1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 // Histogram is an object that aggregates statistics, and can summarize them in
6 // various forms, including ASCII graphical, HTML, and numerically (as a
7 // vector of numbers corresponding to each of the aggregating buckets).
9 // It supports calls to accumulate either time intervals (which are processed
10 // as integral number of milliseconds), or arbitrary integral units.
12 // For Histogram(exponential histogram), LinearHistogram and CustomHistogram,
13 // the minimum for a declared range is 1 (instead of 0), while the maximum is
14 // (HistogramBase::kSampleType_MAX - 1). Currently you can declare histograms
15 // with ranges exceeding those limits (e.g. 0 as minimal or
16 // HistogramBase::kSampleType_MAX as maximal), but those excesses will be
17 // silently clamped to those limits (for backwards compatibility with existing
18 // code). Best practice is to not exceed the limits.
20 // Each use of a histogram with the same name will reference the same underlying
21 // data, so it is safe to record to the same histogram from multiple locations
22 // in the code. It is a runtime error if all uses of the same histogram do not
23 // agree exactly in type, bucket size and range.
25 // For Histogram and LinearHistogram, the maximum for a declared range should
26 // always be larger (not equal) than minimal range. Zero and
27 // HistogramBase::kSampleType_MAX are implicitly added as first and last ranges,
28 // so the smallest legal bucket_count is 3. However CustomHistogram can have
29 // bucket count as 2 (when you give a custom ranges vector containing only 1
31 // For these 3 kinds of histograms, the max bucket count is always
32 // (Histogram::kBucketCount_MAX - 1).
34 // The buckets layout of class Histogram is exponential. For example, buckets
35 // might contain (sequentially) the count of values in the following intervals:
36 // [0,1), [1,2), [2,4), [4,8), [8,16), [16,32), [32,64), [64,infinity)
37 // That bucket allocation would actually result from construction of a histogram
38 // for values between 1 and 64, with 8 buckets, such as:
39 // Histogram count("some name", 1, 64, 8);
40 // Note that the underflow bucket [0,1) and the overflow bucket [64,infinity)
41 // are also counted by the constructor in the user supplied "bucket_count"
43 // The above example has an exponential ratio of 2 (doubling the bucket width
44 // in each consecutive bucket. The Histogram class automatically calculates
45 // the smallest ratio that it can use to construct the number of buckets
46 // selected in the constructor. An another example, if you had 50 buckets,
47 // and millisecond time values from 1 to 10000, then the ratio between
48 // consecutive bucket widths will be approximately somewhere around the 50th
49 // root of 10000. This approach provides very fine grain (narrow) buckets
50 // at the low end of the histogram scale, but allows the histogram to cover a
51 // gigantic range with the addition of very few buckets.
53 // Usually we use macros to define and use a histogram, which are defined in
54 // base/metrics/histogram_macros.h. Note: Callers should include that header
55 // directly if they only access the histogram APIs through macros.
57 // Macros use a pattern involving a function static variable, that is a pointer
58 // to a histogram. This static is explicitly initialized on any thread
59 // that detects a uninitialized (NULL) pointer. The potentially racy
60 // initialization is not a problem as it is always set to point to the same
61 // value (i.e., the FactoryGet always returns the same value). FactoryGet
62 // is also completely thread safe, which results in a completely thread safe,
63 // and relatively fast, set of counters. To avoid races at shutdown, the static
64 // pointer is NOT deleted, and we leak the histograms at process termination.
66 #ifndef BASE_METRICS_HISTOGRAM_H_
67 #define BASE_METRICS_HISTOGRAM_H_
73 #include "base/base_export.h"
74 #include "base/basictypes.h"
75 #include "base/compiler_specific.h"
76 #include "base/gtest_prod_util.h"
77 #include "base/logging.h"
78 #include "base/memory/scoped_ptr.h"
79 #include "base/metrics/bucket_ranges.h"
80 #include "base/metrics/histogram_base.h"
81 // TODO(asvitkine): Migrate callers to to include this directly and remove this.
