1 /*
2 * Copyright (C) 2017 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #define DEBUG false // STOPSHIP if true
18 #include "Log.h"
19
20 #include "ValueMetricProducer.h"
21 #include "../guardrail/StatsdStats.h"
22 #include "../stats_log_util.h"
23
24 #include <limits.h>
25 #include <stdlib.h>
26
27 using android::util::FIELD_COUNT_REPEATED;
28 using android::util::FIELD_TYPE_BOOL;
29 using android::util::FIELD_TYPE_DOUBLE;
30 using android::util::FIELD_TYPE_INT32;
31 using android::util::FIELD_TYPE_INT64;
32 using android::util::FIELD_TYPE_MESSAGE;
33 using android::util::FIELD_TYPE_STRING;
34 using android::util::ProtoOutputStream;
35 using std::map;
36 using std::shared_ptr;
37 using std::unordered_map;
38
39 namespace android {
40 namespace os {
41 namespace statsd {
42
43 // for StatsLogReport
44 const int FIELD_ID_ID = 1;
45 const int FIELD_ID_VALUE_METRICS = 7;
46 const int FIELD_ID_TIME_BASE = 9;
47 const int FIELD_ID_BUCKET_SIZE = 10;
48 const int FIELD_ID_DIMENSION_PATH_IN_WHAT = 11;
49 const int FIELD_ID_IS_ACTIVE = 14;
50 // for ValueMetricDataWrapper
51 const int FIELD_ID_DATA = 1;
52 const int FIELD_ID_SKIPPED = 2;
53 // for SkippedBuckets
54 const int FIELD_ID_SKIPPED_START_MILLIS = 3;
55 const int FIELD_ID_SKIPPED_END_MILLIS = 4;
56 const int FIELD_ID_SKIPPED_DROP_EVENT = 5;
57 // for DumpEvent Proto
58 const int FIELD_ID_BUCKET_DROP_REASON = 1;
59 const int FIELD_ID_DROP_TIME = 2;
60 // for ValueMetricData
61 const int FIELD_ID_DIMENSION_IN_WHAT = 1;
62 const int FIELD_ID_BUCKET_INFO = 3;
63 const int FIELD_ID_DIMENSION_LEAF_IN_WHAT = 4;
64 const int FIELD_ID_SLICE_BY_STATE = 6;
65 // for ValueBucketInfo
66 const int FIELD_ID_VALUE_INDEX = 1;
67 const int FIELD_ID_VALUE_LONG = 2;
68 const int FIELD_ID_VALUE_DOUBLE = 3;
69 const int FIELD_ID_VALUES = 9;
70 const int FIELD_ID_BUCKET_NUM = 4;
71 const int FIELD_ID_START_BUCKET_ELAPSED_MILLIS = 5;
72 const int FIELD_ID_END_BUCKET_ELAPSED_MILLIS = 6;
73 const int FIELD_ID_CONDITION_TRUE_NS = 10;
74
75 const Value ZERO_LONG((int64_t)0);
76 const Value ZERO_DOUBLE((int64_t)0);
77
78 // ValueMetric has a minimum bucket size of 10min so that we don't pull too frequently
ValueMetricProducer(const ConfigKey & key,const ValueMetric & metric,const int conditionIndex,const vector<ConditionState> & initialConditionCache,const sp<ConditionWizard> & conditionWizard,const int whatMatcherIndex,const sp<EventMatcherWizard> & matcherWizard,const int pullTagId,const int64_t timeBaseNs,const int64_t startTimeNs,const sp<StatsPullerManager> & pullerManager,const unordered_map<int,shared_ptr<Activation>> & eventActivationMap,const unordered_map<int,vector<shared_ptr<Activation>>> & eventDeactivationMap,const vector<int> & slicedStateAtoms,const unordered_map<int,unordered_map<int,int64_t>> & stateGroupMap)79 ValueMetricProducer::ValueMetricProducer(
80 const ConfigKey& key, const ValueMetric& metric, const int conditionIndex,
81 const vector<ConditionState>& initialConditionCache,
82 const sp<ConditionWizard>& conditionWizard, const int whatMatcherIndex,
83 const sp<EventMatcherWizard>& matcherWizard, const int pullTagId, const int64_t timeBaseNs,
84 const int64_t startTimeNs, const sp<StatsPullerManager>& pullerManager,
85 const unordered_map<int, shared_ptr<Activation>>& eventActivationMap,
86 const unordered_map<int, vector<shared_ptr<Activation>>>& eventDeactivationMap,
87 const vector<int>& slicedStateAtoms,
88 const unordered_map<int, unordered_map<int, int64_t>>& stateGroupMap)
89 : MetricProducer(metric.id(), key, timeBaseNs, conditionIndex, initialConditionCache,
90 conditionWizard, eventActivationMap, eventDeactivationMap, slicedStateAtoms,
91 stateGroupMap),
92 mWhatMatcherIndex(whatMatcherIndex),
93 mEventMatcherWizard(matcherWizard),
94 mPullerManager(pullerManager),
95 mPullTagId(pullTagId),
96 mIsPulled(pullTagId != -1),
97 mMinBucketSizeNs(metric.min_bucket_size_nanos()),
98 mDimensionSoftLimit(StatsdStats::kAtomDimensionKeySizeLimitMap.find(pullTagId) !=
99 StatsdStats::kAtomDimensionKeySizeLimitMap.end()
100 ? StatsdStats::kAtomDimensionKeySizeLimitMap.at(pullTagId).first
101 : StatsdStats::kDimensionKeySizeSoftLimit),
102 mDimensionHardLimit(StatsdStats::kAtomDimensionKeySizeLimitMap.find(pullTagId) !=
103 StatsdStats::kAtomDimensionKeySizeLimitMap.end()
104 ? StatsdStats::kAtomDimensionKeySizeLimitMap.at(pullTagId).second
105 : StatsdStats::kDimensionKeySizeHardLimit),
106 mUseAbsoluteValueOnReset(metric.use_absolute_value_on_reset()),
107 mAggregationType(metric.aggregation_type()),
108 mUseDiff(metric.has_use_diff() ? metric.use_diff() : (mIsPulled ? true : false)),
109 mValueDirection(metric.value_direction()),
110 mSkipZeroDiffOutput(metric.skip_zero_diff_output()),
111 mUseZeroDefaultBase(metric.use_zero_default_base()),
112 mHasGlobalBase(false),
113 mCurrentBucketIsSkipped(false),
114 mMaxPullDelayNs(metric.max_pull_delay_sec() > 0 ? metric.max_pull_delay_sec() * NS_PER_SEC
115 : StatsdStats::kPullMaxDelayNs),
116 mSplitBucketForAppUpgrade(metric.split_bucket_for_app_upgrade()),
117 // Condition timer will be set later within the constructor after pulling events
118 mConditionTimer(false, timeBaseNs) {
119 int64_t bucketSizeMills = 0;
120 if (metric.has_bucket()) {
121 bucketSizeMills = TimeUnitToBucketSizeInMillisGuardrailed(key.GetUid(), metric.bucket());
122 } else {
123 bucketSizeMills = TimeUnitToBucketSizeInMillis(ONE_HOUR);
124 }
125
126 mBucketSizeNs = bucketSizeMills * 1000000;
127
128 translateFieldMatcher(metric.value_field(), &mFieldMatchers);
129
130 if (metric.has_dimensions_in_what()) {
131 translateFieldMatcher(metric.dimensions_in_what(), &mDimensionsInWhat);
132 mContainANYPositionInDimensionsInWhat = HasPositionANY(metric.dimensions_in_what());
133 mSliceByPositionALL = HasPositionALL(metric.dimensions_in_what());
134 }
135
136 if (metric.links().size() > 0) {
137 for (const auto& link : metric.links()) {
138 Metric2Condition mc;
139 mc.conditionId = link.condition();
140 translateFieldMatcher(link.fields_in_what(), &mc.metricFields);
141 translateFieldMatcher(link.fields_in_condition(), &mc.conditionFields);
142 mMetric2ConditionLinks.push_back(mc);
143 }
144 mConditionSliced = true;
145 }
146
147 for (const auto& stateLink : metric.state_link()) {
148 Metric2State ms;
149 ms.stateAtomId = stateLink.state_atom_id();
150 translateFieldMatcher(stateLink.fields_in_what(), &ms.metricFields);
151 translateFieldMatcher(stateLink.fields_in_state(), &ms.stateFields);
152 mMetric2StateLinks.push_back(ms);
153 }
154
155 int64_t numBucketsForward = calcBucketsForwardCount(startTimeNs);
156 mCurrentBucketNum += numBucketsForward;
157
158 flushIfNeededLocked(startTimeNs);
159
160 if (mIsPulled) {
161 mPullerManager->RegisterReceiver(mPullTagId, mConfigKey, this, getCurrentBucketEndTimeNs(),
162 mBucketSizeNs);
163 }
164
165 // Only do this for partial buckets like first bucket. All other buckets should use
166 // flushIfNeeded to adjust start and end to bucket boundaries.
