]>
Commit | Line | Data |
---|---|---|
254912f3 | 1 | /* |
f70aedc4 | 2 | * Copyright (C) 1996-2021 The Squid Software Foundation and contributors |
bbc27441 AJ |
3 | * |
4 | * Squid software is distributed under GPLv2+ license and includes | |
5 | * contributions from numerous individuals and organizations. | |
6 | * Please see the COPYING and CONTRIBUTORS files for details. | |
254912f3 AR |
7 | */ |
8 | ||
bbc27441 AJ |
9 | /* DEBUG: section 47 Store Directory Routines */ |
10 | ||
f7f3304a | 11 | #include "squid.h" |
ccfbe8f4 | 12 | #include "base/CodeContext.h" |
f5591061 | 13 | #include "base/RunnersRegistry.h" |
254912f3 AR |
14 | #include "base/TextException.h" |
15 | #include "DiskIO/IORequestor.h" | |
16 | #include "DiskIO/IpcIo/IpcIoFile.h" | |
17 | #include "DiskIO/ReadRequest.h" | |
18 | #include "DiskIO/WriteRequest.h" | |
c4ad1349 | 19 | #include "fd.h" |
b3f7fd88 | 20 | #include "fs_io.h" |
582c2af2 | 21 | #include "globals.h" |
21d845b1 | 22 | #include "ipc/mem/Pages.h" |
254912f3 AR |
23 | #include "ipc/Messages.h" |
24 | #include "ipc/Port.h" | |
f5591061 | 25 | #include "ipc/Queue.h" |
6ccfd70a | 26 | #include "ipc/StrandCoord.h" |
9a51593d | 27 | #include "ipc/StrandSearch.h" |
254912f3 | 28 | #include "ipc/UdsOp.h" |
65e41a45 | 29 | #include "sbuf/SBuf.h" |
4d5904f7 | 30 | #include "SquidConfig.h" |
21d845b1 FC |
31 | #include "SquidTime.h" |
32 | #include "StatCounters.h" | |
5bed43d6 | 33 | #include "tools.h" |
254912f3 | 34 | |
1a30fdf5 AJ |
35 | #include <cerrno> |
36 | ||
254912f3 AR |
37 | CBDATA_CLASS_INIT(IpcIoFile); |
38 | ||
f5591061 DK |
39 | /// shared memory segment path to use for IpcIoFile maps |
40 | static const char *const ShmLabel = "io_file"; | |
ea2cdeb6 DK |
41 | /// a single worker-to-disker or disker-to-worker queue capacity; up |
42 | /// to 2*QueueCapacity I/O requests queued between a single worker and | |
43 | /// a single disker | |
44 | // TODO: make configurable or compute from squid.conf settings if possible | |
45 | static const int QueueCapacity = 1024; | |
f5591061 | 46 | |
fa61cefe | 47 | const double IpcIoFile::Timeout = 7; // seconds; XXX: ALL,9 may require more |
b2aa0934 DK |
48 | IpcIoFile::IpcIoFileList IpcIoFile::WaitingForOpen; |
49 | IpcIoFile::IpcIoFilesMap IpcIoFile::IpcIoFiles; | |
829030b5 | 50 | std::unique_ptr<IpcIoFile::Queue> IpcIoFile::queue; |
254912f3 | 51 | |
c792f7fc AR |
52 | bool IpcIoFile::DiskerHandleMoreRequestsScheduled = false; |
53 | ||
7015a149 AR |
54 | static bool DiskerOpen(const SBuf &path, int flags, mode_t mode); |
55 | static void DiskerClose(const SBuf &path); | |
254912f3 | 56 | |
fa61cefe AR |
57 | /// IpcIo wrapper for debugs() streams; XXX: find a better class name |
58 | struct SipcIo { | |
59 | SipcIo(int aWorker, const IpcIoMsg &aMsg, int aDisker): | |
f53969cc | 60 | worker(aWorker), msg(aMsg), disker(aDisker) {} |
fa61cefe AR |
61 | |
62 | int worker; | |
63 | const IpcIoMsg &msg; | |
64 | int disker; | |
65 | }; | |
66 | ||
67 | std::ostream & | |
68 | operator <<(std::ostream &os, const SipcIo &sio) | |
69 | { | |
70 | return os << "ipcIo" << sio.worker << '.' << sio.msg.requestId << | |
d8ee9e8d | 71 | sio.msg.command << sio.disker; |
fa61cefe AR |
