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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
|
from __future__ import annotations
import asyncio
import contextlib
import logging
from collections import defaultdict
from collections.abc import Iterator
from typing import TYPE_CHECKING, Any, Generic, Sized, TypeVar
from distributed.metrics import time
from distributed.shuffle._limiter import ResourceLimiter
if TYPE_CHECKING:
import pyarrow as pa
logger = logging.getLogger("distributed.shuffle")
ShardType = TypeVar("ShardType", bound=Sized)
class ShardsBuffer(Generic[ShardType]):
"""A buffer for P2P shuffle
The objects to buffer are typically bytes belonging to certain shards.
Typically the buffer is implemented on sending and receiving end.
The buffer allows for concurrent writing and buffers shards to reduce overhead of writing.
The shards are typically provided in a format like::
{
"bucket-0": [b"shard1", b"shard2"],
"bucket-1": [b"shard1", b"shard2"],
}
Buckets typically correspond to output partitions.
If exceptions occur during writing, the buffer is automatically closed. Subsequent attempts to write will raise the same exception.
Flushing will not raise an exception. To ensure that the buffer finished successfully, please call `ShardsBuffer.raise_on_exception`
"""
shards: defaultdict[str, list[ShardType]]
sizes: defaultdict[str, int]
concurrency_limit: int
memory_limiter: ResourceLimiter | None
diagnostics: dict[str, float]
max_message_size: int
bytes_total: int
bytes_memory: int
bytes_written: int
bytes_read: int
_accepts_input: bool
_inputs_done: bool
_exception: None | Exception
_tasks: list[asyncio.Task]
_shards_available: asyncio.Condition
_flush_lock: asyncio.Lock
def __init__(
self,
memory_limiter: ResourceLimiter | None,
concurrency_limit: int = 2,
max_message_size: int = -1,
) -> None:
self._accepts_input = True
self.shards = defaultdict(list)
self.sizes = defaultdict(int)
self._exception = None
self.concurrency_limit = concurrency_limit
self._inputs_done = False
self.memory_limiter = memory_limiter
self.diagnostics: dict[str, float] = defaultdict(float)
self._tasks = [
asyncio.create_task(self._background_task())
for _ in range(concurrency_limit)
]
self._shards_available = asyncio.Condition()
self._flush_lock = asyncio.Lock()
self.max_message_size = max_message_size
self.bytes_total = 0
self.bytes_memory = 0
self.bytes_written = 0
self.bytes_read = 0
def heartbeat(self) -> dict[str, Any]:
return {
"memory": self.bytes_memory,
"total": self.bytes_total,
"buckets": len(self.shards),
"written": self.bytes_written,
"read": self.bytes_read,
"diagnostics": self.diagnostics,
"memory_limit": self.memory_limiter._maxvalue if self.memory_limiter else 0,
}
async def process(self, id: str, shards: list[pa.Table], size: int) -> None:
try:
start = time()
try:
await self._process(id, shards)
self.bytes_written += size
except Exception as e:
self._exception = e
self._inputs_done = True
stop = time()
self.diagnostics["avg_size"] = (
0.98 * self.diagnostics["avg_size"] + 0.02 * size
)
self.diagnostics["avg_duration"] = 0.98 * self.diagnostics[
"avg_duration"
] + 0.02 * (stop - start)
finally:
if self.memory_limiter:
await self.memory_limiter.decrease(size)
self.bytes_memory -= size
async def _process(self, id: str, shards: list[ShardType]) -> None:
raise NotImplementedError()
def read(self, id: str) -> ShardType:
raise NotImplementedError()
@property
def empty(self) -> bool:
return not self.shards
async def _background_task(self) -> None:
def _continue() -> bool:
return bool(self.shards or self._inputs_done)
while True:
async with self._shards_available:
await self._shards_available.wait_for(_continue)
if self._inputs_done and not self.shards:
break
part_id = max(self.sizes, key=self.sizes.__getitem__)
if self.max_message_size > 0:
size = 0
shards = []
while size < self.max_message_size:
try:
shard = self.shards[part_id].pop()
shards.append(shard)
s = len(shard)
size += s
self.sizes[part_id] -= s
except IndexError:
break
finally:
if not self.shards[part_id]:
del self.shards[part_id]
assert not self.sizes[part_id]
del self.sizes[part_id]
else:
shards = self.shards.pop(part_id)
size = self.sizes.pop(part_id)
self._shards_available.notify_all()
await self.process(part_id, shards, size)
async def write(self, data: dict[str, list[ShardType]]) -> None:
"""
Writes many objects into the local buffers, blocks until ready for more
Parameters
----------
data: dict
A dictionary mapping destinations to lists of objects that should
be written to that destination
"""
if self._exception:
raise self._exception
if not self._accepts_input or self._inputs_done:
raise RuntimeError(f"Trying to put data in closed {self}.")
if not data:
return
shards = None
size = 0
sizes = {}
for id_, shards in data.items():
size = sum(map(len, shards))
sizes[id_] = size
total_batch_size = sum(sizes.values())
self.bytes_memory += total_batch_size
self.bytes_total += total_batch_size
if self.memory_limiter:
self.memory_limiter.increase(total_batch_size)
async with self._shards_available:
for id_, shards in data.items():
self.shards[id_].extend(shards)
self.sizes[id_] += sizes[id_]
self._shards_available.notify()
if self.memory_limiter:
await self.memory_limiter.wait_for_available()
del data, shards
assert size
def raise_on_exception(self) -> None:
"""Raises an exception if something went wrong during writing"""
if self._exception:
raise self._exception
async def flush(self) -> None:
"""Wait until all writes are finished.
This closes the buffer such that no new writes are allowed
"""
async with self._flush_lock:
self._accepts_input = False
async with self._shards_available:
self._shards_available.notify_all()
await self._shards_available.wait_for(
lambda: not self.shards or self._exception or self._inputs_done
)
self._inputs_done = True
self._shards_available.notify_all()
await asyncio.gather(*self._tasks)
if not self._exception:
assert not self.bytes_memory, (type(self), self.bytes_memory)
async def close(self) -> None:
"""Flush and close the buffer.
This cleans up all allocated resources.
"""
await self.flush()
if not self._exception:
assert not self.bytes_memory, (type(self), self.bytes_memory)
for t in self._tasks:
t.cancel()
self._accepts_input = False
self._inputs_done = True
self.shards.clear()
self.bytes_memory = 0
async with self._shards_available:
self._shards_available.notify_all()
await asyncio.gather(*self._tasks)
async def __aenter__(self) -> "ShardsBuffer":
return self
async def __aexit__(self, exc: Any, typ: Any, traceback: Any) -> None:
await self.close()
@contextlib.contextmanager
def time(self, name: str) -> Iterator[None]:
start = time()
yield
stop = time()
self.diagnostics[name] += stop - start
|