167 // Adjust start for partial bucket
168 mCurrentBucketStartTimeNs = startTimeNs;
169 mConditionTimer.newBucketStart(mCurrentBucketStartTimeNs);
170
171 // Now that activations are processed, start the condition timer if needed.
172 mConditionTimer.onConditionChanged(mIsActive && mCondition == ConditionState::kTrue,
173 mCurrentBucketStartTimeNs);
174
175 VLOG("value metric %lld created. bucket size %lld start_time: %lld", (long long)metric.id(),
176 (long long)mBucketSizeNs, (long long)mTimeBaseNs);
177 }
178
~ValueMetricProducer()179 ValueMetricProducer::~ValueMetricProducer() {
180 VLOG("~ValueMetricProducer() called");
181 if (mIsPulled) {
182 mPullerManager->UnRegisterReceiver(mPullTagId, mConfigKey, this);
183 }
184 }
185
onStateChanged(int64_t eventTimeNs,int32_t atomId,const HashableDimensionKey & primaryKey,const FieldValue & oldState,const FieldValue & newState)186 void ValueMetricProducer::onStateChanged(int64_t eventTimeNs, int32_t atomId,
187 const HashableDimensionKey& primaryKey,
188 const FieldValue& oldState, const FieldValue& newState) {
189 VLOG("ValueMetric %lld onStateChanged time %lld, State %d, key %s, %d -> %d",
190 (long long)mMetricId, (long long)eventTimeNs, atomId, primaryKey.toString().c_str(),
191 oldState.mValue.int_value, newState.mValue.int_value);
192
193 // If old and new states are in the same StateGroup, then we do not need to
194 // pull for this state change.
195 FieldValue oldStateCopy = oldState;
196 FieldValue newStateCopy = newState;
197 mapStateValue(atomId, &oldStateCopy);
198 mapStateValue(atomId, &newStateCopy);
199 if (oldStateCopy == newStateCopy) {
200 return;
201 }
202
203 // If condition is not true or metric is not active, we do not need to pull
204 // for this state change.
205 if (mCondition != ConditionState::kTrue || !mIsActive) {
206 return;
207 }
208
209 bool isEventLate = eventTimeNs < mCurrentBucketStartTimeNs;
210 if (isEventLate) {
211 VLOG("Skip event due to late arrival: %lld vs %lld", (long long)eventTimeNs,
212 (long long)mCurrentBucketStartTimeNs);
213 invalidateCurrentBucket(eventTimeNs, BucketDropReason::EVENT_IN_WRONG_BUCKET);
214 return;
215 }
216 mStateChangePrimaryKey.first = atomId;
217 mStateChangePrimaryKey.second = primaryKey;
218 if (mIsPulled) {
219 pullAndMatchEventsLocked(eventTimeNs);
220 }
221 mStateChangePrimaryKey.first = 0;
222 mStateChangePrimaryKey.second = DEFAULT_DIMENSION_KEY;
223 flushIfNeededLocked(eventTimeNs);
224 }
225
onSlicedConditionMayChangeLocked(bool overallCondition,const int64_t eventTime)226 void ValueMetricProducer::onSlicedConditionMayChangeLocked(bool overallCondition,
227 const int64_t eventTime) {
228 VLOG("Metric %lld onSlicedConditionMayChange", (long long)mMetricId);
229 }
230
dropDataLocked(const int64_t dropTimeNs)231 void ValueMetricProducer::dropDataLocked(const int64_t dropTimeNs) {
232 StatsdStats::getInstance().noteBucketDropped(mMetricId);
233
234 // The current partial bucket is not flushed and does not require a pull,
235 // so the data is still valid.
236 flushIfNeededLocked(dropTimeNs);
237 clearPastBucketsLocked(dropTimeNs);
238 }
239
clearPastBucketsLocked(const int64_t dumpTimeNs)240 void ValueMetricProducer::clearPastBucketsLocked(const int64_t dumpTimeNs) {
241 mPastBuckets.clear();
242 mSkippedBuckets.clear();
243 }
244
onDumpReportLocked(const int64_t dumpTimeNs,const bool include_current_partial_bucket,const bool erase_data,const DumpLatency dumpLatency,std::set<string> * str_set,ProtoOutputStream * protoOutput)245 void ValueMetricProducer::onDumpReportLocked(const int64_t dumpTimeNs,
246 const bool include_current_partial_bucket,
247 const bool erase_data,
248 const DumpLatency dumpLatency,
249 std::set<string> *str_set,
250 ProtoOutputStream* protoOutput) {
251 VLOG("metric %lld dump report now...", (long long)mMetricId);
252 if (include_current_partial_bucket) {
253 // For pull metrics, we need to do a pull at bucket boundaries. If we do not do that the
254 // current bucket will have incomplete data and the next will have the wrong snapshot to do
255 // a diff against. If the condition is false, we are fine since the base data is reset and
256 // we are not tracking anything.
