[folly.git] / folly / stats / BucketedTimeSeries.h

/*
 * Copyright 2017 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 <chrono>
#include <vector>

#include <folly/detail/Stats.h>

namespace folly {

/*
 * A helper clock type to helper older code using BucketedTimeSeries with
 * std::chrono::seconds transition to properly using clock types and time_point
 * objects.
 */
template <typename TT = std::chrono::seconds>
class LegacyStatsClock {
 public:
  using duration = TT;
  using time_point = std::chrono::time_point<LegacyStatsClock, TT>;

  // This clock does not actually implement now(), since the older API
  // did not really specify what clock should be used.  (In practice most
  // callers unfortuantely used wall clock time rather than a monotonic clock.)
};

/*
 * This class represents a bucketed time series which keeps track of values
 * added in the recent past, and merges these values together into a fixed
 * number of buckets to keep a lid on memory use if the number of values
 * added is very large.
 *
 * For example, a BucketedTimeSeries() with duration == 60s and 10 buckets
 * will keep track of 10 6-second buckets, and discard all data added more
 * than 1 minute ago.  As time ticks by, a 6-second bucket at a time will
 * be discarded and new data will go into the newly opened bucket.  Internally,
 * it uses a circular array of buckets that it reuses as time advances.
 *
 * This class assumes that time advances forwards.  The window of time tracked
 * by the timeseries will advance forwards whenever a more recent timestamp is
 * passed to addValue().  While it is possible to pass old time values to
 * addValue(), this will never move the time window backwards.  If the old time
 * value falls outside the tracked window of time, the data point will be
 * ignored.
 *
 * This class is not thread-safe -- use your own synchronization!
 */
template <typename VT, typename CT = LegacyStatsClock<std::chrono::seconds>>
class BucketedTimeSeries {
 public:
  using ValueType = VT;
  using Clock = CT;
  using Duration = typename Clock::duration;
  using TimePoint = typename Clock::time_point;
  using Bucket = detail::Bucket<ValueType>;

  /*
   * Create a new BucketedTimeSeries.
   *
   * This creates a new BucketedTimeSeries with the specified number of
   * buckets, storing data for the specified amount of time.
   *
   * If the duration is 0, the BucketedTimeSeries will track data forever,
   * and does not need the rolling buckets.  The numBuckets parameter is
   * ignored when duration is 0.
   */
  BucketedTimeSeries(size_t numBuckets, Duration duration);

  /*
   * Adds the value 'val' at time 'now'
   *
   * This function expects time to generally move forwards.  The window of time
   * tracked by this time series will move forwards with time.  If 'now' is
   * more recent than any time previously seen, addValue() will automatically
   * call update(now) to advance the time window tracked by this data
   * structure.
   *
   * Values in the recent past may be added to the data structure by passing in
   * a slightly older value of 'now', as long as this time point still falls
   * within the tracked duration.  If 'now' is older than the tracked duration
   * of time, the data point value will be ignored, and addValue() will return
   * false without doing anything else.
   *
   * Returns true on success, or false if now was older than the tracked time
   * window.
   */
  bool addValue(TimePoint now, const ValueType& val);

  /*
   * Adds the value 'val' the given number of 'times' at time 'now'
   */
  bool addValue(TimePoint now, const ValueType& val, int64_t times);

  /*
   * Adds the value 'total' as the sum of 'nsamples' samples
   */
  bool
  addValueAggregated(TimePoint now, const ValueType& total, int64_t nsamples);

  /*
   * Updates the container to the specified time, doing all the necessary
   * work to rotate the buckets and remove any stale data points.
   *
   * The addValue() methods automatically call update() when adding new data
   * points.  However, when reading data from the timeseries, you should make
   * sure to manually call update() before accessing the data.  Otherwise you
   * may be reading stale data if update() has not been called recently.
   *
   * Returns the current bucket index after the update.
   */
  size_t update(TimePoint now);

  /*
   * Reset the timeseries to an empty state,
   * as if no data points have ever been added to it.
   */
  void clear();

  /*
   * Get the latest time that has ever been passed to update() or addValue().
   *
   * If no data has ever been added to this timeseries, 0 will be returned.
   */
  TimePoint getLatestTime() const {
    return latestTime_;
  }

  /*
   * Get the time of the earliest data point stored in this timeseries.
   *
   * If no data has ever been added to this timeseries, 0 will be returned.
   *
   * If isAllTime() is true, this is simply the time when the first data point
   * was recorded.
   *
   * For non-all-time data, the timestamp reflects the first data point still
   * remembered.  As new data points are added, old data will be expired.
   * getEarliestTime() returns the timestamp of the oldest bucket still present
   * in the timeseries.  This will never be older than (getLatestTime() -
   * duration()).
   */
  TimePoint getEarliestTime() const;