72 | } |
73 | ||
d8ee9e8d EB |
74 | /* IpcIo::Command */ |
75 | ||
76 | std::ostream & | |
77 | operator <<(std::ostream &os, const IpcIo::Command command) | |
78 | { | |
79 | switch (command) { | |
80 | case IpcIo::cmdNone: | |
81 | return os << '-'; | |
82 | case IpcIo::cmdOpen: | |
83 | return os << 'o'; | |
84 | case IpcIo::cmdRead: | |
85 | return os << 'r'; | |
86 | case IpcIo::cmdWrite: | |
87 | return os << 'w'; | |
88 | } | |
89 | // unreachable code | |
90 | return os << static_cast<int>(command); | |
91 | } | |
92 | ||
93 | /* IpcIoFile */ | |
94 | ||
254912f3 | 95 | IpcIoFile::IpcIoFile(char const *aDb): |
30fe12bc EB |
96 | dbName(aDb), |
97 | myPid(getpid()), | |
98 | diskId(-1), | |
99 | error_(false), | |
100 | lastRequestId(0), | |
f53969cc SM |
101 | olderRequests(&requestMap1), newerRequests(&requestMap2), |
102 | timeoutCheckScheduled(false) | |
254912f3 | 103 | { |
30fe12bc | 104 | assert(myPid >= 0); |
254912f3 AR |
105 | } |
106 | ||
107 | IpcIoFile::~IpcIoFile() | |
108 | { | |
ebaabe74 AR |
109 | SWALLOW_EXCEPTIONS({ |
110 | if (diskId >= 0) { | |
111 | const auto i = IpcIoFiles.find(diskId); | |
112 | Must(i != IpcIoFiles.end()); | |
113 | Must(i->second == this); | |
114 | IpcIoFiles.erase(i); | |
115 | } | |
116 | }); | |
254912f3 AR |
117 | } |
118 | ||
43ebbac3 AR |
119 | void |
120 | IpcIoFile::configure(const Config &cfg) | |
121 | { | |
122 | DiskFile::configure(cfg); | |
123 | config = cfg; | |
124 | } | |
125 | ||
254912f3 AR |
126 | void |
127 | IpcIoFile::open(int flags, mode_t mode, RefCount<IORequestor> callback) | |
128 | { | |
129 | ioRequestor = callback; | |
130 | Must(diskId < 0); // we do not know our disker yet | |
f5591061 DK |
131 | |
132 | if (!queue.get()) | |
133 | queue.reset(new Queue(ShmLabel, IamWorkerProcess() ? Queue::groupA : Queue::groupB, KidIdentifier)); | |
254912f3 AR |
134 | |
135 | if (IamDiskProcess()) { | |
7015a149 | 136 | error_ = !DiskerOpen(SBuf(dbName.termedBuf()), flags, mode); |
f7091279 DK |
137 | if (error_) |
138 | return; | |
139 | ||
b2aa0934 DK |
140 | diskId = KidIdentifier; |
141 | const bool inserted = | |
142 | IpcIoFiles.insert(std::make_pair(diskId, this)).second; | |
143 | Must(inserted); | |
9a51593d | 144 | |
6c6a656a | 145 | queue->localRateLimit().store(config.ioRate); |
df881a0f | 146 | |
6ccfd70a | 147 | Ipc::StrandMessage::NotifyCoordinator(Ipc::mtRegisterStrand, dbName.termedBuf()); |
f7091279 | 148 | |
e528e570 | 149 | ioRequestor->ioCompletedNotification(); |
254912f3 | 150 | return; |
9a51593d | 151 | } |
254912f3 | 152 | |
4c218615 | 153 | const Ipc::StrandSearchRequest request(dbName); |
9a51593d DK |
154 | Ipc::TypedMsgHdr msg; |
155 | request.pack(msg); | |
1ee292b7 | 156 | Ipc::SendMessage(Ipc::Port::CoordinatorAddr(), msg); |
9a51593d | 157 | |
b2aa0934 DK |
158 | WaitingForOpen.push_back(this); |
159 | ||
160 | eventAdd("IpcIoFile::OpenTimeout", &IpcIoFile::OpenTimeout, | |
161 | this, Timeout, 0, false); // "this" pointer is used as id | |
254912f3 AR |
162 | } |
163 | ||
164 | void | |
6ccfd70a | 165 | IpcIoFile::openCompleted(const Ipc::StrandMessage *const response) |
9199139f | 166 | { |
9a51593d | 167 | Must(diskId < 0); // we do not know our disker yet |
