| 1 | #!/usr/bin/env python3 |
| 2 | # group: rw |
| 3 | # |
| 4 | # Tests for active mirroring |
| 5 | # |
| 6 | # Copyright (C) 2018 Red Hat, Inc. |
| 7 | # |
| 8 | # This program is free software; you can redistribute it and/or modify |
| 9 | # it under the terms of the GNU General Public License as published by |
| 10 | # the Free Software Foundation; either version 2 of the License, or |
| 11 | # (at your option) any later version. |
| 12 | # |
| 13 | # This program is distributed in the hope that it will be useful, |
| 14 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | # GNU General Public License for more details. |
| 17 | # |
| 18 | # You should have received a copy of the GNU General Public License |
| 19 | # along with this program. If not, see <http://www.gnu.org/licenses/>. |
| 20 | # |
| 21 | |
| 22 | import math |
| 23 | import os |
| 24 | import subprocess |
| 25 | import time |
| 26 | from typing import List, Optional |
| 27 | import iotests |
| 28 | from iotests import qemu_img |
| 29 | |
| 30 | source_img = os.path.join(iotests.test_dir, 'source.' + iotests.imgfmt) |
| 31 | target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) |
| 32 | |
| 33 | class TestActiveMirror(iotests.QMPTestCase): |
| 34 | image_len = 128 * 1024 * 1024 # MB |
| 35 | potential_writes_in_flight = True |
| 36 | |
| 37 | def setUp(self): |
| 38 | qemu_img('create', '-f', iotests.imgfmt, source_img, '128M') |
| 39 | qemu_img('create', '-f', iotests.imgfmt, target_img, '128M') |
| 40 | |
| 41 | blk_source = {'id': 'source', |
| 42 | 'if': 'none', |
| 43 | 'node-name': 'source-node', |
| 44 | 'driver': iotests.imgfmt, |
| 45 | 'file': {'driver': 'blkdebug', |
| 46 | 'image': {'driver': 'file', |
| 47 | 'filename': source_img}}} |
| 48 | |
| 49 | blk_target = {'node-name': 'target-node', |
| 50 | 'driver': iotests.imgfmt, |
| 51 | 'file': {'driver': 'file', |
| 52 | 'filename': target_img}} |
| 53 | |
| 54 | self.vm = iotests.VM() |
| 55 | self.vm.add_drive_raw(self.vm.qmp_to_opts(blk_source)) |
| 56 | self.vm.add_blockdev(self.vm.qmp_to_opts(blk_target)) |
| 57 | self.vm.add_device('virtio-blk,id=vblk,drive=source') |
| 58 | self.vm.launch() |
| 59 | |
| 60 | def tearDown(self): |
| 61 | self.vm.shutdown() |
| 62 | |
| 63 | if not self.potential_writes_in_flight: |
| 64 | self.assertTrue(iotests.compare_images(source_img, target_img), |
| 65 | 'mirror target does not match source') |
| 66 | |
| 67 | os.remove(source_img) |
| 68 | os.remove(target_img) |
| 69 | |
| 70 | def doActiveIO(self, sync_source_and_target): |
| 71 | # Fill the source image |
| 72 | self.vm.hmp_qemu_io('source', |
| 73 | 'write -P 1 0 %i' % self.image_len); |
| 74 | |
| 75 | # Start some background requests |
| 76 | for offset in range(1 * self.image_len // 8, 3 * self.image_len // 8, 1024 * 1024): |
| 77 | self.vm.hmp_qemu_io('source', 'aio_write -P 2 %i 1M' % offset) |
| 78 | for offset in range(2 * self.image_len // 8, 3 * self.image_len // 8, 1024 * 1024): |
| 79 | self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) |
| 80 | |
| 81 | # Start the block job |
| 82 | self.vm.cmd('blockdev-mirror', |
| 83 | job_id='mirror', |
| 84 | filter_node_name='mirror-node', |
| 85 | device='source-node', |
| 86 | target='target-node', |
| 87 | sync='full', |
| 88 | copy_mode='write-blocking') |
| 89 | |
| 90 | # Start some more requests |
| 91 | for offset in range(3 * self.image_len // 8, 5 * self.image_len // 8, 1024 * 1024): |
| 92 | self.vm.hmp_qemu_io('source', 'aio_write -P 3 %i 1M' % offset) |
| 93 | for offset in range(4 * self.image_len // 8, 5 * self.image_len // 8, 1024 * 1024): |