  /*
   * Return the number of buckets.
   */
  size_t numBuckets() const {
    return buckets_.size();
  }

  /*
   * Return the maximum duration of data that can be tracked by this
   * BucketedTimeSeries.
   */
  Duration duration() const {
    return duration_;
  }

  /*
   * Returns true if this BucketedTimeSeries stores data for all-time, without
   * ever rolling over into new buckets.
   */
  bool isAllTime() const {
    return (duration_ == Duration(0));
  }

  /*
   * Returns true if no calls to update() have been made since the last call to
   * clear().
   */
  bool empty() const {
    // We set firstTime_ greater than latestTime_ in the constructor and in
    // clear, so we use this to distinguish if the timeseries is empty.
    //
    // Once a data point has been added, latestTime_ will always be greater
    // than or equal to firstTime_.
    return firstTime_ > latestTime_;
  }

  /*
   * Get the amount of time tracked by this timeseries.
   *
   * For an all-time timeseries, this returns the length of time since the
   * first data point was added to the time series.
   *
   * Otherwise, this never returns a value greater than the overall timeseries
   * duration.  If the first data point was recorded less than a full duration
   * ago, the time since the first data point is returned.  If a full duration
   * has elapsed, and we have already thrown away some data, the time since the
   * oldest bucket is returned.
   *
   * For example, say we are tracking 600 seconds worth of data, in 60 buckets.
   * - If less than 600 seconds have elapsed since the first data point,
   *   elapsed() returns the total elapsed time so far.
   * - If more than 600 seconds have elapsed, we have already thrown away some
   *   data.  However, we throw away a full bucket (10 seconds worth) at once,
   *   so at any point in time we have from 590 to 600 seconds worth of data.
   *   elapsed() will therefore return a value between 590 and 600.
   *
   * Note that you generally should call update() before calling elapsed(), to
   * make sure you are not reading stale data.
   */
  Duration elapsed() const;

  /*
   * Get the amount of time tracked by this timeseries, between the specified
   * start and end times.
   *
   * If the timeseries contains data for the entire time range specified, this
   * simply returns (end - start).  However, if start is earlier than
   * getEarliestTime(), this returns (end - getEarliestTime()).
   */
  Duration elapsed(TimePoint start, TimePoint end) const;

  /*
   * Return the sum of all the data points currently tracked by this
   * BucketedTimeSeries.
   *
   * Note that you generally should call update() before calling sum(), to
   * make sure you are not reading stale data.
   */
  const ValueType& sum() const {
    return total_.sum;
  }

  /*
   * Return the number of data points currently tracked by this
   * BucketedTimeSeries.
   *
   * Note that you generally should call update() before calling count(), to
   * make sure you are not reading stale data.
   */
  uint64_t count() const {
    return total_.count;
  }

  /*
   * Return the average value (sum / count).
   *
   * The return type may be specified to control whether floating-point or
   * integer division should be performed.
   *
   * Note that you generally should call update() before calling avg(), to
   * make sure you are not reading stale data.
   */
  template <typename ReturnType=double>
  ReturnType avg() const {
    return total_.template avg<ReturnType>();
  }

  /*
   * Return the sum divided by the elapsed time.
   *
   * Note that you generally should call update() before calling rate(), to
   * make sure you are not reading stale data.
   */
  template <typename ReturnType = double, typename Interval = Duration>
  ReturnType rate() const {
    return rateHelper<ReturnType, Interval>(total_.sum, elapsed());
  }

  /*
   * Return the count divided by the elapsed time.
   *
   * The Interval template parameter causes the elapsed time to be converted to
   * the Interval type before using it.  For example, if Interval is
   * std::chrono::seconds, the return value will be the count per second.
   * If Interval is std::chrono::hours, the return value will be the count per
   * hour.
   *
   * Note that you generally should call update() before calling countRate(),
   * to make sure you are not reading stale data.
   */
  template <typename ReturnType = double, typename Interval = Duration>
  ReturnType countRate() const {
    return rateHelper<ReturnType, Interval>(total_.count, elapsed());
  }

  /*
   * Estimate the sum of the data points that occurred in the specified time
   * period.
   *
   * The range queried is [start, end).
   * That is, start is inclusive, and end is exclusive.
   *
   * Note that data outside of the timeseries duration will no longer be
   * available for use in the estimation.  Specifying a start time earlier than
   * getEarliestTime() will not have much effect, since only data points after
   * that point in time will be counted.
   *
   * Note that the value returned is an estimate, and may not be precise.
   */
  ValueType sum(TimePoint start, TimePoint end) const;