9a51593d DK |
168 | |
169 | if (!response) { | |
82ed74dc AR |
170 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " communication " << |
171 | "channel establishment timeout"); | |
caca86d7 | 172 | error_ = true; |
9a51593d DK |
173 | } else { |
174 | diskId = response->strand.kidId; | |
175 | if (diskId >= 0) { | |
b2aa0934 DK |
176 | const bool inserted = |
177 | IpcIoFiles.insert(std::make_pair(diskId, this)).second; | |
178 | Must(inserted); | |
9a51593d | 179 | } else { |
254912f3 | 180 | error_ = true; |
82ed74dc AR |
181 | debugs(79, DBG_IMPORTANT, "ERROR: no disker claimed " << |
182 | "responsibility for " << dbName); | |
9a51593d DK |
183 | } |
184 | } | |
254912f3 AR |
185 | |
186 | ioRequestor->ioCompletedNotification(); | |
187 | } | |
188 | ||
189 | /** | |
190 | * Alias for IpcIoFile::open(...) | |
191 | \copydoc IpcIoFile::open(int flags, mode_t mode, RefCount<IORequestor> callback) | |
192 | */ | |
193 | void | |
194 | IpcIoFile::create(int flags, mode_t mode, RefCount<IORequestor> callback) | |
195 | { | |
196 | assert(false); // check | |
197 | /* We use the same logic path for open */ | |
198 | open(flags, mode, callback); | |
199 | } | |
200 | ||
201 | void | |
202 | IpcIoFile::close() | |
203 | { | |
204 | assert(ioRequestor != NULL); | |
205 | ||
206 | if (IamDiskProcess()) | |
7015a149 | 207 | DiskerClose(SBuf(dbName.termedBuf())); |
9a51593d | 208 | // XXX: else nothing to do? |
254912f3 AR |
209 | |
210 | ioRequestor->closeCompleted(); | |
211 | } | |
212 | ||
213 | bool | |
214 | IpcIoFile::canRead() const | |
215 | { | |
7e2d5ef6 | 216 | return diskId >= 0 && !error_ && canWait(); |
254912f3 AR |
217 | } |
218 | ||
219 | bool | |
220 | IpcIoFile::canWrite() const | |
221 | { | |
7e2d5ef6 | 222 | return diskId >= 0 && !error_ && canWait(); |
254912f3 AR |
223 | } |
224 | ||
225 | bool | |
226 | IpcIoFile::error() const | |
227 | { | |
228 | return error_; | |
229 | } | |
230 | ||
231 | void | |
232 | IpcIoFile::read(ReadRequest *readRequest) | |
233 | { | |
234 | debugs(79,3, HERE << "(disker" << diskId << ", " << readRequest->len << ", " << | |
9199139f | 235 | readRequest->offset << ")"); |
254912f3 AR |
236 | |
237 | assert(ioRequestor != NULL); | |
254912f3 AR |
238 | assert(readRequest->offset >= 0); |
239 | Must(!error_); | |
240 | ||
241 | //assert(minOffset < 0 || minOffset <= readRequest->offset); | |
242 | //assert(maxOffset < 0 || readRequest->offset + readRequest->len <= (uint64_t)maxOffset); | |
243 | ||
9a51593d | 244 | IpcIoPendingRequest *const pending = new IpcIoPendingRequest(this); |
254912f3 | 245 | pending->readRequest = readRequest; |
7a907247 | 246 | push(pending); |
254912f3 AR |
247 | } |
248 | ||
249 | void | |
250 | IpcIoFile::readCompleted(ReadRequest *readRequest, | |
8ed94021 | 251 | IpcIoMsg *const response) |
254912f3 | 252 | { |
caca86d7 | 253 | bool ioError = false; |
9a51593d | 254 | if (!response) { |
0ef0509e | 255 | debugs(79, 3, HERE << "error: timeout"); |
caca86d7 | 256 | ioError = true; // I/O timeout does not warrant setting error_? |
9a51593d | 257 | } else { |
70e3f706 | 258 | if (response->xerrno) { |
82ed74dc AR |
259 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " read: " << |
