Thread-safety at log processor level.

Test: statsd unit test passed.

Change-Id: Ibe8c8d3cc8297875b16ee385c077b71c87353147
This commit is contained in:
Yangster-mac
2018-01-05 15:44:07 -08:00
parent 20ac944ea0
commit b0d0628a29
3 changed files with 21 additions and 20 deletions

View File

@@ -82,9 +82,7 @@ StatsLogProcessor::~StatsLogProcessor() {
void StatsLogProcessor::onAnomalyAlarmFired(
const uint64_t timestampNs,
unordered_set<sp<const AnomalyAlarm>, SpHash<AnomalyAlarm>> anomalySet) {
// TODO: This is a thread-safety issue. mMetricsManagers could change under our feet.
// TODO: Solution? Lock everything! :(
// TODO: Question: Can we replace the other lock (broadcast), or do we need to supplement it?
std::lock_guard<std::mutex> lock(mMetricsMutex);
for (const auto& itr : mMetricsManagers) {
itr.second->onAnomalyAlarmFired(timestampNs, anomalySet);
}
@@ -92,11 +90,13 @@ void StatsLogProcessor::onAnomalyAlarmFired(
// TODO: what if statsd service restarts? How do we know what logs are already processed before?
void StatsLogProcessor::OnLogEvent(const LogEvent& msg) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
StatsdStats::getInstance().noteAtomLogged(msg.GetTagId(), msg.GetTimestampNs() / NS_PER_SEC);
// pass the event to metrics managers.
for (auto& pair : mMetricsManagers) {
pair.second->onLogEvent(msg);
flushIfNecessary(msg.GetTimestampNs(), pair.first, *(pair.second));
flushIfNecessaryLocked(msg.GetTimestampNs(), pair.first, *(pair.second));
}
// Hard-coded logic to update the isolated uid's in the uid-map.
// The field numbers need to be currently updated by hand with atoms.proto
@@ -116,6 +116,7 @@ void StatsLogProcessor::OnLogEvent(const LogEvent& msg) {
}
void StatsLogProcessor::OnConfigUpdated(const ConfigKey& key, const StatsdConfig& config) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
ALOGD("Updated configuration for key %s", key.ToString().c_str());
sp<MetricsManager> newMetricsManager = new MetricsManager(key, config, mTimeBaseSec, mUidMap);
auto it = mMetricsManagers.find(key);
@@ -142,6 +143,7 @@ void StatsLogProcessor::OnConfigUpdated(const ConfigKey& key, const StatsdConfig
}
size_t StatsLogProcessor::GetMetricsSize(const ConfigKey& key) const {
std::lock_guard<std::mutex> lock(mMetricsMutex);
auto it = mMetricsManagers.find(key);
if (it == mMetricsManagers.end()) {
ALOGW("Config source %s does not exist", key.ToString().c_str());
@@ -152,6 +154,7 @@ size_t StatsLogProcessor::GetMetricsSize(const ConfigKey& key) const {
void StatsLogProcessor::onDumpReport(const ConfigKey& key, const uint64_t& dumpTimeStampNs,
ConfigMetricsReportList* report) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
auto it = mMetricsManagers.find(key);
if (it == mMetricsManagers.end()) {
ALOGW("Config source %s does not exist", key.ToString().c_str());
@@ -165,6 +168,7 @@ void StatsLogProcessor::onDumpReport(const ConfigKey& key, const uint64_t& dumpT
}
void StatsLogProcessor::onDumpReport(const ConfigKey& key, vector<uint8_t>* outData) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
auto it = mMetricsManagers.find(key);
if (it == mMetricsManagers.end()) {
ALOGW("Config source %s does not exist", key.ToString().c_str());
@@ -173,9 +177,7 @@ void StatsLogProcessor::onDumpReport(const ConfigKey& key, vector<uint8_t>* outD
// This allows another broadcast to be sent within the rate-limit period if we get close to
// filling the buffer again soon.
mBroadcastTimesMutex.lock();
mLastBroadcastTimes.erase(key);
mBroadcastTimesMutex.unlock();
ProtoOutputStream proto;
@@ -224,6 +226,7 @@ void StatsLogProcessor::onDumpReport(const ConfigKey& key, vector<uint8_t>* outD
}
void StatsLogProcessor::OnConfigRemoved(const ConfigKey& key) {
std::lock_guard<std::mutex> lock(mMetricsMutex);
auto it = mMetricsManagers.find(key);
if (it != mMetricsManagers.end()) {
mMetricsManagers.erase(it);
@@ -231,14 +234,11 @@ void StatsLogProcessor::OnConfigRemoved(const ConfigKey& key) {
}
StatsdStats::getInstance().noteConfigRemoved(key);
std::lock_guard<std::mutex> lock(mBroadcastTimesMutex);
mLastBroadcastTimes.erase(key);
}
void StatsLogProcessor::flushIfNecessary(uint64_t timestampNs, const ConfigKey& key,
MetricsManager& metricsManager) {
std::lock_guard<std::mutex> lock(mBroadcastTimesMutex);
void StatsLogProcessor::flushIfNecessaryLocked(
uint64_t timestampNs, const ConfigKey& key, MetricsManager& metricsManager) {
auto lastCheckTime = mLastByteSizeTimes.find(key);
if (lastCheckTime != mLastByteSizeTimes.end()) {
if (timestampNs - lastCheckTime->second < StatsdStats::kMinByteSizeCheckPeriodNs) {
@@ -274,6 +274,7 @@ void StatsLogProcessor::flushIfNecessary(uint64_t timestampNs, const ConfigKey&
void StatsLogProcessor::WriteDataToDisk() {
mkdir(STATS_DATA_DIR, S_IRWXU);
std::lock_guard<std::mutex> lock(mMetricsMutex);
for (auto& pair : mMetricsManagers) {
const ConfigKey& key = pair.first;
vector<uint8_t> data;