| 94 | self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) |
| 95 | |
| 96 | # Wait for the READY event |
| 97 | self.wait_ready(drive='mirror') |
| 98 | |
| 99 | # Now start some final requests; all of these (which land on |
| 100 | # the source) should be settled using the active mechanism. |
| 101 | # The mirror code itself asserts that the source BDS's dirty |
| 102 | # bitmap will stay clean between READY and COMPLETED. |
| 103 | for offset in range(5 * self.image_len // 8, 7 * self.image_len // 8, 1024 * 1024): |
| 104 | self.vm.hmp_qemu_io('source', 'aio_write -P 3 %i 1M' % offset) |
| 105 | for offset in range(6 * self.image_len // 8, 7 * self.image_len // 8, 1024 * 1024): |
| 106 | self.vm.hmp_qemu_io('source', 'aio_write -z %i 1M' % offset) |
| 107 | |
| 108 | if sync_source_and_target: |
| 109 | # If source and target should be in sync after the mirror, |
| 110 | # we have to flush before completion |
| 111 | self.vm.hmp_qemu_io('source', 'aio_flush') |
| 112 | self.potential_writes_in_flight = False |
| 113 | |
| 114 | self.complete_and_wait(drive='mirror', wait_ready=False) |
| 115 | |
| 116 | def testActiveIO(self): |
| 117 | self.doActiveIO(False) |
| 118 | |
| 119 | def testActiveIOFlushed(self): |
| 120 | self.doActiveIO(True) |
| 121 | |
| 122 | def testUnalignedActiveIO(self): |
| 123 | # Fill the source image |
| 124 | result = self.vm.hmp_qemu_io('source', 'write -P 1 0 2M') |
| 125 | |
| 126 | # Start the block job (very slowly) |
| 127 | self.vm.cmd('blockdev-mirror', |
| 128 | job_id='mirror', |
| 129 | filter_node_name='mirror-node', |
| 130 | device='source-node', |
| 131 | target='target-node', |
| 132 | sync='full', |
| 133 | copy_mode='write-blocking', |
| 134 | buf_size=(1048576 // 4), |
| 135 | speed=1) |
| 136 | |
| 137 | # Start an unaligned request to a dirty area |
| 138 | result = self.vm.hmp_qemu_io('source', 'write -P 2 %i 1' % (1048576 + 42)) |
| 139 | |
| 140 | # Let the job finish |
| 141 | self.vm.cmd('block-job-set-speed', device='mirror', speed=0) |
| 142 | self.complete_and_wait(drive='mirror') |
| 143 | |
| 144 | self.potential_writes_in_flight = False |
| 145 | |
| 146 | def testIntersectingActiveIO(self): |
| 147 | # Fill the source image |
| 148 | result = self.vm.hmp_qemu_io('source', 'write -P 1 0 2M') |
| 149 | |
| 150 | # Start the block job (very slowly) |
| 151 | self.vm.cmd('blockdev-mirror', |
| 152 | job_id='mirror', |
| 153 | filter_node_name='mirror-node', |
| 154 | device='source-node', |
| 155 | target='target-node', |
| 156 | sync='full', |
| 157 | copy_mode='write-blocking', |
| 158 | speed=1) |
| 159 | |
| 160 | self.vm.hmp_qemu_io('source', 'break write_aio A') |
| 161 | self.vm.hmp_qemu_io('source', 'aio_write 0 1M') # 1 |
| 162 | self.vm.hmp_qemu_io('source', 'wait_break A') |
| 163 | self.vm.hmp_qemu_io('source', 'aio_write 0 2M') # 2 |
| 164 | self.vm.hmp_qemu_io('source', 'aio_write 0 2M') # 3 |
| 165 | |
| 166 | # Now 2 and 3 are in mirror_wait_on_conflicts, waiting for 1 |
| 167 | |
| 168 | self.vm.hmp_qemu_io('source', 'break write_aio B') |
| 169 | self.vm.hmp_qemu_io('source', 'aio_write 1M 2M') # 4 |
| 170 | self.vm.hmp_qemu_io('source', 'wait_break B') |
| 171 | |
| 172 | # 4 doesn't wait for 2 and 3, because they didn't yet set |
| 173 | # in_flight_bitmap. So, nothing prevents 4 to go except for our |
| 174 | # break-point B. |
| 175 | |
| 176 | self.vm.hmp_qemu_io('source', 'resume A') |
| 177 | |
| 178 | # Now we resumed 1, so 2 and 3 goes to the next iteration of while loop |