260 | xstrerr(response->xerrno)); | |
70e3f706 | 261 | ioError = error_ = true; |
9199139f | 262 | } else if (!response->page) { |
82ed74dc AR |
263 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " read ran " << |
264 | "out of shared memory pages"); | |
70e3f706 DK |
265 | ioError = true; |
266 | } else { | |
267 | const char *const buf = Ipc::Mem::PagePointer(response->page); | |
268 | memcpy(readRequest->buf, buf, response->len); | |
269 | } | |
254912f3 | 270 | |
70e3f706 DK |
271 | Ipc::Mem::PutPage(response->page); |
272 | } | |
8ed94021 | 273 | |
caca86d7 | 274 | const ssize_t rlen = ioError ? -1 : (ssize_t)readRequest->len; |
9199139f | 275 | const int errflag = ioError ? DISK_ERROR :DISK_OK; |
20b0e1fe | 276 | ioRequestor->readCompleted(readRequest->buf, rlen, errflag, readRequest); |
254912f3 AR |
277 | } |
278 | ||
279 | void | |
280 | IpcIoFile::write(WriteRequest *writeRequest) | |
281 | { | |
282 | debugs(79,3, HERE << "(disker" << diskId << ", " << writeRequest->len << ", " << | |
9199139f | 283 | writeRequest->offset << ")"); |
254912f3 AR |
284 | |
285 | assert(ioRequestor != NULL); | |
254912f3 AR |
286 | assert(writeRequest->len > 0); // TODO: work around mmap failures on zero-len? |
287 | assert(writeRequest->offset >= 0); | |
288 | Must(!error_); | |
289 | ||
290 | //assert(minOffset < 0 || minOffset <= writeRequest->offset); | |
291 | //assert(maxOffset < 0 || writeRequest->offset + writeRequest->len <= (uint64_t)maxOffset); | |
292 | ||
9a51593d | 293 | IpcIoPendingRequest *const pending = new IpcIoPendingRequest(this); |
254912f3 | 294 | pending->writeRequest = writeRequest; |
7a907247 | 295 | push(pending); |
254912f3 AR |
296 | } |
297 | ||
298 | void | |
299 | IpcIoFile::writeCompleted(WriteRequest *writeRequest, | |
9a51593d | 300 | const IpcIoMsg *const response) |
254912f3 | 301 | { |
caca86d7 | 302 | bool ioError = false; |
9a51593d | 303 | if (!response) { |
4e4f7fd9 | 304 | debugs(79, 3, "disker " << diskId << " timeout"); |
caca86d7 | 305 | ioError = true; // I/O timeout does not warrant setting error_? |
9199139f | 306 | } else if (response->xerrno) { |
4e4f7fd9 AR |
307 | debugs(79, DBG_IMPORTANT, "ERROR: disker " << diskId << |
308 | " error writing " << writeRequest->len << " bytes at " << | |
309 | writeRequest->offset << ": " << xstrerr(response->xerrno) << | |
310 | "; this worker will stop using " << dbName); | |
9a51593d | 311 | ioError = error_ = true; |
9199139f | 312 | } else if (response->len != writeRequest->len) { |
4e4f7fd9 AR |
313 | debugs(79, DBG_IMPORTANT, "ERROR: disker " << diskId << " wrote " << |
314 | response->len << " instead of " << writeRequest->len << | |
315 | " bytes (offset " << writeRequest->offset << "); " << | |
316 | "this worker will stop using " << dbName); | |
254912f3 AR |
317 | error_ = true; |
318 | } | |
319 | ||
320 | if (writeRequest->free_func) | |
321 | (writeRequest->free_func)(const_cast<char*>(writeRequest->buf)); // broken API? | |
322 | ||
caca86d7 | 323 | if (!ioError) { |
254912f3 | 324 | debugs(79,5, HERE << "wrote " << writeRequest->len << " to disker" << |
9199139f AR |
325 | diskId << " at " << writeRequest->offset); |
326 | } | |
