Persist active metric status across system server
Previously, all metrics/configs would deactivate on system server death.
Now, active status is restored.
Bug: 129717537
Test: bit statsd_test:*
Test: libprotoutil_test:ProtoOutputStream*
Change-Id: Idf372457f60a931a2d00176a5eab58c534a25e41
diff --git a/cmds/statsd/src/StatsLogProcessor.cpp b/cmds/statsd/src/StatsLogProcessor.cpp
index ec02b12..4e0a8eb 100644
--- a/cmds/statsd/src/StatsLogProcessor.cpp
+++ b/cmds/statsd/src/StatsLogProcessor.cpp
@@ -611,11 +611,8 @@
void StatsLogProcessor::SaveActiveConfigsToDisk(int64_t currentTimeNs) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
-
const int64_t timeNs = getElapsedRealtimeNs();
// Do not write to disk if we already have in the last few seconds.
- // This is to avoid overwriting files that would have the same name if we
- // write twice in the same second.
if (static_cast<unsigned long long> (timeNs) <
mLastActiveMetricsWriteNs + WRITE_DATA_COOL_DOWN_SEC * NS_PER_SEC) {
ALOGI("Statsd skipping writing active metrics to disk. Already wrote data in last %d seconds",
@@ -625,13 +622,7 @@
mLastActiveMetricsWriteNs = timeNs;
ProtoOutputStream proto;
- for (const auto& pair : mMetricsManagers) {
- const sp<MetricsManager>& metricsManager = pair.second;
- uint64_t configToken = proto.start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED |
- FIELD_ID_ACTIVE_CONFIG_LIST_CONFIG);
- metricsManager->writeActiveConfigToProtoOutputStream(currentTimeNs, &proto);
- proto.end(configToken);
- }
+ WriteActiveConfigsToProtoOutputStreamLocked(currentTimeNs, DEVICE_SHUTDOWN, &proto);
string file_name = StringPrintf("%s/active_metrics", STATS_ACTIVE_METRIC_DIR);
StorageManager::deleteFile(file_name.c_str());
@@ -644,9 +635,24 @@
proto.flush(fd.get());
}
+void StatsLogProcessor::WriteActiveConfigsToProtoOutputStream(
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto) {
+ std::lock_guard<std::mutex> lock(mMetricsMutex);
+ WriteActiveConfigsToProtoOutputStreamLocked(currentTimeNs, reason, proto);
+}
+
+void StatsLogProcessor::WriteActiveConfigsToProtoOutputStreamLocked(
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto) {
+ for (const auto& pair : mMetricsManagers) {
+ const sp<MetricsManager>& metricsManager = pair.second;
+ uint64_t configToken = proto->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED |
+ FIELD_ID_ACTIVE_CONFIG_LIST_CONFIG);
+ metricsManager->writeActiveConfigToProtoOutputStream(currentTimeNs, reason, proto);
+ proto->end(configToken);
+ }
+}
void StatsLogProcessor::LoadActiveConfigsFromDisk() {
std::lock_guard<std::mutex> lock(mMetricsMutex);
-
string file_name = StringPrintf("%s/active_metrics", STATS_ACTIVE_METRIC_DIR);
int fd = open(file_name.c_str(), O_RDONLY | O_CLOEXEC);
if (-1 == fd) {
@@ -670,6 +676,19 @@
StorageManager::deleteFile(file_name.c_str());
return;
}
+ // Passing in mTimeBaseNs only works as long as we only load from disk is when statsd starts.