257 bool pullNeeded = mIsPulled && mCondition == ConditionState::kTrue;
258 if (pullNeeded) {
259 switch (dumpLatency) {
260 case FAST:
261 invalidateCurrentBucket(dumpTimeNs, BucketDropReason::DUMP_REPORT_REQUESTED);
262 break;
263 case NO_TIME_CONSTRAINTS:
264 pullAndMatchEventsLocked(dumpTimeNs);
265 break;
266 }
267 }
268 flushCurrentBucketLocked(dumpTimeNs, dumpTimeNs);
269 }
270 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_ID, (long long)mMetricId);
271 protoOutput->write(FIELD_TYPE_BOOL | FIELD_ID_IS_ACTIVE, isActiveLocked());
272
273 if (mPastBuckets.empty() && mSkippedBuckets.empty()) {
274 return;
275 }
276 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_TIME_BASE, (long long)mTimeBaseNs);
277 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_BUCKET_SIZE, (long long)mBucketSizeNs);
278 // Fills the dimension path if not slicing by ALL.
279 if (!mSliceByPositionALL) {
280 if (!mDimensionsInWhat.empty()) {
281 uint64_t dimenPathToken =
282 protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_ID_DIMENSION_PATH_IN_WHAT);
283 writeDimensionPathToProto(mDimensionsInWhat, protoOutput);
284 protoOutput->end(dimenPathToken);
285 }
286 }
287
288 uint64_t protoToken = protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_ID_VALUE_METRICS);
289
290 for (const auto& skippedBucket : mSkippedBuckets) {
291 uint64_t wrapperToken =
292 protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED | FIELD_ID_SKIPPED);
293 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_SKIPPED_START_MILLIS,
294 (long long)(NanoToMillis(skippedBucket.bucketStartTimeNs)));
295 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_SKIPPED_END_MILLIS,
296 (long long)(NanoToMillis(skippedBucket.bucketEndTimeNs)));
297 for (const auto& dropEvent : skippedBucket.dropEvents) {
298 uint64_t dropEventToken = protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED |
299 FIELD_ID_SKIPPED_DROP_EVENT);
300 protoOutput->write(FIELD_TYPE_INT32 | FIELD_ID_BUCKET_DROP_REASON, dropEvent.reason);
301 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_DROP_TIME,
302 (long long)(NanoToMillis(dropEvent.dropTimeNs)));
303 ;
304 protoOutput->end(dropEventToken);
305 }
306 protoOutput->end(wrapperToken);
307 }
308
309 for (const auto& pair : mPastBuckets) {
310 const MetricDimensionKey& dimensionKey = pair.first;
311 VLOG(" dimension key %s", dimensionKey.toString().c_str());
312 uint64_t wrapperToken =
313 protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED | FIELD_ID_DATA);
314
315 // First fill dimension.
316 if (mSliceByPositionALL) {
317 uint64_t dimensionToken =
318 protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_ID_DIMENSION_IN_WHAT);
319 writeDimensionToProto(dimensionKey.getDimensionKeyInWhat(), str_set, protoOutput);
320 protoOutput->end(dimensionToken);
321 } else {
322 writeDimensionLeafNodesToProto(dimensionKey.getDimensionKeyInWhat(),
323 FIELD_ID_DIMENSION_LEAF_IN_WHAT, str_set, protoOutput);
324 }
325
326 // Then fill slice_by_state.
327 for (auto state : dimensionKey.getStateValuesKey().getValues()) {
328 uint64_t stateToken = protoOutput->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED |
329 FIELD_ID_SLICE_BY_STATE);
330 writeStateToProto(state, protoOutput);
331 protoOutput->end(stateToken);
332 }
333
334 // Then fill bucket_info (ValueBucketInfo).
335 for (const auto& bucket : pair.second) {
336 uint64_t bucketInfoToken = protoOutput->start(
337 FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED | FIELD_ID_BUCKET_INFO);
338
339 if (bucket.mBucketEndNs - bucket.mBucketStartNs != mBucketSizeNs) {
340 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_START_BUCKET_ELAPSED_MILLIS,
341 (long long)NanoToMillis(bucket.mBucketStartNs));
342 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_END_BUCKET_ELAPSED_MILLIS,
343 (long long)NanoToMillis(bucket.mBucketEndNs));
344 } else {
345 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_BUCKET_NUM,
346 (long long)(getBucketNumFromEndTimeNs(bucket.mBucketEndNs)));
347 }
348 // only write the condition timer value if the metric has a condition.
349 if (mConditionTrackerIndex >= 0) {
350 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_CONDITION_TRUE_NS,
351 (long long)bucket.mConditionTrueNs);
352 }
353 for (int i = 0; i < (int)bucket.valueIndex.size(); i++) {
354 int index = bucket.valueIndex[i];
355 const Value& value = bucket.values[i];
356 uint64_t valueToken = protoOutput->start(
357 FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED | FIELD_ID_VALUES);
358 protoOutput->write(FIELD_TYPE_INT32 | FIELD_ID_VALUE_INDEX,
359 index);
360 if (value.getType() == LONG) {
361 protoOutput->write(FIELD_TYPE_INT64 | FIELD_ID_VALUE_LONG,
362 (long long)value.long_value);
363 VLOG("\t bucket [%lld - %lld] value %d: %lld", (long long)bucket.mBucketStartNs,
364 (long long)bucket.mBucketEndNs, index, (long long)value.long_value);
365 } else if (value.getType() == DOUBLE) {
366 protoOutput->write(FIELD_TYPE_DOUBLE | FIELD_ID_VALUE_DOUBLE,
367 value.double_value);
368 VLOG("\t bucket [%lld - %lld] value %d: %.2f", (long long)bucket.mBucketStartNs,
369 (long long)bucket.mBucketEndNs, index, value.double_value);
370 } else {
371 VLOG("Wrong value type for ValueMetric output: %d", value.getType());
372 }
373 protoOutput->end(valueToken);
374 }
375 protoOutput->end(bucketInfoToken);
376 }
377 protoOutput->end(wrapperToken);
378 }
379 protoOutput->end(protoToken);
380
381 VLOG("metric %lld dump report now...", (long long)mMetricId);
382 if (erase_data) {
383 mPastBuckets.clear();
384 mSkippedBuckets.clear();
385 }
386 }
387
invalidateCurrentBucketWithoutResetBase(const int64_t dropTimeNs,const BucketDropReason reason)388 void ValueMetricProducer::invalidateCurrentBucketWithoutResetBase(const int64_t dropTimeNs,
389 const BucketDropReason reason) {
390 if (!mCurrentBucketIsSkipped) {
391 // Only report to StatsdStats once per invalid bucket.