| 179 | # in mirror_wait_on_conflicts(). They don't exit, as bitmap is dirty |
| 180 | # due to request 4. |
| 181 | # In the past at that point 2 and 3 would wait for each other producing |
| 182 | # a dead-lock. Now this is fixed and they will wait for request 4. |
| 183 | |
| 184 | self.vm.hmp_qemu_io('source', 'resume B') |
| 185 | |
| 186 | # After resuming 4, one of 2 and 3 goes first and set in_flight_bitmap, |
| 187 | # so the other will wait for it. |
| 188 | |
| 189 | self.vm.cmd('block-job-set-speed', device='mirror', speed=0) |
| 190 | self.complete_and_wait(drive='mirror') |
| 191 | |
| 192 | self.potential_writes_in_flight = False |
| 193 | |
| 194 | def testUnalignedSmallerThanGranularityWriteZeroes(self): |
| 195 | # Fill the source image |
| 196 | self.vm.hmp_qemu_io('source', 'write -P 1 0 %i' % self.image_len); |
| 197 | |
| 198 | # Start the block job |
| 199 | self.vm.cmd('blockdev-mirror', |
| 200 | job_id='mirror', |
| 201 | filter_node_name='mirror-node', |
| 202 | device='source-node', |
| 203 | target='target-node', |
| 204 | sync='full', |
| 205 | copy_mode='write-blocking') |
| 206 | |
| 207 | # Wait for the READY event |
| 208 | self.wait_ready(drive='mirror') |
| 209 | |
| 210 | for offset in range(6 * self.image_len // 8, 7 * self.image_len // 8, 1024 * 1024): |
| 211 | self.vm.hmp_qemu_io('source', 'aio_write -z %i 512' % (offset + 512)) |
| 212 | |
| 213 | self.complete_and_wait(drive='mirror', wait_ready=False) |
| 214 | |
| 215 | class TestThrottledWithNbdExportBase(iotests.QMPTestCase): |
| 216 | image_len = 128 * 1024 * 1024 # MB |
| 217 | iops: Optional[int] = None |
| 218 | background_processes: List['subprocess.Popen[str]'] = [] |
| 219 | |
| 220 | def setUp(self): |
| 221 | # Must be set by subclasses |
| 222 | self.assertIsNotNone(self.iops) |
| 223 | |
| 224 | qemu_img('create', '-f', iotests.imgfmt, source_img, '128M') |
| 225 | qemu_img('create', '-f', iotests.imgfmt, target_img, '128M') |
| 226 | |
| 227 | self.vm = iotests.VM() |
| 228 | self.vm.launch() |
| 229 | |
| 230 | self.vm.cmd('object-add', **{ |
| 231 | 'qom-type': 'throttle-group', |
| 232 | 'id': 'thrgr', |
| 233 | 'limits': { |
| 234 | 'iops-total': self.iops, |
| 235 | 'iops-total-max': self.iops |
| 236 | } |
| 237 | }) |
| 238 | |
| 239 | self.vm.cmd('blockdev-add', **{ |
| 240 | 'node-name': 'source-node', |
| 241 | 'driver': 'throttle', |
| 242 | 'throttle-group': 'thrgr', |
| 243 | 'file': { |
| 244 | 'driver': iotests.imgfmt, |
| 245 | 'file': { |
| 246 | 'driver': 'file', |
| 247 | 'filename': source_img |
| 248 | } |
| 249 | } |
| 250 | }) |
| 251 | |
| 252 | self.vm.cmd('blockdev-add', **{ |
| 253 | 'node-name': 'target-node', |
| 254 | 'driver': iotests.imgfmt, |
| 255 | 'file': { |
| 256 | 'driver': 'file', |
| 257 | 'filename': target_img |
| 258 | } |
| 259 | }) |
| 260 | |
| 261 | self.nbd_sock = iotests.file_path('nbd.sock', |
| 262 | base_dir=iotests.sock_dir) |
| 263 | self.nbd_url = f'nbd+unix:///source-node?socket={self.nbd_sock}' |
| 264 | |
| 265 | self.vm.cmd('nbd-server-start', addr={ |
| 266 | 'type': 'unix', |
| 267 | 'data': { |
| 268 | 'path': self.nbd_sock |
| 269 | } |
| 270 | }) |
| 271 | |
| 272 | self.vm.cmd('block-export-add', id='exp0', type='nbd', |
| 273 | node_name='source-node', writable=True) |
| 274 | |
| 275 | def tearDown(self): |
| 276 | # Wait for background requests to settle |
| 277 | try: |
| 278 | while True: |
| 279 | p = self.background_processes.pop() |
| 280 | while True: |
| 281 | try: |
| 282 | p.wait(timeout=0.0) |
| 283 | break |
| 284 | except subprocess.TimeoutExpired: |