+ SetConfigsActiveStateLocked(activeConfigList, mTimeBaseNs);
+ StorageManager::deleteFile(file_name.c_str());
+}
+
+void StatsLogProcessor::SetConfigsActiveState(const ActiveConfigList& activeConfigList,
+ int64_t currentTimeNs) {
+ std::lock_guard<std::mutex> lock(mMetricsMutex);
+ SetConfigsActiveStateLocked(activeConfigList, currentTimeNs);
+}
+
+void StatsLogProcessor::SetConfigsActiveStateLocked(const ActiveConfigList& activeConfigList,
+ int64_t currentTimeNs) {
for (int i = 0; i < activeConfigList.config_size(); i++) {
const auto& config = activeConfigList.config(i);
ConfigKey key(config.uid(), config.id());
@@ -679,11 +698,9 @@
continue;
}
VLOG("Setting active config %s", key.ToString().c_str());
- it->second->loadActiveConfig(config, mTimeBaseNs);
+ it->second->loadActiveConfig(config, currentTimeNs);
}
VLOG("Successfully loaded %d active configs.", activeConfigList.config_size());
-
- StorageManager::deleteFile(file_name.c_str());
}
void StatsLogProcessor::WriteDataToDiskLocked(const DumpReportReason dumpReportReason,
diff --git a/cmds/statsd/src/StatsLogProcessor.h b/cmds/statsd/src/StatsLogProcessor.h
index 0dc597b..92aa425 100644
--- a/cmds/statsd/src/StatsLogProcessor.h
+++ b/cmds/statsd/src/StatsLogProcessor.h
@@ -31,17 +31,6 @@
namespace os {
namespace statsd {
-// Keep this in sync with DumpReportReason enum in stats_log.proto
-enum DumpReportReason {
- DEVICE_SHUTDOWN = 1,
- CONFIG_UPDATED = 2,
- CONFIG_REMOVED = 3,
- GET_DATA_CALLED = 4,
- ADB_DUMP = 5,
- CONFIG_RESET = 6,
- STATSCOMPANION_DIED = 7,
- TERMINATION_SIGNAL_RECEIVED = 8
-};
class StatsLogProcessor : public ConfigListener {
public:
@@ -92,9 +81,16 @@
/* Persist configs containing metrics with active activations to disk. */
void SaveActiveConfigsToDisk(int64_t currentTimeNs);
+ /* Writes the current active status/ttl for all configs and metrics to ProtoOutputStream. */
+ void WriteActiveConfigsToProtoOutputStream(
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto);
+
/* Load configs containing metrics with active activations from disk. */
void LoadActiveConfigsFromDisk();
+ /* Sets the active status/ttl for all configs and metrics to the status in ActiveConfigList. */
+ void SetConfigsActiveState(const ActiveConfigList& activeConfigList, int64_t currentTimeNs);
+
// Reset all configs.
void resetConfigs();
@@ -158,6 +154,12 @@
void GetActiveConfigsLocked(const int uid, vector<int64_t>& outActiveConfigs);
+ void WriteActiveConfigsToProtoOutputStreamLocked(
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto);
+
+ void SetConfigsActiveStateLocked(const ActiveConfigList& activeConfigList,
+ int64_t currentTimeNs);
+
void WriteDataToDiskLocked(const DumpReportReason dumpReportReason,
const DumpLatency dumpLatency);
void WriteDataToDiskLocked(const ConfigKey& key, const int64_t timestampNs,
@@ -224,6 +226,7 @@
FRIEND_TEST(StatsLogProcessorTest, TestActivationOnBootMultipleActivations);
FRIEND_TEST(StatsLogProcessorTest,
TestActivationOnBootMultipleActivationsDifferentActivationTypes);
+ FRIEND_TEST(StatsLogProcessorTest, TestActivationsPersistAcrossSystemServerRestart);
FRIEND_TEST(WakelockDurationE2eTest, TestAggregatedPredicateDimensionsForSumDuration1);