392 StatsdStats::getInstance().noteInvalidatedBucket(mMetricId);
393 }
394
395 skipCurrentBucket(dropTimeNs, reason);
396 }
397
invalidateCurrentBucket(const int64_t dropTimeNs,const BucketDropReason reason)398 void ValueMetricProducer::invalidateCurrentBucket(const int64_t dropTimeNs,
399 const BucketDropReason reason) {
400 invalidateCurrentBucketWithoutResetBase(dropTimeNs, reason);
401 resetBase();
402 }
403
skipCurrentBucket(const int64_t dropTimeNs,const BucketDropReason reason)404 void ValueMetricProducer::skipCurrentBucket(const int64_t dropTimeNs,
405 const BucketDropReason reason) {
406 if (!maxDropEventsReached()) {
407 mCurrentSkippedBucket.dropEvents.emplace_back(buildDropEvent(dropTimeNs, reason));
408 }
409 mCurrentBucketIsSkipped = true;
410 }
411
resetBase()412 void ValueMetricProducer::resetBase() {
413 for (auto& slice : mCurrentBaseInfo) {
414 for (auto& baseInfo : slice.second) {
415 baseInfo.hasBase = false;
416 }
417 }
418 mHasGlobalBase = false;
419 }
420
421 // Handle active state change. Active state change is treated like a condition change:
422 // - drop bucket if active state change event arrives too late
423 // - if condition is true, pull data on active state changes
424 // - ConditionTimer tracks changes based on AND of condition and active state.
onActiveStateChangedLocked(const int64_t & eventTimeNs)425 void ValueMetricProducer::onActiveStateChangedLocked(const int64_t& eventTimeNs) {
426 bool isEventTooLate = eventTimeNs < mCurrentBucketStartTimeNs;
427 if (isEventTooLate) {
428 // Drop bucket because event arrived too late, ie. we are missing data for this bucket.
429 StatsdStats::getInstance().noteLateLogEventSkipped(mMetricId);
430 invalidateCurrentBucket(eventTimeNs, BucketDropReason::EVENT_IN_WRONG_BUCKET);
431 }
432
433 // Call parent method once we've verified the validity of current bucket.
434 MetricProducer::onActiveStateChangedLocked(eventTimeNs);
435
436 if (ConditionState::kTrue != mCondition) {
437 return;
438 }
439
440 // Pull on active state changes.
441 if (!isEventTooLate) {
442 if (mIsPulled) {
443 pullAndMatchEventsLocked(eventTimeNs);
444 }
445 // When active state changes from true to false, clear diff base but don't
446 // reset other counters as we may accumulate more value in the bucket.
447 if (mUseDiff && !mIsActive) {
448 resetBase();
449 }
450 }
451
452 flushIfNeededLocked(eventTimeNs);
453
454 // Let condition timer know of new active state.
455 mConditionTimer.onConditionChanged(mIsActive, eventTimeNs);
456 }
457
onConditionChangedLocked(const bool condition,const int64_t eventTimeNs)458 void ValueMetricProducer::onConditionChangedLocked(const bool condition,
459 const int64_t eventTimeNs) {
460 ConditionState newCondition = condition ? ConditionState::kTrue : ConditionState::kFalse;
461 bool isEventTooLate = eventTimeNs < mCurrentBucketStartTimeNs;
462
463 // If the config is not active, skip the event.
464 if (!mIsActive) {
465 mCondition = isEventTooLate ? ConditionState::kUnknown : newCondition;
466 return;
467 }
468
469 // If the event arrived late, mark the bucket as invalid and skip the event.
470 if (isEventTooLate) {
471 VLOG("Skip event due to late arrival: %lld vs %lld", (long long)eventTimeNs,
472 (long long)mCurrentBucketStartTimeNs);
473 StatsdStats::getInstance().noteLateLogEventSkipped(mMetricId);
474 StatsdStats::getInstance().noteConditionChangeInNextBucket(mMetricId);
475 invalidateCurrentBucket(eventTimeNs, BucketDropReason::EVENT_IN_WRONG_BUCKET);
476 mCondition = ConditionState::kUnknown;
477 mConditionTimer.onConditionChanged(mCondition, eventTimeNs);
478 return;
479 }
480
481 // If the previous condition was unknown, mark the bucket as invalid
482 // because the bucket will contain partial data. For example, the condition
483 // change might happen close to the end of the bucket and we might miss a
484 // lot of data.
485 //
486 // We still want to pull to set the base.
487 if (mCondition == ConditionState::kUnknown) {
488 invalidateCurrentBucket(eventTimeNs, BucketDropReason::CONDITION_UNKNOWN);
489 }
490
491 // Pull and match for the following condition change cases:
492 // unknown/false -> true - condition changed
493 // true -> false - condition changed
494 // true -> true - old condition was true so we can flush the bucket at the
495 // end if needed.
496 //
497 // We don’t need to pull for unknown -> false or false -> false.
498 //
499 // onConditionChangedLocked might happen on bucket boundaries if this is
500 // called before #onDataPulled.
501 if (mIsPulled &&
502 (newCondition == ConditionState::kTrue || mCondition == ConditionState::kTrue)) {
503 pullAndMatchEventsLocked(eventTimeNs);
504 }
505
506 // For metrics that use diff, when condition changes from true to false,
507 // clear diff base but don't reset other counts because we may accumulate
508 // more value in the bucket.
509 if (mUseDiff &&
510 (mCondition == ConditionState::kTrue && newCondition == ConditionState::kFalse)) {
511 resetBase();
512 }
513
514 // Update condition state after pulling.
515 mCondition = newCondition;
516
517 flushIfNeededLocked(eventTimeNs);
518 mConditionTimer.onConditionChanged(mCondition, eventTimeNs);
519 }
520
prepareFirstBucketLocked()521 void ValueMetricProducer::prepareFirstBucketLocked() {
522 // Kicks off the puller immediately if condition is true and diff based.
523 if (mIsActive && mIsPulled && mCondition == ConditionState::kTrue && mUseDiff) {
524 pullAndMatchEventsLocked(mCurrentBucketStartTimeNs);
525 }
526 }
527
pullAndMatchEventsLocked(const int64_t timestampNs)528 void ValueMetricProducer::pullAndMatchEventsLocked(const int64_t timestampNs) {
529 vector<std::shared_ptr<LogEvent>> allData;
530 if (!mPullerManager->Pull(mPullTagId, mConfigKey, timestampNs, &allData)) {
531 ALOGE("Stats puller failed for tag: %d at %lld", mPullTagId, (long long)timestampNs);
532 invalidateCurrentBucket(timestampNs, BucketDropReason::PULL_FAILED);
533 return;
534 }
535
536 accumulateEvents(allData, timestampNs, timestampNs);
537 }
538
calcPreviousBucketEndTime(const int64_t currentTimeNs)539 int64_t ValueMetricProducer::calcPreviousBucketEndTime(const int64_t currentTimeNs) {
540 return mTimeBaseNs + ((currentTimeNs - mTimeBaseNs) / mBucketSizeNs) * mBucketSizeNs;
541 }
542
543 // By design, statsd pulls data at bucket boundaries using AlarmManager. These pulls are likely
544 // to be delayed. Other events like condition changes or app upgrade which are not based on
545 // AlarmManager might have arrived earlier and close the bucket.
onDataPulled(const std::vector<std::shared_ptr<LogEvent>> & allData,bool pullSuccess,int64_t originalPullTimeNs)546 void ValueMetricProducer::onDataPulled(const std::vector<std::shared_ptr<LogEvent>>& allData,
547 bool pullSuccess, int64_t originalPullTimeNs) {
548 std::lock_guard<std::mutex> lock(mMutex);
549 if (mCondition == ConditionState::kTrue) {
550 // If the pull failed, we won't be able to compute a diff.
