Coverage for bzfs_main/parallel_batch_cmd.py: 100%

37 statements  

« prev     ^ index     » next       coverage.py v7.11.0, created at 2025-11-07 04:44 +0000

1# Copyright 2024 Wolfgang Hoschek AT mac DOT com 

2# 

3# Licensed under the Apache License, Version 2.0 (the "License"); 

4# you may not use this file except in compliance with the License. 

5# You may obtain a copy of the License at 

6# 

7# http://www.apache.org/licenses/LICENSE-2.0 

8# 

9# Unless required by applicable law or agreed to in writing, software 

10# distributed under the License is distributed on an "AS IS" BASIS, 

11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 

12# See the License for the specific language governing permissions and 

13# limitations under the License. 

14# 

15"""Helpers for running CLI commands in (sequential or parallel) batches, without exceeding operating system limits. 

16 

17The batch size aka max_batch_items splits one CLI command into one or more CLI commands. The resulting commands are executed 

18sequentially (via functions *_batched()), or in parallel across max_workers threads (via functions *_parallel()). 

19 

20The degree of parallelism (max_workers) is specified by the job (via --threads). 

21Batch size is a trade-off between resource consumption, latency, bandwidth and throughput. 

22 

23Example: 

24-------- 

25 

26- max_batch_items=1 (seq or par): 

27``` 

28zfs list -t snapshot d1 

29zfs list -t snapshot d2 

30zfs list -t snapshot d3 

31zfs list -t snapshot d4 

32``` 

33 

34- max_batch_items=2 (seq or par): 

35``` 

36zfs list -t snapshot d1 d2 

37zfs list -t snapshot d3 d4 

38 

39- max_batch_items=N (seq or par): 

40``` 

41zfs list -t snapshot d1 d2 d3 d4 

42``` 

43""" 

44 

45from __future__ import ( 

46 annotations, 

47) 

48import sys 

49from collections.abc import ( 

50 Iterable, 

51 Iterator, 

52) 

53from typing import ( 

54 TYPE_CHECKING, 

55 Any, 

56 Callable, 

57 TypeVar, 

58) 

59 

60from bzfs_main.connection import ( 

61 SHARED, 

62 ConnectionPool, 

63 try_ssh_command, 

64) 

65from bzfs_main.parallel_iterator import ( 

66 batch_cmd_iterator, 

67 get_max_command_line_bytes, 

68 parallel_iterator, 

69) 

70from bzfs_main.utils import ( 

71 LOG_TRACE, 

72 drain, 

73) 

74 

75if TYPE_CHECKING: # pragma: no cover - for type hints only 

76 from bzfs_main.bzfs import ( 

77 Job, 

78 ) 

79 from bzfs_main.configuration import ( 

80 Remote, 

81 ) 

82 

83T = TypeVar("T") 

84 

85 

86def run_ssh_cmd_batched( 

87 job: Job, 

88 r: Remote, 

89 cmd: list[str], 

90 cmd_args: Iterable[str], 

91 fn: Callable[[list[str]], Any], 

92 max_batch_items: int = 2**29, 

93 sep: str = " ", 

94) -> None: 

95 """Runs ssh command for each sequential batch of args, without creating a cmdline that's too big for the OS to handle.""" 

96 drain(itr_ssh_cmd_batched(job, r, cmd, cmd_args, fn, max_batch_items=max_batch_items, sep=sep)) 

97 

98 

99def itr_ssh_cmd_batched( 

100 job: Job, 

101 r: Remote, 

102 cmd: list[str], 

103 cmd_args: Iterable[str], 

104 fn: Callable[[list[str]], T], 

105 max_batch_items: int = 2**29, 

106 sep: str = " ", 

107) -> Iterator[T]: 

108 """Runs fn(cmd_args) in sequential batches w/ cmd, without creating a cmdline that's too big for the OS to handle.""" 

109 max_bytes: int = _max_batch_bytes(job, r, cmd, sep) 

110 return batch_cmd_iterator(cmd_args=cmd_args, fn=fn, max_batch_items=max_batch_items, max_batch_bytes=max_bytes, sep=sep) 

111 

112 

113def run_ssh_cmd_parallel( 

114 job: Job, 

115 r: Remote, 

116 cmd_args_list: Iterable[tuple[list[str], Iterable[str]]], 

117 fn: Callable[[list[str], list[str]], Any], 

118 max_batch_items: int = 2**29, 

119) -> None: 

120 """Runs multiple ssh commands in parallel, batching each set of args.""" 