82 #include "base/metrics/histogram_macros.h"
83 #include "base/metrics/histogram_samples.h"
84 #include "base/time/time.h"
91 class BooleanHistogram
;
92 class CustomHistogram
;
94 class LinearHistogram
;
97 class BASE_EXPORT Histogram
: public HistogramBase
{
99 // Initialize maximum number of buckets in histograms as 16,384.
100 static const size_t kBucketCount_MAX
;
102 typedef std::vector
<Count
> Counts
;
104 //----------------------------------------------------------------------------
105 // For a valid histogram, input should follow these restrictions:
106 // minimum > 0 (if a minimum below 1 is specified, it will implicitly be
107 // normalized up to 1)
109 // buckets > 2 [minimum buckets needed: underflow, overflow and the range]
111 // buckets <= (maximum - minimum + 2) - this is to ensure that we don't have
112 // more buckets than the range of numbers; having more buckets than 1 per
113 // value in the range would be nonsensical.
114 static HistogramBase
* FactoryGet(const std::string
& name
,
119 static HistogramBase
* FactoryTimeGet(const std::string
& name
,
120 base::TimeDelta minimum
,
121 base::TimeDelta maximum
,
125 static void InitializeBucketRanges(Sample minimum
,
127 BucketRanges
* ranges
);
129 // This constant if for FindCorruption. Since snapshots of histograms are
130 // taken asynchronously relative to sampling, and our counting code currently
131 // does not prevent race conditions, it is pretty likely that we'll catch a
132 // redundant count that doesn't match the sample count. We allow for a
133 // certain amount of slop before flagging this as an inconsistency. Even with
134 // an inconsistency, we'll snapshot it again (for UMA in about a half hour),
135 // so we'll eventually get the data, if it was not the result of a corruption.
136 static const int kCommonRaceBasedCountMismatch
;
138 // Check to see if bucket ranges, counts and tallies in the snapshot are
139 // consistent with the bucket ranges and checksums in our histogram. This can
140 // produce a false-alarm if a race occurred in the reading of the data during
141 // a SnapShot process, but should otherwise be false at all times (unless we
142 // have memory over-writes, or DRAM failures).
143 int FindCorruption(const HistogramSamples
& samples
) const override
;
145 //----------------------------------------------------------------------------
146 // Accessors for factory construction, serialization and testing.
147 //----------------------------------------------------------------------------
148 Sample
declared_min() const { return declared_min_
; }
149 Sample
declared_max() const { return declared_max_
; }
150 virtual Sample
ranges(size_t i
) const;
151 virtual size_t bucket_count() const;
152 const BucketRanges
* bucket_ranges() const { return bucket_ranges_
; }
154 // This function validates histogram construction arguments. It returns false
155 // if some of the arguments are totally bad.
156 // Note. Currently it allow some bad input, e.g. 0 as minimum, but silently
157 // converts it to good input: 1.
158 // TODO(kaiwang): Be more restrict and return false for any bad input, and
159 // make this a readonly validating function.
160 static bool InspectConstructionArguments(const std::string
& name
,
163 size_t* bucket_count
);
165 // HistogramBase implementation:
166 HistogramType
GetHistogramType() const override
;
167 bool HasConstructionArguments(Sample expected_minimum
,
168 Sample expected_maximum
,
169 size_t expected_bucket_count
) const override
;
170 void Add(Sample value
) override
;
171 scoped_ptr
<HistogramSamples
> SnapshotSamples() const override
;
172 void AddSamples(const HistogramSamples
& samples
) override
;
173 bool AddSamplesFromPickle(PickleIterator
* iter
) override
;
174 void WriteHTMLGraph(std::string
* output
) const override
;
175 void WriteAscii(std::string
* output
) const override
;
178 // |ranges| should contain the underflow and overflow buckets. See top
179 // comments for example.
180 Histogram(const std::string
& name
,
183 const BucketRanges
* ranges
);
185 ~Histogram() override
;
187 // HistogramBase implementation:
188 bool SerializeInfoImpl(Pickle
* pickle
) const override
;
190 // Method to override to skip the display of the i'th bucket if it's empty.