551 if (!pullSuccess) {
552 invalidateCurrentBucket(originalPullTimeNs, BucketDropReason::PULL_FAILED);
553 } else {
554 bool isEventLate = originalPullTimeNs < getCurrentBucketEndTimeNs();
555 if (isEventLate) {
556 // If the event is late, we are in the middle of a bucket. Just
557 // process the data without trying to snap the data to the nearest bucket.
558 accumulateEvents(allData, originalPullTimeNs, originalPullTimeNs);
559 } else {
560 // For scheduled pulled data, the effective event time is snap to the nearest
561 // bucket end. In the case of waking up from a deep sleep state, we will
562 // attribute to the previous bucket end. If the sleep was long but not very
563 // long, we will be in the immediate next bucket. Previous bucket may get a
564 // larger number as we pull at a later time than real bucket end.
565 //
566 // If the sleep was very long, we skip more than one bucket before sleep. In
567 // this case, if the diff base will be cleared and this new data will serve as
568 // new diff base.
569 int64_t bucketEndTime = calcPreviousBucketEndTime(originalPullTimeNs) - 1;
570 StatsdStats::getInstance().noteBucketBoundaryDelayNs(
571 mMetricId, originalPullTimeNs - bucketEndTime);
572 accumulateEvents(allData, originalPullTimeNs, bucketEndTime);
573 }
574 }
575 }
576
577 // We can probably flush the bucket. Since we used bucketEndTime when calling
578 // #onMatchedLogEventInternalLocked, the current bucket will not have been flushed.
579 flushIfNeededLocked(originalPullTimeNs);
580 }
581
accumulateEvents(const std::vector<std::shared_ptr<LogEvent>> & allData,int64_t originalPullTimeNs,int64_t eventElapsedTimeNs)582 void ValueMetricProducer::accumulateEvents(const std::vector<std::shared_ptr<LogEvent>>& allData,
583 int64_t originalPullTimeNs, int64_t eventElapsedTimeNs) {
584 bool isEventLate = eventElapsedTimeNs < mCurrentBucketStartTimeNs;
585 if (isEventLate) {
586 VLOG("Skip bucket end pull due to late arrival: %lld vs %lld",
587 (long long)eventElapsedTimeNs, (long long)mCurrentBucketStartTimeNs);
588 StatsdStats::getInstance().noteLateLogEventSkipped(mMetricId);
589 invalidateCurrentBucket(eventElapsedTimeNs, BucketDropReason::EVENT_IN_WRONG_BUCKET);
590 return;
591 }
592
593 const int64_t elapsedRealtimeNs = getElapsedRealtimeNs();
594 const int64_t pullDelayNs = elapsedRealtimeNs - originalPullTimeNs;
595 StatsdStats::getInstance().notePullDelay(mPullTagId, pullDelayNs);
596 if (pullDelayNs > mMaxPullDelayNs) {
597 ALOGE("Pull finish too late for atom %d, longer than %lld", mPullTagId,
598 (long long)mMaxPullDelayNs);
599 StatsdStats::getInstance().notePullExceedMaxDelay(mPullTagId);
600 // We are missing one pull from the bucket which means we will not have a complete view of
601 // what's going on.
602 invalidateCurrentBucket(eventElapsedTimeNs, BucketDropReason::PULL_DELAYED);
603 return;
604 }
605
606 mMatchedMetricDimensionKeys.clear();
607 for (const auto& data : allData) {
608 LogEvent localCopy = data->makeCopy();
609 if (mEventMatcherWizard->matchLogEvent(localCopy, mWhatMatcherIndex) ==
610 MatchingState::kMatched) {
611 localCopy.setElapsedTimestampNs(eventElapsedTimeNs);
612 onMatchedLogEventLocked(mWhatMatcherIndex, localCopy);
613 }
614 }
615 // If a key that is:
616 // 1. Tracked in mCurrentSlicedBucket and
617 // 2. A superset of the current mStateChangePrimaryKey
618 // was not found in the new pulled data (i.e. not in mMatchedDimensionInWhatKeys)
619 // then we need to reset the base.
620 for (auto& slice : mCurrentSlicedBucket) {
621 const auto& whatKey = slice.first.getDimensionKeyInWhat();
622 bool presentInPulledData =
623 mMatchedMetricDimensionKeys.find(whatKey) != mMatchedMetricDimensionKeys.end();
624 if (!presentInPulledData && whatKey.contains(mStateChangePrimaryKey.second)) {
625 auto it = mCurrentBaseInfo.find(whatKey);
626 for (auto& baseInfo : it->second) {
627 baseInfo.hasBase = false;
628 }
629 }
630 }
631 mMatchedMetricDimensionKeys.clear();
632 mHasGlobalBase = true;
633
634 // If we reach the guardrail, we might have dropped some data which means the bucket is
635 // incomplete.
636 //
637 // The base also needs to be reset. If we do not have the full data, we might
638 // incorrectly compute the diff when mUseZeroDefaultBase is true since an existing key
639 // might be missing from mCurrentSlicedBucket.
640 if (hasReachedGuardRailLimit()) {
641 invalidateCurrentBucket(eventElapsedTimeNs, BucketDropReason::DIMENSION_GUARDRAIL_REACHED);
642 mCurrentSlicedBucket.clear();
643 }
644 }
645
dumpStatesLocked(FILE * out,bool verbose) const646 void ValueMetricProducer::dumpStatesLocked(FILE* out, bool verbose) const {
647 if (mCurrentSlicedBucket.size() == 0) {
648 return;
649 }
650
651 fprintf(out, "ValueMetric %lld dimension size %lu\n", (long long)mMetricId,
652 (unsigned long)mCurrentSlicedBucket.size());
653 if (verbose) {
654 for (const auto& it : mCurrentSlicedBucket) {
655 for (const auto& interval : it.second) {
656 fprintf(out, "\t(what)%s\t(states)%s (value)%s\n",
657 it.first.getDimensionKeyInWhat().toString().c_str(),
658 it.first.getStateValuesKey().toString().c_str(),
659 interval.value.toString().c_str());
660 }
661 }
662 }
663 }
664
hasReachedGuardRailLimit() const665 bool ValueMetricProducer::hasReachedGuardRailLimit() const {
666 return mCurrentSlicedBucket.size() >= mDimensionHardLimit;
667 }
668
hitGuardRailLocked(const MetricDimensionKey & newKey)669 bool ValueMetricProducer::hitGuardRailLocked(const MetricDimensionKey& newKey) {
670 // ===========GuardRail==============
671 // 1. Report the tuple count if the tuple count > soft limit
672 if (mCurrentSlicedBucket.find(newKey) != mCurrentSlicedBucket.end()) {
673 return false;
674 }
675 if (mCurrentSlicedBucket.size() > mDimensionSoftLimit - 1) {
676 size_t newTupleCount = mCurrentSlicedBucket.size() + 1;
677 StatsdStats::getInstance().noteMetricDimensionSize(mConfigKey, mMetricId, newTupleCount);
678 // 2. Don't add more tuples, we are above the allowed threshold. Drop the data.