254912f3 | 327 | |
caca86d7 | 328 | const ssize_t rlen = ioError ? 0 : (ssize_t)writeRequest->len; |
9199139f | 329 | const int errflag = ioError ? DISK_ERROR :DISK_OK; |
254912f3 AR |
330 | ioRequestor->writeCompleted(errflag, rlen, writeRequest); |
331 | } | |
332 | ||
333 | bool | |
334 | IpcIoFile::ioInProgress() const | |
335 | { | |
9a51593d | 336 | return !olderRequests->empty() || !newerRequests->empty(); |
254912f3 AR |
337 | } |
338 | ||
9a51593d | 339 | /// track a new pending request |
254912f3 | 340 | void |
28bd45ba | 341 | IpcIoFile::trackPendingRequest(const unsigned int id, IpcIoPendingRequest *const pending) |
254912f3 | 342 | { |
28bd45ba AR |
343 | const std::pair<RequestMap::iterator,bool> result = |
344 | newerRequests->insert(std::make_pair(id, pending)); | |
345 | Must(result.second); // failures means that id was not unique | |
9a51593d DK |
346 | if (!timeoutCheckScheduled) |
347 | scheduleTimeoutCheck(); | |
348 | } | |
254912f3 | 349 | |
9a51593d DK |
350 | /// push an I/O request to disker |
351 | void | |
7a907247 | 352 | IpcIoFile::push(IpcIoPendingRequest *const pending) |
9a51593d | 353 | { |
fa61cefe | 354 | // prevent queue overflows: check for responses to earlier requests |
28bd45ba | 355 | // warning: this call may result in indirect push() recursion |
ccfbe8f4 AR |
356 | CallService(nullptr, [] { |
357 | HandleResponses("before push"); | |
358 | }); | |
fa61cefe | 359 | |
a1c98830 | 360 | debugs(47, 7, HERE); |
9a51593d | 361 | Must(diskId >= 0); |
7a907247 DK |
362 | Must(pending); |
363 | Must(pending->readRequest || pending->writeRequest); | |
9a51593d DK |
364 | |
365 | IpcIoMsg ipcIo; | |
8ed94021 | 366 | try { |
28bd45ba AR |
367 | if (++lastRequestId == 0) // don't use zero value as requestId |
368 | ++lastRequestId; | |
8ed94021 | 369 | ipcIo.requestId = lastRequestId; |
0a11e039 | 370 | ipcIo.start = current_time; |
30fe12bc | 371 | ipcIo.workerPid = myPid; |
8ed94021 DK |
372 | if (pending->readRequest) { |
373 | ipcIo.command = IpcIo::cmdRead; | |
374 | ipcIo.offset = pending->readRequest->offset; | |
375 | ipcIo.len = pending->readRequest->len; | |
376 | } else { // pending->writeRequest | |
377 | Must(pending->writeRequest->len <= Ipc::Mem::PageSize()); | |
551f8a18 | 378 | if (!Ipc::Mem::GetPage(Ipc::Mem::PageId::ioPage, ipcIo.page)) { |
8ed94021 | 379 | ipcIo.len = 0; |
551f8a18 | 380 | throw TexcHere("run out of shared memory pages for IPC I/O"); |
8ed94021 DK |
381 | } |
382 | ipcIo.command = IpcIo::cmdWrite; | |
383 | ipcIo.offset = pending->writeRequest->offset; | |
384 | ipcIo.len = pending->writeRequest->len; | |
385 | char *const buf = Ipc::Mem::PagePointer(ipcIo.page); | |
386 | memcpy(buf, pending->writeRequest->buf, ipcIo.len); // optimize away | |
387 | } | |
254912f3 | 388 | |
f5591061 | 389 | debugs(47, 7, HERE << "pushing " << SipcIo(KidIdentifier, ipcIo, diskId)); |
fa61cefe | 390 | |
387e1d45 EB |
391 | // protect DiskerHandleRequest() from pop queue overflow |
392 | if (pendingRequests() >= QueueCapacity) | |
393 | throw Ipc::OneToOneUniQueue::Full(); | |
394 | ||
f5591061 | 395 | if (queue->push(diskId, ipcIo)) |
fa61cefe | 396 | Notify(diskId); // must notify disker |
