Merge "Add more statistics on failure cases for value metrics" (5222946e) · Commits · e / os / android_frameworks_base

cmds/statsd/src/external/StatsCompanionServicePuller.cpp

+4 −0

Original line number	Diff line number	Diff line
		@@ -55,6 +55,10 @@ bool StatsCompanionServicePuller::PullInternal(vector<shared_ptr<LogEvent> >* da
		Status status = statsCompanionServiceCopy->pullData(mTagId, &returned_value);
		if (!status.isOk()) {
		ALOGW("StatsCompanionServicePuller::pull failed for %d", mTagId);
		StatsdStats::getInstance().noteStatsCompanionPullFailed(mTagId);
		if (status.exceptionCode() == Status::Exception::EX_TRANSACTION_FAILED) {
		StatsdStats::getInstance().noteStatsCompanionPullBinderTransactionFailed(mTagId);
		}
		return false;
		}
		data->clear();

cmds/statsd/src/external/StatsPullerManager.cpp

+3 −0

Original line number	Diff line number	Diff line
		@@ -231,6 +231,9 @@ bool StatsPullerManager::Pull(int tagId, vector<shared_ptr<LogEvent>>* data) {
		if (kAllPullAtomInfo.find(tagId) != kAllPullAtomInfo.end()) {
		bool ret = kAllPullAtomInfo.find(tagId)->second.puller->Pull(data);
		VLOG("pulled %d items", (int)data->size());
		if (!ret) {
		StatsdStats::getInstance().notePullFailed(tagId);
		}
		return ret;
		} else {
		VLOG("Unknown tagId %d", tagId);

cmds/statsd/src/guardrail/StatsdStats.cpp

+59 −0

Original line number	Diff line number	Diff line
		@@ -403,6 +403,60 @@ void StatsdStats::noteSystemServerRestart(int32_t timeSec) {
		mSystemServerRestartSec.push_back(timeSec);
		}

		void StatsdStats::notePullFailed(int atomId) {
		lock_guard<std::mutex> lock(mLock);
		mPulledAtomStats[atomId].pullFailed++;
		}

		void StatsdStats::noteStatsCompanionPullFailed(int atomId) {
		lock_guard<std::mutex> lock(mLock);
		mPulledAtomStats[atomId].statsCompanionPullFailed++;
		}

		void StatsdStats::noteStatsCompanionPullBinderTransactionFailed(int atomId) {
		lock_guard<std::mutex> lock(mLock);
		mPulledAtomStats[atomId].statsCompanionPullBinderTransactionFailed++;
		}

		void StatsdStats::noteEmptyData(int atomId) {
		lock_guard<std::mutex> lock(mLock);
		mPulledAtomStats[atomId].emptyData++;
		}

		void StatsdStats::noteHardDimensionLimitReached(int metricId) {
		lock_guard<std::mutex> lock(mLock);
		getAtomMetricStats(metricId).hardDimensionLimitReached++;
		}

		void StatsdStats::noteLateLogEventSkipped(int metricId) {
		lock_guard<std::mutex> lock(mLock);
		getAtomMetricStats(metricId).lateLogEventSkipped++;
		}

		void StatsdStats::noteSkippedForwardBuckets(int metricId) {
		lock_guard<std::mutex> lock(mLock);
		getAtomMetricStats(metricId).skippedForwardBuckets++;
		}

		void StatsdStats::noteBadValueType(int metricId) {
		lock_guard<std::mutex> lock(mLock);
		getAtomMetricStats(metricId).badValueType++;
		}

		void StatsdStats::noteConditionChangeInNextBucket(int metricId) {
		lock_guard<std::mutex> lock(mLock);
		getAtomMetricStats(metricId).conditionChangeInNextBucket++;
		}