191 virtual bool PrintEmptyBucket(size_t index
) const;
193 // Get normalized size, relative to the ranges(i).
194 virtual double GetBucketSize(Count current
, size_t i
) const;
196 // Return a string description of what goes in a given bucket.
197 // Most commonly this is the numeric value, but in derived classes it may
198 // be a name (or string description) given to the bucket.
199 virtual const std::string
GetAsciiBucketRange(size_t it
) const;
202 // Allow tests to corrupt our innards for testing purposes.
203 FRIEND_TEST_ALL_PREFIXES(HistogramTest
, BoundsTest
);
204 FRIEND_TEST_ALL_PREFIXES(HistogramTest
, BucketPlacementTest
);
205 FRIEND_TEST_ALL_PREFIXES(HistogramTest
, CorruptBucketBounds
);
206 FRIEND_TEST_ALL_PREFIXES(HistogramTest
, CorruptSampleCounts
);
207 FRIEND_TEST_ALL_PREFIXES(HistogramTest
, NameMatchTest
);
209 friend class StatisticsRecorder
; // To allow it to delete duplicates.
210 friend class StatisticsRecorderTest
;
212 friend BASE_EXPORT_PRIVATE HistogramBase
* DeserializeHistogramInfo(
213 PickleIterator
* iter
);
214 static HistogramBase
* DeserializeInfoImpl(PickleIterator
* iter
);
216 // Implementation of SnapshotSamples function.
217 scoped_ptr
<SampleVector
> SnapshotSampleVector() const;
219 //----------------------------------------------------------------------------
220 // Helpers for emitting Ascii graphic. Each method appends data to output.
222 void WriteAsciiImpl(bool graph_it
,
223 const std::string
& newline
,
224 std::string
* output
) const;
226 // Find out how large (graphically) the largest bucket will appear to be.
227 double GetPeakBucketSize(const SampleVector
& samples
) const;
229 // Write a common header message describing this histogram.
230 void WriteAsciiHeader(const SampleVector
& samples
,
232 std::string
* output
) const;
234 // Write information about previous, current, and next buckets.
235 // Information such as cumulative percentage, etc.
236 void WriteAsciiBucketContext(const int64 past
, const Count current
,
237 const int64 remaining
, const size_t i
,
238 std::string
* output
) const;
240 // WriteJSON calls these.
241 void GetParameters(DictionaryValue
* params
) const override
;
243 void GetCountAndBucketData(Count
* count
,
245 ListValue
* buckets
) const override
;
247 // Does not own this object. Should get from StatisticsRecorder.
248 const BucketRanges
* bucket_ranges_
;
250 Sample declared_min_
; // Less than this goes into the first bucket.
251 Sample declared_max_
; // Over this goes into the last bucket.
253 // Finally, provide the state that changes with the addition of each new
255 scoped_ptr
<SampleVector
> samples_
;
257 DISALLOW_COPY_AND_ASSIGN(Histogram
);
260 //------------------------------------------------------------------------------
262 // LinearHistogram is a more traditional histogram, with evenly spaced
264 class BASE_EXPORT LinearHistogram
: public Histogram
{
266 ~LinearHistogram() override
;
268 /* minimum should start from 1. 0 is as minimum is invalid. 0 is an implicit
269 default underflow bucket. */
270 static HistogramBase
* FactoryGet(const std::string
& name
,
275 static HistogramBase
* FactoryTimeGet(const std::string
& name
,
281 struct DescriptionPair
{
283 const char* description
; // Null means end of a list of pairs.
286 // Create a LinearHistogram and store a list of number/text values for use in
287 // writing the histogram graph.
288 // |descriptions| can be NULL, which means no special descriptions to set. If
289 // it's not NULL, the last element in the array must has a NULL in its
290 // "description" field.