| 285 | self.vm.qtest(f'clock_step {1 * 1000 * 1000 * 1000}') |
| 286 | try: |
| 287 | p.kill() |
| 288 | p.stdout.close() |
| 289 | except: |
| 290 | pass |
| 291 | except IndexError: |
| 292 | pass |
| 293 | |
| 294 | # Cancel ongoing block jobs |
| 295 | for job in self.vm.qmp('query-jobs')['return']: |
| 296 | self.vm.qmp('block-job-cancel', device=job['id'], force=True) |
| 297 | |
| 298 | while True: |
| 299 | self.vm.qtest(f'clock_step {1 * 1000 * 1000 * 1000}') |
| 300 | if len(self.vm.qmp('query-jobs')['return']) == 0: |
| 301 | break |
| 302 | |
| 303 | self.vm.shutdown() |
| 304 | os.remove(source_img) |
| 305 | os.remove(target_img) |
| 306 | |
| 307 | |
| 308 | class TestLowThrottledWithNbdExport(TestThrottledWithNbdExportBase): |
| 309 | iops = 16 |
| 310 | |
| 311 | @iotests.skip_flaky("https://gitlab.com/qemu-project/qemu/-/work_items/3513") |
| 312 | def testUnderLoad(self): |
| 313 | ''' |
| 314 | Throttle the source node, then issue a whole bunch of external requests |
| 315 | while the mirror job (in write-blocking mode) is running. We want to |
| 316 | see background requests being issued even while the source is under |
| 317 | full load by active writes, so that progress can be made towards READY. |
| 318 | ''' |
| 319 | |
| 320 | # Fill the first half of the source image; do not fill the second half, |
| 321 | # that is where we will have active requests occur. This ensures that |
| 322 | # active mirroring itself will not directly contribute to the job's |
| 323 | # progress (because when the job was started, those areas were not |
| 324 | # intended to be copied, so active mirroring will only lead to not |
| 325 | # losing progress, but also not making any). |
| 326 | self.vm.hmp_qemu_io('source-node', |
| 327 | f'aio_write -P 1 0 {self.image_len // 2}') |
| 328 | self.vm.qtest(f'clock_step {1 * 1000 * 1000 * 1000}') |
| 329 | |
| 330 | # Launch the mirror job |
| 331 | mirror_buf_size = 65536 |
| 332 | self.vm.cmd('blockdev-mirror', |
| 333 | job_id='mirror', |
| 334 | filter_node_name='mirror-node', |
| 335 | device='source-node', |
| 336 | target='target-node', |
| 337 | sync='full', |
| 338 | copy_mode='write-blocking', |
| 339 | buf_size=mirror_buf_size) |
| 340 | |
| 341 | # We create the external requests via qemu-io processes on the NBD |
| 342 | # server. Have their offset start in the middle of the image so they |
| 343 | # do not overlap with the background requests (which start from the |
| 344 | # beginning). |
| 345 | active_request_offset = self.image_len // 2 |
| 346 | active_request_len = 4096 |
| 347 | |
| 348 | # Create enough requests to saturate the node for 5 seconds |
| 349 | for _ in range(0, 5 * self.iops): |
| 350 | req = f'write -P 42 {active_request_offset} {active_request_len}' |
| 351 | active_request_offset += active_request_len |
| 352 | p = iotests.qemu_io_popen('-f', 'nbd', self.nbd_url, '-c', req) |
| 353 | self.background_processes += [p] |
| 354 | |
| 355 | # Now advance the clock one I/O operation at a time by the 4 seconds |
| 356 | # (i.e. one less than 5). We expect the mirror job to issue background |
| 357 | # operations here, even though active requests are still in flight. |
| 358 | # The active requests will take precedence, however, because they have |
| 359 | # been issued earlier than mirror's background requests. |
| 360 | # Once the active requests we have started above are done (i.e. after 5 |
| 361 | # virtual seconds), we expect those background requests to be worked |
| 362 | # on. We only advance 4 seconds here to avoid race conditions. |
| 363 | for _ in range(0, 4 * self.iops): |