FRIEND_TEST(WakelockDurationE2eTest, TestAggregatedPredicateDimensionsForSumDuration2);
diff --git a/cmds/statsd/src/StatsService.cpp b/cmds/statsd/src/StatsService.cpp
index 623a1f2..8191d37 100644
--- a/cmds/statsd/src/StatsService.cpp
+++ b/cmds/statsd/src/StatsService.cpp
@@ -1478,8 +1478,21 @@
StatsdStats::getInstance().noteSystemServerRestart(getWallClockSec());
if (mProcessor != nullptr) {
ALOGW("Reset statsd upon system server restarts.");
+ int64_t systemServerRestartNs = getElapsedRealtimeNs();
+ ProtoOutputStream proto;
+ mProcessor->WriteActiveConfigsToProtoOutputStream(systemServerRestartNs,
+ STATSCOMPANION_DIED, &proto);
+
mProcessor->WriteDataToDisk(STATSCOMPANION_DIED, FAST);
mProcessor->resetConfigs();
+
+ std::string serializedActiveConfigs;
+ if (proto.serializeToString(&serializedActiveConfigs)) {
+ ActiveConfigList activeConfigs;
+ if (activeConfigs.ParseFromString(serializedActiveConfigs)) {
+ mProcessor->SetConfigsActiveState(activeConfigs, systemServerRestartNs);
+ }
+ }
}
mAnomalyAlarmMonitor->setStatsCompanionService(nullptr);
mPeriodicAlarmMonitor->setStatsCompanionService(nullptr);
diff --git a/cmds/statsd/src/StatsService.h b/cmds/statsd/src/StatsService.h
index 936f7db..a4e6d7f 100644
--- a/cmds/statsd/src/StatsService.h
+++ b/cmds/statsd/src/StatsService.h
@@ -427,6 +427,7 @@
std::shared_ptr<LogEventQueue> mEventQueue;
+ FRIEND_TEST(StatsLogProcessorTest, TestActivationsPersistAcrossSystemServerRestart);
FRIEND_TEST(StatsServiceTest, TestAddConfig_simple);
FRIEND_TEST(StatsServiceTest, TestAddConfig_empty);
FRIEND_TEST(StatsServiceTest, TestAddConfig_invalid);
diff --git a/cmds/statsd/src/active_config_list.proto b/cmds/statsd/src/active_config_list.proto
index ef8e50b..9929833 100644
--- a/cmds/statsd/src/active_config_list.proto
+++ b/cmds/statsd/src/active_config_list.proto
@@ -26,7 +26,18 @@
// Time left in activation. When this proto is loaded after device boot,
// the activation should be set to active for this duration.
+ // This field will only be set when the state is ACTIVE
optional int64 remaining_ttl_nanos = 2;
+
+ enum State {
+ UNNKNOWN = 0;
+ // This metric should activate for remaining_ttl_nanos when we load the activations.
+ ACTIVE = 1;
+ // When we load the activations, this metric should activate on next boot for the tll
+ // specified in the config.
+ ACTIVATE_ON_BOOT = 2;
+ }
+ optional State state = 3;
}
message ActiveMetric {
diff --git a/cmds/statsd/src/metrics/MetricProducer.cpp b/cmds/statsd/src/metrics/MetricProducer.cpp
index 9ad7f09..d913427 100644
--- a/cmds/statsd/src/metrics/MetricProducer.cpp
+++ b/cmds/statsd/src/metrics/MetricProducer.cpp
@@ -19,6 +19,7 @@
#include "MetricProducer.h"
using android::util::FIELD_COUNT_REPEATED;
+using android::util::FIELD_TYPE_ENUM;
using android::util::FIELD_TYPE_INT32;
using android::util::FIELD_TYPE_INT64;
using android::util::FIELD_TYPE_MESSAGE;
@@ -37,6 +38,7 @@
// for ActiveEventActivation
const int FIELD_ID_ACTIVE_EVENT_ACTIVATION_ATOM_MATCHER_INDEX = 1;
const int FIELD_ID_ACTIVE_EVENT_ACTIVATION_REMAINING_TTL_NANOS = 2;
+const int FIELD_ID_ACTIVE_EVENT_ACTIVATION_STATE = 3;