679 if (hasReachedGuardRailLimit()) {
680 ALOGE("ValueMetric %lld dropping data for dimension key %s", (long long)mMetricId,
681 newKey.toString().c_str());
682 StatsdStats::getInstance().noteHardDimensionLimitReached(mMetricId);
683 return true;
684 }
685 }
686
687 return false;
688 }
689
hitFullBucketGuardRailLocked(const MetricDimensionKey & newKey)690 bool ValueMetricProducer::hitFullBucketGuardRailLocked(const MetricDimensionKey& newKey) {
691 // ===========GuardRail==============
692 // 1. Report the tuple count if the tuple count > soft limit
693 if (mCurrentFullBucket.find(newKey) != mCurrentFullBucket.end()) {
694 return false;
695 }
696 if (mCurrentFullBucket.size() > mDimensionSoftLimit - 1) {
697 size_t newTupleCount = mCurrentFullBucket.size() + 1;
698 // 2. Don't add more tuples, we are above the allowed threshold. Drop the data.
699 if (newTupleCount > mDimensionHardLimit) {
700 ALOGE("ValueMetric %lld dropping data for full bucket dimension key %s",
701 (long long)mMetricId,
702 newKey.toString().c_str());
703 return true;
704 }
705 }
706
707 return false;
708 }
709
getDoubleOrLong(const LogEvent & event,const Matcher & matcher,Value & ret)710 bool getDoubleOrLong(const LogEvent& event, const Matcher& matcher, Value& ret) {
711 for (const FieldValue& value : event.getValues()) {
712 if (value.mField.matches(matcher)) {
713 switch (value.mValue.type) {
714 case INT:
715 ret.setLong(value.mValue.int_value);
716 break;
717 case LONG:
718 ret.setLong(value.mValue.long_value);
719 break;
720 case FLOAT:
721 ret.setDouble(value.mValue.float_value);
722 break;
723 case DOUBLE:
724 ret.setDouble(value.mValue.double_value);
725 break;
726 default:
727 return false;
728 break;
729 }
730 return true;
731 }
732 }
733 return false;
734 }
735
onMatchedLogEventInternalLocked(const size_t matcherIndex,const MetricDimensionKey & eventKey,const ConditionKey & conditionKey,bool condition,const LogEvent & event,const map<int,HashableDimensionKey> & statePrimaryKeys)736 void ValueMetricProducer::onMatchedLogEventInternalLocked(
737 const size_t matcherIndex, const MetricDimensionKey& eventKey,
738 const ConditionKey& conditionKey, bool condition, const LogEvent& event,
739 const map<int, HashableDimensionKey>& statePrimaryKeys) {
740 auto whatKey = eventKey.getDimensionKeyInWhat();
741 auto stateKey = eventKey.getStateValuesKey();
742
743 // Skip this event if a state changed occurred for a different primary key.
744 auto it = statePrimaryKeys.find(mStateChangePrimaryKey.first);
745 // Check that both the atom id and the primary key are equal.
746 if (it != statePrimaryKeys.end() && it->second != mStateChangePrimaryKey.second) {
747 VLOG("ValueMetric skip event with primary key %s because state change primary key "
748 "is %s",
749 it->second.toString().c_str(), mStateChangePrimaryKey.second.toString().c_str());
750 return;
751 }
752
753 int64_t eventTimeNs = event.GetElapsedTimestampNs();
754 if (eventTimeNs < mCurrentBucketStartTimeNs) {
755 VLOG("Skip event due to late arrival: %lld vs %lld", (long long)eventTimeNs,
756 (long long)mCurrentBucketStartTimeNs);
757 return;
758 }
759 mMatchedMetricDimensionKeys.insert(whatKey);
760
761 if (!mIsPulled) {
762 // We cannot flush without doing a pull first.
763 flushIfNeededLocked(eventTimeNs);
764 }
765
766 // We should not accumulate the data for pushed metrics when the condition is false.
767 bool shouldSkipForPushMetric = !mIsPulled && !condition;
768 // For pulled metrics, there are two cases:
769 // - to compute diffs, we need to process all the state changes
770 // - for non-diffs metrics, we should ignore the data if the condition wasn't true. If we have a
771 // state change from
772 // + True -> True: we should process the data, it might be a bucket boundary
773 // + True -> False: we als need to process the data.
774 bool shouldSkipForPulledMetric = mIsPulled && !mUseDiff
775 && mCondition != ConditionState::kTrue;
776 if (shouldSkipForPushMetric || shouldSkipForPulledMetric) {
777 VLOG("ValueMetric skip event because condition is false and we are not using diff (for "
778 "pulled metric)");
779 return;
780 }
781
782 if (hitGuardRailLocked(eventKey)) {
783 return;
784 }
785
786 vector<BaseInfo>& baseInfos = mCurrentBaseInfo[whatKey];
787 if (baseInfos.size() < mFieldMatchers.size()) {
788 VLOG("Resizing number of intervals to %d", (int)mFieldMatchers.size());
789 baseInfos.resize(mFieldMatchers.size());
790 }
791
792 for (BaseInfo& baseInfo : baseInfos) {
793 if (!baseInfo.hasCurrentState) {
794 baseInfo.currentState = getUnknownStateKey();
795 baseInfo.hasCurrentState = true;
796 }
797 }
798
799 // We need to get the intervals stored with the previous state key so we can
800 // close these value intervals.
801 const auto oldStateKey = baseInfos[0].currentState;
802 vector<Interval>& intervals = mCurrentSlicedBucket[MetricDimensionKey(whatKey, oldStateKey)];
803 if (intervals.size() < mFieldMatchers.size()) {
804 VLOG("Resizing number of intervals to %d", (int)mFieldMatchers.size());
805 intervals.resize(mFieldMatchers.size());
806 }
807
808 // We only use anomaly detection under certain cases.
809 // N.B.: The anomaly detection cases were modified in order to fix an issue with value metrics
810 // containing multiple values. We tried to retain all previous behaviour, but we are unsure the
811 // previous behaviour was correct. At the time of the fix, anomaly detection had no owner.
812 // Whoever next works on it should look into the cases where it is triggered in this function.
813 // Discussion here: http://ag/6124370.
814 bool useAnomalyDetection = true;
815
816 for (int i = 0; i < (int)mFieldMatchers.size(); i++) {
817 const Matcher& matcher = mFieldMatchers[i];
818 BaseInfo& baseInfo = baseInfos[i];
819 Interval& interval = intervals[i];
820 interval.valueIndex = i;
821 Value value;
822 baseInfo.hasCurrentState = true;
823 baseInfo.currentState = stateKey;
824 if (!getDoubleOrLong(event, matcher, value)) {
825 VLOG("Failed to get value %d from event %s", i, event.ToString().c_str());
826 StatsdStats::getInstance().noteBadValueType(mMetricId);
827 return;
828 }
829 interval.seenNewData = true;
830
831 if (mUseDiff) {
832 if (!baseInfo.hasBase) {
833 if (mHasGlobalBase && mUseZeroDefaultBase) {
834 // The bucket has global base. This key does not.
835 // Optionally use zero as base.
836 baseInfo.base = (value.type == LONG ? ZERO_LONG : ZERO_DOUBLE);
837 baseInfo.hasBase = true;
838 } else {
839 // no base. just update base and return.