28bd45ba | 397 | trackPendingRequest(ipcIo.requestId, pending); |
f5591061 | 398 | } catch (const Queue::Full &) { |
82ed74dc AR |
399 | debugs(47, DBG_IMPORTANT, "ERROR: worker I/O push queue for " << |
400 | dbName << " overflow: " << | |
fa61cefe AR |
401 | SipcIo(KidIdentifier, ipcIo, diskId)); // TODO: report queue len |
402 | // TODO: grow queue size | |
d0c5cf9e CG |
403 | if (ipcIo.page) |
404 | Ipc::Mem::PutPage(ipcIo.page); | |
9199139f | 405 | |
0ef0509e | 406 | pending->completeIo(NULL); |
8ed94021 DK |
407 | delete pending; |
408 | } catch (const TextException &e) { | |
82ed74dc | 409 | debugs(47, DBG_IMPORTANT, "ERROR: " << dbName << " exception: " << e.what()); |
0ef0509e | 410 | pending->completeIo(NULL); |
7a907247 | 411 | delete pending; |
9a51593d | 412 | } |
254912f3 AR |
413 | } |
414 | ||
0a11e039 AR |
415 | /// whether we think there is enough time to complete the I/O |
416 | bool | |
9199139f AR |
417 | IpcIoFile::canWait() const |
418 | { | |
43ebbac3 | 419 | if (!config.ioTimeout) |
0a11e039 AR |
420 | return true; // no timeout specified |
421 | ||
422 | IpcIoMsg oldestIo; | |
5aac671b | 423 | if (!queue->findOldest(diskId, oldestIo) || oldestIo.start.tv_sec <= 0) |
0a11e039 AR |
424 | return true; // we cannot estimate expected wait time; assume it is OK |
425 | ||
55939a01 AR |
426 | const int oldestWait = tvSubMsec(oldestIo.start, current_time); |
427 | ||
428 | int rateWait = -1; // time in millisecons | |
6c6a656a | 429 | const int ioRate = queue->rateLimit(diskId).load(); |
55939a01 AR |
430 | if (ioRate > 0) { |
431 | // if there are N requests pending, the new one will wait at | |
432 | // least N/max-swap-rate seconds | |
75017bc9 | 433 | rateWait = static_cast<int>(1e3 * queue->outSize(diskId) / ioRate); |
55939a01 AR |
434 | // adjust N/max-swap-rate value based on the queue "balance" |
435 | // member, in case we have been borrowing time against future | |
436 | // I/O already | |
437 | rateWait += queue->balance(diskId); | |
438 | } | |
439 | ||
440 | const int expectedWait = max(oldestWait, rateWait); | |
0a11e039 | 441 | if (expectedWait < 0 || |
43ebbac3 | 442 | static_cast<time_msec_t>(expectedWait) < config.ioTimeout) |
2f8abb64 | 443 | return true; // expected wait time is acceptable |
0a11e039 | 444 | |
c792f7fc AR |
445 | debugs(47,2, HERE << "cannot wait: " << expectedWait << |
446 | " oldest: " << SipcIo(KidIdentifier, oldestIo, diskId)); | |
0a11e039 AR |
447 | return false; // do not want to wait that long |
448 | } | |
449 | ||
9a51593d | 450 | /// called when coordinator responds to worker open request |
254912f3 | 451 | void |
6ccfd70a | 452 | IpcIoFile::HandleOpenResponse(const Ipc::StrandMessage &response) |
254912f3 | 453 | { |
9a51593d | 454 | debugs(47, 7, HERE << "coordinator response to open request"); |
b2aa0934 | 455 | for (IpcIoFileList::iterator i = WaitingForOpen.begin(); |
9199139f | 456 | i != WaitingForOpen.end(); ++i) { |
b2aa0934 DK |
457 | if (response.strand.tag == (*i)->dbName) { |
458 | (*i)->openCompleted(&response); | |
459 | WaitingForOpen.erase(i); | |
460 | return; | |
461 | } | |
462 | } | |
463 | ||
464 | debugs(47, 4, HERE << "LATE disker response to open for " << | |
465 | response.strand.tag); | |
