1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
|
#include "clickhouse_config.h"
#if USE_AWS_S3
#include <IO/WriteBufferFromS3TaskTracker.h>
namespace ProfileEvents
{
extern const Event WriteBufferFromS3WaitInflightLimitMicroseconds;
}
namespace DB
{
WriteBufferFromS3::TaskTracker::TaskTracker(ThreadPoolCallbackRunner<void> scheduler_, size_t max_tasks_inflight_, LogSeriesLimiterPtr limitedLog_)
: is_async(bool(scheduler_))
, scheduler(scheduler_ ? std::move(scheduler_) : syncRunner())
, max_tasks_inflight(max_tasks_inflight_)
, limitedLog(limitedLog_)
{}
WriteBufferFromS3::TaskTracker::~TaskTracker()
{
safeWaitAll();
}
ThreadPoolCallbackRunner<void> WriteBufferFromS3::TaskTracker::syncRunner()
{
return [](Callback && callback, int64_t) mutable -> std::future<void>
{
auto package = std::packaged_task<void()>(std::move(callback));
/// No exceptions are propagated, exceptions are packed to future
package();
return package.get_future();
};
}
void WriteBufferFromS3::TaskTracker::waitAll()
{
/// Exceptions are propagated
for (auto & future : futures)
{
future.get();
}
futures.clear();
std::lock_guard lock(mutex);
finished_futures.clear();
}
void WriteBufferFromS3::TaskTracker::safeWaitAll()
{
for (auto & future : futures)
{
if (future.valid())
{
try
{
/// Exceptions are not propagated
future.get();
} catch (...)
{
/// But at least they are printed
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
}
futures.clear();
std::lock_guard lock(mutex);
finished_futures.clear();
}
void WriteBufferFromS3::TaskTracker::waitIfAny()
{
if (futures.empty())
return;
Stopwatch watch;
{
std::lock_guard lock(mutex);
for (auto & it : finished_futures)
{
/// actually that call might lock this thread until the future is set finally
/// however that won't lock us for long, the task is about to finish when the pointer appears in the `finished_futures`
it->get();
/// in case of exception in `it->get()`
/// it it not necessary to remove `it` from list `futures`
/// `TaskTracker` has to be destroyed after any exception occurs, for this `safeWaitAll` is called.
/// `safeWaitAll` handles invalid futures in the list `futures`
futures.erase(it);
}
finished_futures.clear();
}
watch.stop();
ProfileEvents::increment(ProfileEvents::WriteBufferFromS3WaitInflightLimitMicroseconds, watch.elapsedMicroseconds());
}
void WriteBufferFromS3::TaskTracker::add(Callback && func)
{
/// All this fuzz is about 2 things. This is the most critical place of TaskTracker.
/// The first is not to fail insertion in the list `futures`.
/// In order to face it, the element is allocated at the end of the list `futures` in advance.
/// The second is not to fail the notification of the task.
/// In order to face it, the list element, which would be inserted to the list `finished_futures`,
/// is allocated in advance as an other list `pre_allocated_finished` with one element inside.
/// preallocation for the first issue
futures.emplace_back();
auto future_placeholder = std::prev(futures.end());
/// preallocation for the second issue
FinishedList pre_allocated_finished {future_placeholder};
Callback func_with_notification = [&, my_func = std::move(func), my_pre_allocated_finished = std::move(pre_allocated_finished)]() mutable
{
SCOPE_EXIT({
DENY_ALLOCATIONS_IN_SCOPE;
std::lock_guard lock(mutex);
finished_futures.splice(finished_futures.end(), my_pre_allocated_finished);
has_finished.notify_one();
});
my_func();
};
/// this move is nothrow
*future_placeholder = scheduler(std::move(func_with_notification), Priority{});
waitTilInflightShrink();
}
void WriteBufferFromS3::TaskTracker::waitTilInflightShrink()
{
if (!max_tasks_inflight)
return;
if (futures.size() >= max_tasks_inflight)
LOG_TEST(limitedLog, "have to wait some tasks finish, in queue {}, limit {}", futures.size(), max_tasks_inflight);
Stopwatch watch;
/// Alternative approach is to wait until at least futures.size() - max_tasks_inflight element are finished
/// However the faster finished task is collected the faster CH checks if there is an exception
/// The faster an exception is propagated the lesser time is spent for cancellation
while (futures.size() >= max_tasks_inflight)
{
std::unique_lock lock(mutex);
has_finished.wait(lock, [this] () TSA_REQUIRES(mutex) { return !finished_futures.empty(); });
for (auto & it : finished_futures)
{
it->get();
futures.erase(it);
}
finished_futures.clear();
}
watch.stop();
ProfileEvents::increment(ProfileEvents::WriteBufferFromS3WaitInflightLimitMicroseconds, watch.elapsedMicroseconds());
}
bool WriteBufferFromS3::TaskTracker::isAsync() const
{
return is_async;
}
}
#endif
|