void MetricProducer::onMatchedLogEventLocked(const size_t matcherIndex, const LogEvent& event) {
if (!mIsActive) {
@@ -165,17 +167,21 @@
continue;
}
auto& activation = it->second;
- // We don't want to change the ttl for future activations, so we set the start_ns
- // such that start_ns + ttl_ns == currentTimeNs + remaining_ttl_nanos
- activation->start_ns =
- currentTimeNs + activeEventActivation.remaining_ttl_nanos() - activation->ttl_ns;
- activation->state = ActivationState::kActive;
- mIsActive = true;
+ if (activeEventActivation.state() == ActiveEventActivation::ACTIVE) {
+ // We don't want to change the ttl for future activations, so we set the start_ns
+ // such that start_ns + ttl_ns == currentTimeNs + remaining_ttl_nanos
+ activation->start_ns =
+ currentTimeNs + activeEventActivation.remaining_ttl_nanos() - activation->ttl_ns;
+ activation->state = ActivationState::kActive;
+ mIsActive = true;
+ } else if (activeEventActivation.state() == ActiveEventActivation::ACTIVATE_ON_BOOT) {
+ activation->state = ActivationState::kActiveOnBoot;
+ }
}
}
void MetricProducer::writeActiveMetricToProtoOutputStream(
- int64_t currentTimeNs, ProtoOutputStream* proto) {
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto) {
proto->write(FIELD_TYPE_INT64 | FIELD_ID_ACTIVE_METRIC_ID, (long long)mMetricId);
for (auto& it : mEventActivationMap) {
const int atom_matcher_index = it.first;
@@ -196,9 +202,22 @@
activation->start_ns + activation->ttl_ns - currentTimeNs;
proto->write(FIELD_TYPE_INT64 | FIELD_ID_ACTIVE_EVENT_ACTIVATION_REMAINING_TTL_NANOS,
(long long)remainingTtlNs);
+ proto->write(FIELD_TYPE_ENUM | FIELD_ID_ACTIVE_EVENT_ACTIVATION_STATE,
+ ActiveEventActivation::ACTIVE);
+
} else if (ActivationState::kActiveOnBoot == activation->state) {
- proto->write(FIELD_TYPE_INT64 | FIELD_ID_ACTIVE_EVENT_ACTIVATION_REMAINING_TTL_NANOS,
- (long long)activation->ttl_ns);
+ if (reason == DEVICE_SHUTDOWN || reason == TERMINATION_SIGNAL_RECEIVED) {
+ proto->write(
+ FIELD_TYPE_INT64 | FIELD_ID_ACTIVE_EVENT_ACTIVATION_REMAINING_TTL_NANOS,
+ (long long)activation->ttl_ns);
+ proto->write(FIELD_TYPE_ENUM | FIELD_ID_ACTIVE_EVENT_ACTIVATION_STATE,
+ ActiveEventActivation::ACTIVE);
+ } else if (reason == STATSCOMPANION_DIED) {
+ // We are saving because of system server death, not due to a device shutdown.
+ // Next time we load, we do not want to activate metrics that activate on boot.
+ proto->write(FIELD_TYPE_ENUM | FIELD_ID_ACTIVE_EVENT_ACTIVATION_STATE,
+ ActiveEventActivation::ACTIVATE_ON_BOOT);
+ }
}
proto->end(activationToken);
}
diff --git a/cmds/statsd/src/metrics/MetricProducer.h b/cmds/statsd/src/metrics/MetricProducer.h
index ec3484c..3ddbef4 100644
--- a/cmds/statsd/src/metrics/MetricProducer.h
+++ b/cmds/statsd/src/metrics/MetricProducer.h
@@ -35,6 +35,18 @@
namespace os {
namespace statsd {
+// Keep this in sync with DumpReportReason enum in stats_log.proto
+enum DumpReportReason {
+ DEVICE_SHUTDOWN = 1,
+ CONFIG_UPDATED = 2,
+ CONFIG_REMOVED = 3,
+ GET_DATA_CALLED = 4,
+ ADB_DUMP = 5,
+ CONFIG_RESET = 6,
+ STATSCOMPANION_DIED = 7,
+ TERMINATION_SIGNAL_RECEIVED = 8
+};
+
// If the metric has no activation requirement, it will be active once the metric producer is
// created.
