {
284 nearestAllowed = estimateAllowed(now);
285 return nearestAllowed - now;
289 * Track a task is being closed.
291 * <p>This method does not verify there was any task open.
292 * This call can empty the collection of open tasks, that special case should be handled.
294 * @param now tick number corresponding to caller's present
295 * @param enqueuedTicks see TransmitQueue#recordCompletion
296 * @param transmitTicks see TransmitQueue#recordCompletion
297 * @param execNanos see TransmitQueue#recordCompletion
299 private void unsafeCloseTask(final long now, final long enqueuedTicks, final long transmitTicks,
300 final long execNanos) {
309 * Track a task is being opened.
311 * <p>This method does not aggregate delays, allowing the caller to sidestep the throttling.
312 * This call can make the collection of open tasks non-empty, that special case should be handled.
314 * @param now tick number corresponding to caller's present
316 private void openTaskWithoutThrottle(final long now) {
318 transitFromIdle(now);
324 * Update lastIdle as a new "last" just hapened.
326 private void transitFromIdle(final long now) {
327 lastIdle = Math.max(now, lastIdle);
331 * Update elapsedBeforeIdle as the "before" has jast moved.
333 private void transitToIdle(final long now) {
334 elapsedBeforeIdle += Math.max(0, now - lastIdle);
337 // Protected abstract read-only methods.
340 * Give an estimate of a fair delay, assuming delays caused by other opened tasks are ignored.
342 * @param now tick number corresponding to caller's present
343 * @return delay (in ticks) after which another openTask() would be fair to be called by the same thread again
345 protected abstract long estimateIsolatedDelay(long now);