840 baseInfo.base = value;
841 baseInfo.hasBase = true;
842 // If we're missing a base, do not use anomaly detection on incomplete data
843 useAnomalyDetection = false;
844 // Continue (instead of return) here in order to set baseInfo.base and
845 // baseInfo.hasBase for other baseInfos
846 continue;
847 }
848 }
849
850 Value diff;
851 switch (mValueDirection) {
852 case ValueMetric::INCREASING:
853 if (value >= baseInfo.base) {
854 diff = value - baseInfo.base;
855 } else if (mUseAbsoluteValueOnReset) {
856 diff = value;
857 } else {
858 VLOG("Unexpected decreasing value");
859 StatsdStats::getInstance().notePullDataError(mPullTagId);
860 baseInfo.base = value;
861 // If we've got bad data, do not use anomaly detection
862 useAnomalyDetection = false;
863 continue;
864 }
865 break;
866 case ValueMetric::DECREASING:
867 if (baseInfo.base >= value) {
868 diff = baseInfo.base - value;
869 } else if (mUseAbsoluteValueOnReset) {
870 diff = value;
871 } else {
872 VLOG("Unexpected increasing value");
873 StatsdStats::getInstance().notePullDataError(mPullTagId);
874 baseInfo.base = value;
875 // If we've got bad data, do not use anomaly detection
876 useAnomalyDetection = false;
877 continue;
878 }
879 break;
880 case ValueMetric::ANY:
881 diff = value - baseInfo.base;
882 break;
883 default:
884 break;
885 }
886 baseInfo.base = value;
887 value = diff;
888 }
889
890 if (interval.hasValue) {
891 switch (mAggregationType) {
892 case ValueMetric::SUM:
893 // for AVG, we add up and take average when flushing the bucket
894 case ValueMetric::AVG:
895 interval.value += value;
896 break;
897 case ValueMetric::MIN:
898 interval.value = std::min(value, interval.value);
899 break;
900 case ValueMetric::MAX:
901 interval.value = std::max(value, interval.value);
902 break;
903 default:
904 break;
905 }
906 } else {
907 interval.value = value;
908 interval.hasValue = true;
909 }
910 interval.sampleSize += 1;
911 }
912
913 // Only trigger the tracker if all intervals are correct
914 if (useAnomalyDetection) {
915 // TODO: propgate proper values down stream when anomaly support doubles
916 long wholeBucketVal = intervals[0].value.long_value;
917 auto prev = mCurrentFullBucket.find(eventKey);
918 if (prev != mCurrentFullBucket.end()) {
919 wholeBucketVal += prev->second;
920 }
921 for (auto& tracker : mAnomalyTrackers) {
922 tracker->detectAndDeclareAnomaly(eventTimeNs, mCurrentBucketNum, mMetricId, eventKey,
923 wholeBucketVal);
924 }
925 }
926 }
927
928 // For pulled metrics, we always need to make sure we do a pull before flushing the bucket
929 // if mCondition is true!
flushIfNeededLocked(const int64_t & eventTimeNs)930 void ValueMetricProducer::flushIfNeededLocked(const int64_t& eventTimeNs) {
931 int64_t currentBucketEndTimeNs = getCurrentBucketEndTimeNs();
932 if (eventTimeNs < currentBucketEndTimeNs) {
933 VLOG("eventTime is %lld, less than current bucket end time %lld", (long long)eventTimeNs,
934 (long long)(currentBucketEndTimeNs));
935 return;
936 }
937 int64_t numBucketsForward = calcBucketsForwardCount(eventTimeNs);
938 int64_t nextBucketStartTimeNs = currentBucketEndTimeNs + (numBucketsForward - 1) * mBucketSizeNs;
939 flushCurrentBucketLocked(eventTimeNs, nextBucketStartTimeNs);
940 }
941
calcBucketsForwardCount(const int64_t & eventTimeNs) const942 int64_t ValueMetricProducer::calcBucketsForwardCount(const int64_t& eventTimeNs) const {
943 int64_t currentBucketEndTimeNs = getCurrentBucketEndTimeNs();
944 if (eventTimeNs < currentBucketEndTimeNs) {
945 return 0;
946 }
947 return 1 + (eventTimeNs - currentBucketEndTimeNs) / mBucketSizeNs;
948 }
949
flushCurrentBucketLocked(const int64_t & eventTimeNs,const int64_t & nextBucketStartTimeNs)950 void ValueMetricProducer::flushCurrentBucketLocked(const int64_t& eventTimeNs,
951 const int64_t& nextBucketStartTimeNs) {
952 if (mCondition == ConditionState::kUnknown) {
953 StatsdStats::getInstance().noteBucketUnknownCondition(mMetricId);
954 invalidateCurrentBucketWithoutResetBase(eventTimeNs, BucketDropReason::CONDITION_UNKNOWN);
955 }
956
957 VLOG("finalizing bucket for %ld, dumping %d slices", (long)mCurrentBucketStartTimeNs,
958 (int)mCurrentSlicedBucket.size());
959
960 int64_t fullBucketEndTimeNs = getCurrentBucketEndTimeNs();
961 int64_t bucketEndTime = fullBucketEndTimeNs;
962 int64_t numBucketsForward = calcBucketsForwardCount(eventTimeNs);
963
964 // Skip buckets if this is a pulled metric or a pushed metric that is diffed.
965 if (numBucketsForward > 1 && (mIsPulled || mUseDiff)) {
966
967 VLOG("Skipping forward %lld buckets", (long long)numBucketsForward);
968 StatsdStats::getInstance().noteSkippedForwardBuckets(mMetricId);
969 // Something went wrong. Maybe the device was sleeping for a long time. It is better
970 // to mark the current bucket as invalid. The last pull might have been successful through.
971 invalidateCurrentBucketWithoutResetBase(eventTimeNs,
972 BucketDropReason::MULTIPLE_BUCKETS_SKIPPED);
973 // End the bucket at the next bucket start time so the entire interval is skipped.
974 bucketEndTime = nextBucketStartTimeNs;
975 } else if (eventTimeNs < fullBucketEndTimeNs) {
976 bucketEndTime = eventTimeNs;
977 }
978
979 // Close the current bucket.
980 int64_t conditionTrueDuration = mConditionTimer.newBucketStart(bucketEndTime);
981 bool isBucketLargeEnough = bucketEndTime - mCurrentBucketStartTimeNs >= mMinBucketSizeNs;
982 if (!isBucketLargeEnough) {
983 skipCurrentBucket(eventTimeNs, BucketDropReason::BUCKET_TOO_SMALL);
984 }
985 if (!mCurrentBucketIsSkipped) {
986 bool bucketHasData = false;
987 // The current bucket is large enough to keep.
988 for (const auto& slice : mCurrentSlicedBucket) {
989 ValueBucket bucket = buildPartialBucket(bucketEndTime, slice.second);
990 bucket.mConditionTrueNs = conditionTrueDuration;
991 // it will auto create new vector of ValuebucketInfo if the key is not found.