291 static HistogramBase
* FactoryGetWithRangeDescription(
292 const std::string
& name
,
297 const DescriptionPair descriptions
[]);
299 static void InitializeBucketRanges(Sample minimum
,
301 BucketRanges
* ranges
);
303 // Overridden from Histogram:
304 HistogramType
GetHistogramType() const override
;
307 LinearHistogram(const std::string
& name
,
310 const BucketRanges
* ranges
);
312 double GetBucketSize(Count current
, size_t i
) const override
;
314 // If we have a description for a bucket, then return that. Otherwise
315 // let parent class provide a (numeric) description.
316 const std::string
GetAsciiBucketRange(size_t i
) const override
;
318 // Skip printing of name for numeric range if we have a name (and if this is
320 bool PrintEmptyBucket(size_t index
) const override
;
323 friend BASE_EXPORT_PRIVATE HistogramBase
* DeserializeHistogramInfo(
324 PickleIterator
* iter
);
325 static HistogramBase
* DeserializeInfoImpl(PickleIterator
* iter
);
327 // For some ranges, we store a printable description of a bucket range.
328 // If there is no description, then GetAsciiBucketRange() uses parent class
329 // to provide a description.
330 typedef std::map
<Sample
, std::string
> BucketDescriptionMap
;
331 BucketDescriptionMap bucket_description_
;
333 DISALLOW_COPY_AND_ASSIGN(LinearHistogram
);
336 //------------------------------------------------------------------------------
338 // BooleanHistogram is a histogram for booleans.
339 class BASE_EXPORT BooleanHistogram
: public LinearHistogram
{
341 static HistogramBase
* FactoryGet(const std::string
& name
, int32 flags
);
343 HistogramType
GetHistogramType() const override
;
346 BooleanHistogram(const std::string
& name
, const BucketRanges
* ranges
);
348 friend BASE_EXPORT_PRIVATE HistogramBase
* DeserializeHistogramInfo(
349 PickleIterator
* iter
);
350 static HistogramBase
* DeserializeInfoImpl(PickleIterator
* iter
);
352 DISALLOW_COPY_AND_ASSIGN(BooleanHistogram
);
355 //------------------------------------------------------------------------------
357 // CustomHistogram is a histogram for a set of custom integers.
358 class BASE_EXPORT CustomHistogram
: public Histogram
{
360 // |custom_ranges| contains a vector of limits on ranges. Each limit should be
361 // > 0 and < kSampleType_MAX. (Currently 0 is still accepted for backward
362 // compatibility). The limits can be unordered or contain duplication, but
363 // client should not depend on this.
364 static HistogramBase
* FactoryGet(const std::string
& name
,
365 const std::vector
<Sample
>& custom_ranges
,
368 // Overridden from Histogram:
369 HistogramType
GetHistogramType() const override
;
371 // Helper method for transforming an array of valid enumeration values
372 // to the std::vector<int> expected by UMA_HISTOGRAM_CUSTOM_ENUMERATION.
373 // This function ensures that a guard bucket exists right after any
374 // valid sample value (unless the next higher sample is also a valid value),
375 // so that invalid samples never fall into the same bucket as valid samples.
376 // TODO(kaiwang): Change name to ArrayToCustomEnumRanges.
377 static std::vector
<Sample
> ArrayToCustomRanges(const Sample
* values
,
380 CustomHistogram(const std::string
& name
,
381 const BucketRanges
* ranges
);
383 // HistogramBase implementation:
384 bool SerializeInfoImpl(Pickle
* pickle
) const override
;
386 double GetBucketSize(Count current
, size_t i
) const override
;
389 friend BASE_EXPORT_PRIVATE HistogramBase
* DeserializeHistogramInfo(
390 PickleIterator
* iter
);
391 static HistogramBase
* DeserializeInfoImpl(PickleIterator
* iter
);
393 static bool ValidateCustomRanges(const std::vector
<Sample
>& custom_ranges
);
394 static BucketRanges
* CreateBucketRangesFromCustomRanges(
395 const std::vector
<Sample
>& custom_ranges
);
397 DISALLOW_COPY_AND_ASSIGN(CustomHistogram
);
402 #endif // BASE_METRICS_HISTOGRAM_H_