// If the metric needs to be activated by atoms, the metric producer will start
@@ -244,7 +256,7 @@
void flushIfExpire(int64_t elapsedTimestampNs);
void writeActiveMetricToProtoOutputStream(
- int64_t currentTimeNs, ProtoOutputStream* proto);
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto);
protected:
virtual void onConditionChangedLocked(const bool condition, const int64_t eventTime) = 0;
virtual void onSlicedConditionMayChangeLocked(bool overallCondition,
@@ -268,8 +280,6 @@
return mIsActive;
}
- void prepActiveForBootIfNecessaryLocked(int64_t currentTimeNs);
-
void loadActiveMetricLocked(const ActiveMetric& activeMetric, int64_t currentTimeNs);
virtual void prepareFirstBucketLocked() {};
@@ -412,6 +422,7 @@
FRIEND_TEST(StatsLogProcessorTest, TestActivationOnBootMultipleActivations);
FRIEND_TEST(StatsLogProcessorTest,
TestActivationOnBootMultipleActivationsDifferentActivationTypes);
+ FRIEND_TEST(StatsLogProcessorTest, TestActivationsPersistAcrossSystemServerRestart);
};
} // namespace statsd
diff --git a/cmds/statsd/src/metrics/MetricsManager.cpp b/cmds/statsd/src/metrics/MetricsManager.cpp
index 947f377..207a7dd 100644
--- a/cmds/statsd/src/metrics/MetricsManager.cpp
+++ b/cmds/statsd/src/metrics/MetricsManager.cpp
@@ -528,14 +528,14 @@
}
void MetricsManager::writeActiveConfigToProtoOutputStream(
- int64_t currentTimeNs, ProtoOutputStream* proto) {
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto) {
proto->write(FIELD_TYPE_INT64 | FIELD_ID_ACTIVE_CONFIG_ID, (long long)mConfigKey.GetId());
proto->write(FIELD_TYPE_INT32 | FIELD_ID_ACTIVE_CONFIG_UID, mConfigKey.GetUid());
for (int metricIndex : mMetricIndexesWithActivation) {
const auto& metric = mAllMetricProducers[metricIndex];
const uint64_t metricToken = proto->start(FIELD_TYPE_MESSAGE | FIELD_COUNT_REPEATED |
FIELD_ID_ACTIVE_CONFIG_METRIC);
- metric->writeActiveMetricToProtoOutputStream(currentTimeNs, proto);
+ metric->writeActiveMetricToProtoOutputStream(currentTimeNs, reason, proto);
proto->end(metricToken);
}
}
diff --git a/cmds/statsd/src/metrics/MetricsManager.h b/cmds/statsd/src/metrics/MetricsManager.h
index 818131e..da3be06 100644
--- a/cmds/statsd/src/metrics/MetricsManager.h
+++ b/cmds/statsd/src/metrics/MetricsManager.h
@@ -141,7 +141,7 @@
void loadActiveConfig(const ActiveConfig& config, int64_t currentTimeNs);
void writeActiveConfigToProtoOutputStream(
- int64_t currentTimeNs, ProtoOutputStream* proto);
+ int64_t currentTimeNs, const DumpReportReason reason, ProtoOutputStream* proto);
private:
// For test only.
@@ -290,6 +290,7 @@
FRIEND_TEST(StatsLogProcessorTest, TestActivationOnBootMultipleActivations);
FRIEND_TEST(StatsLogProcessorTest,
TestActivationOnBootMultipleActivationsDifferentActivationTypes);
+ FRIEND_TEST(StatsLogProcessorTest, TestActivationsPersistAcrossSystemServerRestart);
};
} // namespace statsd