992 if (bucket.valueIndex.size() > 0) {
993 auto& bucketList = mPastBuckets[slice.first];
994 bucketList.push_back(bucket);
995 bucketHasData = true;
996 }
997 }
998 if (!bucketHasData) {
999 skipCurrentBucket(eventTimeNs, BucketDropReason::NO_DATA);
1000 }
1001 }
1002
1003 if (mCurrentBucketIsSkipped) {
1004 mCurrentSkippedBucket.bucketStartTimeNs = mCurrentBucketStartTimeNs;
1005 mCurrentSkippedBucket.bucketEndTimeNs = bucketEndTime;
1006 mSkippedBuckets.emplace_back(mCurrentSkippedBucket);
1007 }
1008
1009 // This means that the current bucket was not flushed before a forced bucket split.
1010 // This can happen if an app update or a dump report with include_current_partial_bucket is
1011 // requested before we get a chance to flush the bucket due to receiving new data, either from
1012 // the statsd socket or the StatsPullerManager.
1013 if (bucketEndTime < nextBucketStartTimeNs) {
1014 SkippedBucket bucketInGap;
1015 bucketInGap.bucketStartTimeNs = bucketEndTime;
1016 bucketInGap.bucketEndTimeNs = nextBucketStartTimeNs;
1017 bucketInGap.dropEvents.emplace_back(
1018 buildDropEvent(eventTimeNs, BucketDropReason::NO_DATA));
1019 mSkippedBuckets.emplace_back(bucketInGap);
1020 }
1021
1022 appendToFullBucket(eventTimeNs > fullBucketEndTimeNs);
1023 initCurrentSlicedBucket(nextBucketStartTimeNs);
1024 // Update the condition timer again, in case we skipped buckets.
1025 mConditionTimer.newBucketStart(nextBucketStartTimeNs);
1026 mCurrentBucketNum += numBucketsForward;
1027 }
1028
buildPartialBucket(int64_t bucketEndTime,const std::vector<Interval> & intervals)1029 ValueBucket ValueMetricProducer::buildPartialBucket(int64_t bucketEndTime,
1030 const std::vector<Interval>& intervals) {
1031 ValueBucket bucket;
1032 bucket.mBucketStartNs = mCurrentBucketStartTimeNs;
1033 bucket.mBucketEndNs = bucketEndTime;
1034 for (const auto& interval : intervals) {
1035 if (interval.hasValue) {
1036 // skip the output if the diff is zero
1037 if (mSkipZeroDiffOutput && mUseDiff && interval.value.isZero()) {
1038 continue;
1039 }
1040 bucket.valueIndex.push_back(interval.valueIndex);
1041 if (mAggregationType != ValueMetric::AVG) {
1042 bucket.values.push_back(interval.value);
1043 } else {
1044 double sum = interval.value.type == LONG ? (double)interval.value.long_value
1045 : interval.value.double_value;
1046 bucket.values.push_back(Value((double)sum / interval.sampleSize));
1047 }
1048 }
1049 }
1050 return bucket;
1051 }
1052
initCurrentSlicedBucket(int64_t nextBucketStartTimeNs)1053 void ValueMetricProducer::initCurrentSlicedBucket(int64_t nextBucketStartTimeNs) {
1054 StatsdStats::getInstance().noteBucketCount(mMetricId);
1055 // Cleanup data structure to aggregate values.
1056 for (auto it = mCurrentSlicedBucket.begin(); it != mCurrentSlicedBucket.end();) {
1057 bool obsolete = true;
1058 for (auto& interval : it->second) {
1059 interval.hasValue = false;
1060 interval.sampleSize = 0;
1061 if (interval.seenNewData) {
1062 obsolete = false;
1063 }
1064 interval.seenNewData = false;
1065 }
1066
1067 if (obsolete) {
1068 it = mCurrentSlicedBucket.erase(it);
1069 } else {
1070 it++;
1071 }
1072 // TODO(b/157655103): remove mCurrentBaseInfo entries when obsolete
1073 }
1074
1075 mCurrentBucketIsSkipped = false;
1076 mCurrentSkippedBucket.reset();
1077
1078 // If we do not have a global base when the condition is true,
1079 // we will have incomplete bucket for the next bucket.
1080 if (mUseDiff && !mHasGlobalBase && mCondition) {
1081 mCurrentBucketIsSkipped = false;
1082 }
1083 mCurrentBucketStartTimeNs = nextBucketStartTimeNs;
1084 VLOG("metric %lld: new bucket start time: %lld", (long long)mMetricId,
1085 (long long)mCurrentBucketStartTimeNs);
1086 }
1087
appendToFullBucket(const bool isFullBucketReached)1088 void ValueMetricProducer::appendToFullBucket(const bool isFullBucketReached) {
1089 if (mCurrentBucketIsSkipped) {
1090 if (isFullBucketReached) {
1091 // If the bucket is invalid, we ignore the full bucket since it contains invalid data.
1092 mCurrentFullBucket.clear();
1093 }
1094 // Current bucket is invalid, we do not add it to the full bucket.
1095 return;
1096 }
1097
1098 if (isFullBucketReached) { // If full bucket, send to anomaly tracker.
1099 // Accumulate partial buckets with current value and then send to anomaly tracker.
1100 if (mCurrentFullBucket.size() > 0) {
1101 for (const auto& slice : mCurrentSlicedBucket) {
1102 if (hitFullBucketGuardRailLocked(slice.first)) {
1103 continue;
1104 }
1105 // TODO: fix this when anomaly can accept double values
1106 auto& interval = slice.second[0];
1107 if (interval.hasValue) {
1108 mCurrentFullBucket[slice.first] += interval.value.long_value;
1109 }
1110 }
1111 for (const auto& slice : mCurrentFullBucket) {
1112 for (auto& tracker : mAnomalyTrackers) {
1113 if (tracker != nullptr) {
1114 tracker->addPastBucket(slice.first, slice.second, mCurrentBucketNum);
1115 }
1116 }
1117 }
1118 mCurrentFullBucket.clear();
1119 } else {
1120 // Skip aggregating the partial buckets since there's no previous partial bucket.
1121 for (const auto& slice : mCurrentSlicedBucket) {
1122 for (auto& tracker : mAnomalyTrackers) {
1123 if (tracker != nullptr) {
1124 // TODO: fix this when anomaly can accept double values
1125 auto& interval = slice.second[0];
1126 if (interval.hasValue) {
1127 tracker->addPastBucket(slice.first, interval.value.long_value,
1128 mCurrentBucketNum);
1129 }
1130 }
1131 }
1132 }
1133 }
1134 } else {
1135 // Accumulate partial bucket.
1136 for (const auto& slice : mCurrentSlicedBucket) {
1137 // TODO: fix this when anomaly can accept double values
1138 auto& interval = slice.second[0];
1139 if (interval.hasValue) {
1140 mCurrentFullBucket[slice.first] += interval.value.long_value;
1141 }
1142 }
1143 }
1144 }
1145
byteSizeLocked() const1146 size_t ValueMetricProducer::byteSizeLocked() const {
1147 size_t totalSize = 0;
1148 for (const auto& pair : mPastBuckets) {
1149 totalSize += pair.second.size() * kBucketSize;
1150 }
1151 return totalSize;
1152 }
1153
1154 } // namespace statsd
1155 } // namespace os
1156 } // namespace android
1157