View File

@@ -57,7 +57,7 @@ public:
void WriteDataToDisk();
private:
mutable mutex mBroadcastTimesMutex;
mutable mutex mMetricsMutex;
std::unordered_map<ConfigKey, sp<MetricsManager>> mMetricsManagers;
@@ -72,8 +72,8 @@ private:
/* Check if we should send a broadcast if approaching memory limits and if we're over, we
* actually delete the data. */
void flushIfNecessary(uint64_t timestampNs, const ConfigKey& key,
MetricsManager& metricsManager);
void flushIfNecessaryLocked(uint64_t timestampNs, const ConfigKey& key,
MetricsManager& metricsManager);
// Function used to send a broadcast so that receiver for the config key can call getData
// to retrieve the stored data.

View File

@@ -60,9 +60,9 @@ TEST(StatsLogProcessorTest, TestRateLimitByteSize) {
// Expect only the first flush to trigger a check for byte size since the last two are
// rate-limited.
EXPECT_CALL(mockMetricsManager, byteSize()).Times(1);
p.flushIfNecessary(99, key, mockMetricsManager);
p.flushIfNecessary(100, key, mockMetricsManager);
p.flushIfNecessary(101, key, mockMetricsManager);
p.flushIfNecessaryLocked(99, key, mockMetricsManager);
p.flushIfNecessaryLocked(100, key, mockMetricsManager);
p.flushIfNecessaryLocked(101, key, mockMetricsManager);
}
TEST(StatsLogProcessorTest, TestRateLimitBroadcast) {
@@ -80,12 +80,12 @@ TEST(StatsLogProcessorTest, TestRateLimitBroadcast) {
.WillRepeatedly(Return(int(StatsdStats::kMaxMetricsBytesPerConfig * .95)));
// Expect only one broadcast despite always returning a size that should trigger broadcast.
p.flushIfNecessary(1, key, mockMetricsManager);
p.flushIfNecessaryLocked(1, key, mockMetricsManager);
EXPECT_EQ(1, broadcastCount);
// This next call to flush should not trigger a broadcast.
p.mLastByteSizeTimes.clear(); // Force another check for byte size.
p.flushIfNecessary(2, key, mockMetricsManager);
p.flushIfNecessaryLocked(2, key, mockMetricsManager);
EXPECT_EQ(1, broadcastCount);
}
@@ -106,7 +106,7 @@ TEST(StatsLogProcessorTest, TestDropWhenByteSizeTooLarge) {
EXPECT_CALL(mockMetricsManager, onDumpReport(_)).Times(1);
// Expect to call the onDumpReport and skip the broadcast.
p.flushIfNecessary(1, key, mockMetricsManager);
p.flushIfNecessaryLocked(1, key, mockMetricsManager);
EXPECT_EQ(0, broadcastCount);
}