1 /*
2 * Copyright (C) 2023 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 #include <stats_event.h>
18 #include <stats_socket_loss_reporter.h>
19 #include <unistd.h>
20
21 #include <vector>
22
23 #include "stats_statsdsocketlog.h"
24 #include "utils.h"
25
StatsSocketLossReporter()26 StatsSocketLossReporter::StatsSocketLossReporter() : mUid(getuid()) {
27 }
28
~StatsSocketLossReporter()29 StatsSocketLossReporter::~StatsSocketLossReporter() {
30 // try to dump loss stats since there might be pending data which have been not sent earlier
31 // due to:
32 // - cool down timer was active
33 // - no input atoms to trigger loss info dump after cooldown timer expired
34 dumpAtomsLossStats(true);
35 }
36
getInstance()37 StatsSocketLossReporter& StatsSocketLossReporter::getInstance() {
38 static StatsSocketLossReporter instance;
39 return instance;
40 }
41
noteDrop(int32_t error,int32_t atomId)42 void StatsSocketLossReporter::noteDrop(int32_t error, int32_t atomId) {
43 using namespace android::os::statsdsocket;
44
45 const int64_t currentRealtimeTsNanos = get_elapsed_realtime_ns();
46
47 // The intention is to skip self counting, however the timestamps still need to be updated
48 // to know when was last failed attempt to log atom.
49 // This is required for more accurate cool down timer work
50 if (mFirstTsNanos == 0) {
51 mFirstTsNanos.store(currentRealtimeTsNanos, std::memory_order_relaxed);
52 }
53 mLastTsNanos.store(currentRealtimeTsNanos, std::memory_order_relaxed);
54
55 if (atomId == STATS_SOCKET_LOSS_REPORTED) {
56 // avoid self counting due to write to socket might fail during dumpAtomsLossStats()
57 // also due to mutex is not re-entrant and is already locked by dumpAtomsLossStats() API,
58 // return to avoid deadlock
59 // alternative is to consider std::recursive_mutex
60 return;
61 }
62
63 std::unique_lock<std::mutex> lock(mMutex);
64
65 // using unordered_map is more CPU efficient vs vectors, however will require some
66 // postprocessing before writing into the socket
67 const LossInfoKey key = std::make_pair(error, atomId);
68 auto counterIt = mLossInfo.find(key);
69 if (counterIt != mLossInfo.end()) {
70 ++counterIt->second;
71 } else if (mLossInfo.size() < kMaxAtomTagsCount) {
72 mLossInfo[key] = 1;
73 } else {
74 mOverflowCounter++;
75 }
76 }
77
dumpAtomsLossStats(bool forceDump)78 void StatsSocketLossReporter::dumpAtomsLossStats(bool forceDump) {
79 using namespace android::os::statsdsocket;
80
81 const int64_t currentRealtimeTsNanos = get_elapsed_realtime_ns();
82
83 if (!forceDump && isCooldownTimerActive(currentRealtimeTsNanos)) {
84 // To avoid socket flooding with more STATS_SOCKET_LOSS_REPORTED atoms,
85 // which have high probability of write failures, the cooldown timer approach is applied:
86 // - start cooldown timer for 10us for every failed dump
87 // - before writing STATS_SOCKET_LOSS_REPORTED do check the timestamp to keep some delay
88 return;
89 }
90
91 // intention to hold mutex here during the stats_write() to avoid data copy overhead
92 std::unique_lock<std::mutex> lock(mMutex);
93 if (mLossInfo.size() == 0) {
94 return;
95 }
96
97 // populate temp vectors to be written into the socket
98 std::vector<int> errors(mLossInfo.size());
99 std::vector<int> tags(mLossInfo.size());
100 std::vector<int> counts(mLossInfo.size());
101
102 auto lossInfoIt = mLossInfo.begin();
103 for (size_t i = 0; i < mLossInfo.size(); i++, lossInfoIt++) {
104 const LossInfoKey& key = lossInfoIt->first;
105 errors[i] = key.first;
106 tags[i] = key.second;
107 counts[i] = lossInfoIt->second;
108 }
109
110 // below call might lead to socket loss event - intention is to avoid self counting
111 const int ret = stats_write(STATS_SOCKET_LOSS_REPORTED, mUid, mFirstTsNanos, mLastTsNanos,
112 mOverflowCounter, errors, tags, counts);
113 if (ret > 0) {
114 // Otherwise, in case of failure we preserve all socket loss information between dumps.
115 // When above write failed - the socket loss stats are not discarded
116 // and would be re-send during next attempt.
117 mOverflowCounter = 0;
118 mLossInfo.clear();
119
120 mFirstTsNanos.store(0, std::memory_order_relaxed);
121 mLastTsNanos.store(0, std::memory_order_relaxed);
122 }
123 // since the delay before next attempt is significantly larger than this API call
124 // duration it is ok to have correctness of timestamp in a range of 10us
125 startCooldownTimer(currentRealtimeTsNanos);
126 }
127
startCooldownTimer(int64_t elapsedRealtimeNanos)128 void StatsSocketLossReporter::startCooldownTimer(int64_t elapsedRealtimeNanos) {
129 mCooldownTimerFinishAtNanos = elapsedRealtimeNanos + kCoolDownTimerDurationNanos;
130 }
131
isCooldownTimerActive(int64_t elapsedRealtimeNanos) const132 bool StatsSocketLossReporter::isCooldownTimerActive(int64_t elapsedRealtimeNanos) const {
133 return mCooldownTimerFinishAtNanos > elapsedRealtimeNanos;
134 }
135