| 364 | step = math.ceil(1 * 1000 * 1000 * 1000 / self.iops) |
| 365 | self.vm.qtest(f'clock_step {step}') |
| 366 | |
| 367 | # Note how much remains to be done until the mirror job is finished |
| 368 | job_status = self.vm.qmp('query-jobs')['return'][0] |
| 369 | start_remaining = job_status['total-progress'] - \ |
| 370 | job_status['current-progress'] |
| 371 | |
| 372 | # Create a whole bunch of more active requests |
| 373 | for _ in range(0, 10 * self.iops): |
| 374 | req = f'write -P 42 {active_request_offset} {active_request_len}' |
| 375 | active_request_offset += active_request_len |
| 376 | p = iotests.qemu_io_popen('-f', 'nbd', self.nbd_url, '-c', req) |
| 377 | self.background_processes += [p] |
| 378 | |
| 379 | # Let the clock advance more. After 1 second, as noted above, we |
| 380 | # expect the background requests to be worked on. Give them a couple |
| 381 | # of seconds (specifically 4) to see their impact. |
| 382 | for _ in range(0, 5 * self.iops): |
| 383 | step = math.ceil(1 * 1000 * 1000 * 1000 / self.iops) |
| 384 | self.vm.qtest(f'clock_step {step}') |
| 385 | |
| 386 | # Note how much remains to be done now. We expect this number to be |
| 387 | # reduced thanks to those background requests. |
| 388 | job_status = self.vm.qmp('query-jobs')['return'][0] |
| 389 | end_remaining = job_status['total-progress'] - \ |
| 390 | job_status['current-progress'] |
| 391 | |
| 392 | # See that indeed progress was being made on the job, even while the |
| 393 | # node was saturated with active requests |
| 394 | self.assertGreater(start_remaining - end_remaining, 0) |
| 395 | |
| 396 | |
| 397 | class TestHighThrottledWithNbdExport(TestThrottledWithNbdExportBase): |
| 398 | iops = 1024 |
| 399 | |
| 400 | def testActiveOnCreation(self): |
| 401 | ''' |
| 402 | Issue requests on the mirror source node right as the mirror is |
| 403 | instated. It's possible that requests occur before the actual job is |
| 404 | created, but after the node has been put into the graph. Write |
| 405 | requests across the node must in that case be forwarded to the source |
| 406 | node without attempting to mirror them (there is no job object yet, so |
| 407 | attempting to access it would cause a segfault). |
| 408 | We do this with a lightly throttled node (i.e. quite high IOPS limit). |
| 409 | Using throttling seems to increase reproductivity, but if the limit is |
| 410 | too low, all requests allowed per second will be submitted before |
| 411 | mirror_start_job() gets to the problematic point. |
| 412 | ''' |
| 413 | |
| 414 | # Let qemu-img bench create write requests (enough for two seconds on |
| 415 | # the virtual clock) |
| 416 | bench_args = ['bench', '-w', '-d', '1024', '-f', 'nbd', |
| 417 | '-c', str(self.iops * 2), self.nbd_url] |
| 418 | p = iotests.qemu_tool_popen(iotests.qemu_img_args + bench_args) |
| 419 | self.background_processes += [p] |
| 420 | |
| 421 | # Give qemu-img bench time to start up and issue requests |
| 422 | time.sleep(1.0) |
| 423 | # Flush the request queue, so new requests can come in right as we |
| 424 | # start blockdev-mirror |
| 425 | self.vm.qtest(f'clock_step {1 * 1000 * 1000 * 1000}') |
| 426 | |
| 427 | self.vm.cmd('blockdev-mirror', |
| 428 | job_id='mirror', |
| 429 | device='source-node', |
| 430 | target='target-node', |
| 431 | sync='full', |
| 432 | copy_mode='write-blocking') |
| 433 | |
| 434 | |
| 435 | if __name__ == '__main__': |
| 436 | iotests.main(supported_fmts=['qcow2', 'raw'], |
| 437 | supported_protocols=['file'], |
| 438 | require_hmp=True) |