		StatsdStats::AtomMetricStats& StatsdStats::getAtomMetricStats(int metricId) {
		auto atomMetricStatsIter = mAtomMetricStats.find(metricId);
		if (atomMetricStatsIter != mAtomMetricStats.end()) {
		return atomMetricStatsIter->second;
		}
		auto emplaceResult = mAtomMetricStats.emplace(metricId, AtomMetricStats());
		return emplaceResult.first->second;
		}

		void StatsdStats::reset() {
		lock_guard<std::mutex> lock(mLock);
		resetInternalLocked();
		@@ -442,6 +496,7 @@ void StatsdStats::resetInternalLocked() {
		pullStats.second.pullTimeout = 0;
		pullStats.second.pullExceedMaxDelay = 0;
		}
		mAtomMetricStats.clear();
		}

		string buildTimeString(int64_t timeSec) {
		@@ -713,6 +768,10 @@ void StatsdStats::dumpStats(std::vector<uint8_t>* output, bool reset) {
		android::os::statsd::writePullerStatsToStream(pair, &proto);
		}

		for (const auto& pair : mAtomMetricStats) {
		android::os::statsd::writeAtomMetricStatsToStream(pair, &proto);
		}

		if (mAnomalyAlarmRegisteredStats > 0) {
		uint64_t token = proto.start(FIELD_TYPE_MESSAGE \| FIELD_ID_ANOMALY_ALARM_STATS);
		proto.write(FIELD_TYPE_INT32 \| FIELD_ID_ANOMALY_ALARMS_REGISTERED,

cmds/statsd/src/guardrail/StatsdStats.h

+68 −0

Original line number	Diff line number	Diff line
		@@ -317,6 +317,53 @@ public:
		*/
		void noteLogLost(int32_t wallClockTimeSec, int32_t count, int lastError);

		/**
		* Records that the pull of an atom has failed
		*/
		void notePullFailed(int atomId);

		/**
		* Records that the pull of StatsCompanionService atom has failed
		*/
		void noteStatsCompanionPullFailed(int atomId);

		/**
		* Records that the pull of a StatsCompanionService atom has failed due to a failed binder
		* transaction. This can happen when StatsCompanionService returns too
		* much data (the max Binder parcel size is 1MB)
		*/
		void noteStatsCompanionPullBinderTransactionFailed(int atomId);

		/**
		* A pull with no data occurred
		*/
		void noteEmptyData(int atomId);

		/**
		* Hard limit was reached in the cardinality of an atom
		*/
		void noteHardDimensionLimitReached(int atomId);

		/**
		* A log event was too late, arrived in the wrong bucket and was skipped
		*/
		void noteLateLogEventSkipped(int atomId);

		/**
		* Buckets were skipped as time elapsed without any data for them
		*/
		void noteSkippedForwardBuckets(int atomId);

		/**
		* An unsupported value type was received
		*/
		void noteBadValueType(int atomId);

		/**
		* A condition change was too late, arrived in the wrong bucket and was skipped
		*/
		void noteConditionChangeInNextBucket(int atomId);

		/**
		* Reset the historical stats. Including all stats in icebox, and the tracked stats about
		* metrics, matchers, and atoms. The active configs will be kept and StatsdStats will continue
		@@ -349,8 +396,20 @@ public:
		long dataError = 0;
		long pullTimeout = 0;
		long pullExceedMaxDelay = 0;
		long pullFailed = 0;
		long statsCompanionPullFailed = 0;
		long statsCompanionPullBinderTransactionFailed = 0;
		long emptyData = 0;
		} PulledAtomStats;

		typedef struct {
		long hardDimensionLimitReached = 0;
		long lateLogEventSkipped = 0;
		long skippedForwardBuckets = 0;
		long badValueType = 0;
		long conditionChangeInNextBucket = 0;
		} AtomMetricStats;

		private:
		StatsdStats();

		@@ -378,6 +437,9 @@ private:
		// Maps PullAtomId to its stats. The size is capped by the puller atom counts.
		std::map<int, PulledAtomStats> mPulledAtomStats;

		// Maps metric ID to its stats. The size is capped by the number of metrics.
		std::map<int, AtomMetricStats> mAtomMetricStats;

		struct LogLossStats {
		LogLossStats(int32_t sec, int32_t count, int32_t error)
		: mWallClockSec(sec), mCount(count), mLastError(error) {
		@@ -414,6 +476,12 @@ private:

		void addToIceBoxLocked(std::shared_ptr<ConfigStats>& stats);