  /*
   * Estimate the number of data points that occurred in the specified time
   * period.
   *
   * The same caveats documented in the sum(TimePoint start, TimePoint end)
   * comments apply here as well.
   */
  uint64_t count(TimePoint start, TimePoint end) const;

  /*
   * Estimate the average value during the specified time period.
   *
   * The same caveats documented in the sum(TimePoint start, TimePoint end)
   * comments apply here as well.
   */
  template <typename ReturnType = double>
  ReturnType avg(TimePoint start, TimePoint end) const;

  /*
   * Estimate the rate during the specified time period.
   *
   * The same caveats documented in the sum(TimePoint start, TimePoint end)
   * comments apply here as well.
   */
  template <typename ReturnType = double, typename Interval = Duration>
  ReturnType rate(TimePoint start, TimePoint end) const {
    ValueType intervalSum = sum(start, end);
    Duration interval = elapsed(start, end);
    return rateHelper<ReturnType, Interval>(intervalSum, interval);
  }

  /*
   * Estimate the rate of data points being added during the specified time
   * period.
   *
   * The same caveats documented in the sum(TimePoint start, TimePoint end)
   * comments apply here as well.
   */
  template <typename ReturnType = double, typename Interval = Duration>
  ReturnType countRate(TimePoint start, TimePoint end) const {
    uint64_t intervalCount = count(start, end);
    Duration interval = elapsed(start, end);
    return rateHelper<ReturnType, Interval>(intervalCount, interval);
  }

  /*
   * Invoke a function for each bucket.
   *
   * The function will take as arguments the bucket index,
   * the bucket start time, and the start time of the subsequent bucket.
   *
   * It should return true to continue iterating through the buckets, and false
   * to break out of the loop and stop, without calling the function on any
   * more buckets.
   *
   * bool function(const Bucket& bucket, TimePoint bucketStart,
   *               TimePoint nextBucketStart)
   */
  template <typename Function>
  void forEachBucket(Function fn) const;

  /*
   * Get the index for the bucket containing the specified time.
   *
   * Note that the index is only valid if this time actually falls within one
   * of the current buckets.  If you pass in a value more recent than
   * getLatestTime() or older than (getLatestTime() - elapsed()), the index
   * returned will not be valid.
   *
   * This method may not be called for all-time data.
   */
  size_t getBucketIdx(TimePoint time) const;

  /*
   * Get the bucket at the specified index.
   *
   * This method may not be called for all-time data.
   */
  const Bucket& getBucketByIndex(size_t idx) const {
    return buckets_[idx];
  }

  /*
   * Compute the bucket index that the specified time falls into,
   * as well as the bucket start time and the next bucket's start time.
   *
   * This method may not be called for all-time data.
   */
  void getBucketInfo(
      TimePoint time,
      size_t* bucketIdx,
      TimePoint* bucketStart,
      TimePoint* nextBucketStart) const;

  /*
   * Legacy APIs that accept a Duration parameters rather than TimePoint.
   *
   * These treat the Duration as relative to the clock epoch.
   * Prefer using the correct TimePoint-based APIs instead.  These APIs will
   * eventually be deprecated and removed.
   */
  bool addValue(Duration now, const ValueType& val) {
    return addValueAggregated(TimePoint(now), val, 1);
  }
  bool addValue(Duration now, const ValueType& val, int64_t times) {
    return addValueAggregated(TimePoint(now), val * times, times);
  }
  bool
  addValueAggregated(Duration now, const ValueType& total, int64_t nsamples) {
    return addValueAggregated(TimePoint(now), total, nsamples);
  }
  size_t update(Duration now) {
    return update(TimePoint(now));
  }

 private:
  template <typename ReturnType = double, typename Interval = Duration>
  ReturnType rateHelper(ReturnType numerator, Duration elapsedTime) const {
    return detail::rateHelper<ReturnType, Duration, Interval>(
        numerator, elapsedTime);
  }

  TimePoint getEarliestTimeNonEmpty() const;
  size_t updateBuckets(TimePoint now);

  ValueType rangeAdjust(
      TimePoint bucketStart,
      TimePoint nextBucketStart,
      TimePoint start,
      TimePoint end,
      ValueType input) const;

  template <typename Function>
  void forEachBucket(TimePoint start, TimePoint end, Function fn) const;

  TimePoint firstTime_; // time of first update() since clear()/constructor
  TimePoint latestTime_; // time of last update()
  Duration duration_; // total duration ("window length") of the time series

  Bucket total_;                 // sum and count of everything in time series
  std::vector<Bucket> buckets_;  // actual buckets of values
};

} // folly