121 drain(itr_ssh_cmd_parallel(job, r, cmd_args_list, fn=fn, max_batch_items=max_batch_items, ordered=False)) 

122 

123 

124def itr_ssh_cmd_parallel( 

125 job: Job, 

126 r: Remote, 

127 cmd_args_list: Iterable[tuple[list[str], Iterable[str]]], 

128 fn: Callable[[list[str], list[str]], T], 

129 max_batch_items: int = 2**29, 

130 ordered: bool = True, 

131) -> Iterator[T]: 

132 """Streams results from multiple parallel (batched) SSH commands. 

133 

134 When ordered=True, preserves the order of the batches as provided by cmd_args_list (i.e. yields results in the same order 

135 as the input), not in completion order. When ordered=False, yields results as they complete for minimum latency. 

136 """ 

137 return parallel_iterator( 

138 iterator_builder=lambda executr: ( 

139 itr_ssh_cmd_batched( 

140 job, r, cmd, cmd_args, lambda batch, cmd=cmd: executr.submit(fn, cmd, batch), max_batch_items=max_batch_items # type: ignore[misc] 

141 ) 

142 for cmd, cmd_args in cmd_args_list 

143 ), 

144 max_workers=job.max_workers[r.location], 

145 ordered=ordered, 

146 termination_event=job.termination_event, 

147 ) 

148 

149 

150def zfs_list_snapshots_in_parallel( 

151 job: Job, r: Remote, cmd: list[str], datasets: list[str], ordered: bool = True 

152) -> Iterator[list[str]]: 

153 """Runs 'zfs list -t snapshot' on multiple datasets at the same time. 

154 

155 Implemented with a time and space efficient streaming algorithm; easily scales to millions of datasets and any number of 

156 snapshots. Attempts to use at least 4 datasets per remote cmd mini batch to reflect increased communication latency. 

157 """ 

158 max_workers: int = job.max_workers[r.location] 

159 max_batch_items: int = min( 

160 job.max_datasets_per_minibatch_on_list_snaps[r.location], 

161 max( 

162 len(datasets) // (max_workers * 8), 

163 4 if r.ssh_user_host else 1, 

164 ), 

165 ) 

166 return itr_ssh_cmd_parallel( 

167 job, 

168 r, 

169 [(cmd, datasets)], 

170 fn=lambda cmd, batch: (try_ssh_command(job, r, LOG_TRACE, cmd=cmd + batch) or "").splitlines(), 

171 max_batch_items=max_batch_items, 

172 ordered=ordered, 

173 ) 

174 

175 

176def _max_batch_bytes(job: Job, r: Remote, cmd: list[str], sep: str) -> int: 

177 """Avoids creating a cmdline that's too big for the OS to handle. 

178 

179 The calculation subtracts 'header_bytes', which accounts for the full SSH invocation (including control socket/options) 

180 plus the fixed subcommand prefix, so that the remaining budget is reserved exclusively for the batched arguments. 

181 """ 

182 assert isinstance(sep, str) 

183 max_bytes: int = min(_get_max_command_line_bytes(job, "local"), _get_max_command_line_bytes(job, r.location)) 

184 # Max size of a single argument is 128KB on Linux - https://lists.gnu.org/archive/html/bug-bash/2020-09/msg00095.html 

185 max_bytes = max_bytes if sep == " " else min(max_bytes, 128 * 1024 - 1) # e.g. 'zfs destroy foo@s1,s2,...,sN' 

186 conn_pool: ConnectionPool = job.params.connection_pools[r.location].pool(SHARED) 

187 with conn_pool.connection() as conn: 

188 cmd = conn.ssh_cmd + cmd 

189 header_bytes: int = len(" ".join(cmd).encode(sys.getfilesystemencoding())) 

190 return max_bytes - header_bytes 

191 

192 

193def _get_max_command_line_bytes(job: Job, location: str, os_name: str | None = None) -> int: 

194 """Remote flavor of os.sysconf("SC_ARG_MAX") - size(os.environb) - safety margin""" 

195 os_name = os_name if os_name else job.params.available_programs[location].get("os") 

196 os_name = os_name if os_name else "n/a" 

197 max_bytes = get_max_command_line_bytes(os_name) 

198 if job.max_command_line_bytes is not None: 

199 return job.max_command_line_bytes # for testing only 

200 else: 

201 return max_bytes