		/**
		* Get a reference to AtomMetricStats for a metric. If none exists, create it. The reference
		* will live as long as `this`.
		*/
		StatsdStats::AtomMetricStats& getAtomMetricStats(int metricId);

		FRIEND_TEST(StatsdStatsTest, TestValidConfigAdd);
		FRIEND_TEST(StatsdStatsTest, TestInvalidConfigAdd);
		FRIEND_TEST(StatsdStatsTest, TestConfigRemove);

cmds/statsd/src/metrics/ValueMetricProducer.cpp

+12 −0

Original line number	Diff line number	Diff line
		@@ -322,6 +322,7 @@ void ValueMetricProducer::onConditionChangedLocked(const bool condition,
		if (eventTimeNs < mCurrentBucketStartTimeNs) {
		VLOG("Skip event due to late arrival: %lld vs %lld", (long long)eventTimeNs,
		(long long)mCurrentBucketStartTimeNs);
		StatsdStats::getInstance().noteConditionChangeInNextBucket(mMetricId);
		return;
		}

		@@ -359,6 +360,12 @@ void ValueMetricProducer::pullAndMatchEventsLocked(const int64_t timestampNs) {
		}
		StatsdStats::getInstance().notePullDelay(mPullTagId, pullDelayNs);

		if (timestampNs < mCurrentBucketStartTimeNs) {
		// The data will be skipped in onMatchedLogEventInternalLocked, but we don't want to report
		// for every event, just the pull
		StatsdStats::getInstance().noteLateLogEventSkipped(mMetricId);
		}

		for (const auto& data : allData) {
		// make a copy before doing and changes
		LogEvent localCopy = data->makeCopy();
		@@ -380,6 +387,7 @@ void ValueMetricProducer::onDataPulled(const std::vector<std::shared_ptr<LogEven
		if (mCondition) {
		if (allData.size() == 0) {
		VLOG("Data pulled is empty");
		StatsdStats::getInstance().noteEmptyData(mPullTagId);
		return;
		}
		// For scheduled pulled data, the effective event time is snap to the nearest
		@@ -394,6 +402,7 @@ void ValueMetricProducer::onDataPulled(const std::vector<std::shared_ptr<LogEven
		if (bucketEndTime < mCurrentBucketStartTimeNs) {
		VLOG("Skip bucket end pull due to late arrival: %lld vs %lld", (long long)bucketEndTime,
		(long long)mCurrentBucketStartTimeNs);
		StatsdStats::getInstance().noteLateLogEventSkipped(mMetricId);
		return;
		}
		for (const auto& data : allData) {
		@@ -442,6 +451,7 @@ bool ValueMetricProducer::hitGuardRailLocked(const MetricDimensionKey& newKey) {
		if (newTupleCount > mDimensionHardLimit) {
		ALOGE("ValueMetric %lld dropping data for dimension key %s", (long long)mMetricId,
		newKey.toString().c_str());
		StatsdStats::getInstance().noteHardDimensionLimitReached(mMetricId);
		return true;
		}
		}
		@@ -539,6 +549,7 @@ void ValueMetricProducer::onMatchedLogEventInternalLocked(const size_t matcherIn
		Value value;
		if (!getDoubleOrLong(event, matcher, value)) {
		VLOG("Failed to get value %d from event %s", i, event.ToString().c_str());
		StatsdStats::getInstance().noteBadValueType(mMetricId);
		return;
		}
		interval.seenNewData = true;
		@@ -656,6 +667,7 @@ void ValueMetricProducer::flushIfNeededLocked(const int64_t& eventTimeNs) {

		if (numBucketsForward > 1) {
		VLOG("Skipping forward %lld buckets", (long long)numBucketsForward);
		StatsdStats::getInstance().noteSkippedForwardBuckets(mMetricId);
		// take base again in future good bucket.
		resetBase();
		}