466 | // nothing we can do about it; completeIo() has been called already | |
9a51593d | 467 | } |
254912f3 | 468 | |
9a51593d | 469 | void |
f5591061 | 470 | IpcIoFile::HandleResponses(const char *const when) |
fa61cefe AR |
471 | { |
472 | debugs(47, 4, HERE << "popping all " << when); | |
fa61cefe AR |
473 | IpcIoMsg ipcIo; |
474 | // get all responses we can: since we are not pushing, this will stop | |
f5591061 DK |
475 | int diskId; |
476 | while (queue->pop(diskId, ipcIo)) { | |
477 | const IpcIoFilesMap::const_iterator i = IpcIoFiles.find(diskId); | |
478 | Must(i != IpcIoFiles.end()); // TODO: warn but continue | |
479 | i->second->handleResponse(ipcIo); | |
480 | } | |
9a51593d | 481 | } |
254912f3 | 482 | |
9a51593d | 483 | void |
8ed94021 | 484 | IpcIoFile::handleResponse(IpcIoMsg &ipcIo) |
9a51593d DK |
485 | { |
486 | const int requestId = ipcIo.requestId; | |
fa61cefe | 487 | |
9a51593d DK |
488 | Must(requestId); |
489 | if (IpcIoPendingRequest *const pending = dequeueRequest(requestId)) { | |
ccfbe8f4 AR |
490 | CallBack(pending->codeContext, [&] { |
491 | debugs(47, 7, "popped disker response to " << SipcIo(KidIdentifier, ipcIo, diskId)); | |
30fe12bc EB |
492 | if (myPid == ipcIo.workerPid) |
493 | pending->completeIo(&ipcIo); | |
494 | else | |
495 | debugs(47, 5, "ignoring response meant for our predecessor PID: " << ipcIo.workerPid); | |
ccfbe8f4 AR |
496 | delete pending; // XXX: leaking if throwing |
497 | }); | |
caca86d7 | 498 | } else { |
ccfbe8f4 | 499 | debugs(47, 4, "LATE disker response to " << SipcIo(KidIdentifier, ipcIo, diskId)); |
caca86d7 AR |
500 | // nothing we can do about it; completeIo() has been called already |
501 | } | |
254912f3 AR |
502 | } |
503 | ||
254912f3 | 504 | void |
9a51593d | 505 | IpcIoFile::Notify(const int peerId) |
254912f3 | 506 | { |
fa61cefe | 507 | // TODO: Count and report the total number of notifications, pops, pushes. |
a1c98830 | 508 | debugs(47, 7, HERE << "kid" << peerId); |
9a51593d DK |
509 | Ipc::TypedMsgHdr msg; |
510 | msg.setType(Ipc::mtIpcIoNotification); // TODO: add proper message type? | |
511 | msg.putInt(KidIdentifier); | |
1ee292b7 | 512 | const String addr = Ipc::Port::MakeAddr(Ipc::strandAddrLabel, peerId); |
9a51593d DK |
513 | Ipc::SendMessage(addr, msg); |
514 | } | |
515 | ||
516 | void | |
517 | IpcIoFile::HandleNotification(const Ipc::TypedMsgHdr &msg) | |
518 | { | |
fa61cefe AR |
519 | const int from = msg.getInt(); |
520 | debugs(47, 7, HERE << "from " << from); | |
f5591061 DK |
521 | queue->clearReaderSignal(from); |
522 | if (IamDiskProcess()) | |
523 | DiskerHandleRequests(); | |
524 | else | |
525 | HandleResponses("after notification"); | |
9a51593d DK |
526 | } |
527 | ||
d8ee9e8d EB |
528 | void |
529 | IpcIoFile::StatQueue(std::ostream &os) | |
530 | { | |
531 | if (queue.get()) { | |
532 | os << "SMP disk I/O queues:\n"; | |
533 | queue->stat<IpcIoMsg>(os); | |
534 | } | |
535 | } | |
536 | ||
b2aa0934 DK |
537 | /// handles open request timeout |
538 | void | |
539 | IpcIoFile::OpenTimeout(void *const param) | |
540 | { | |
541 | Must(param); | |
542 | // the pointer is used for comparison only and not dereferenced | |
543 | const IpcIoFile *const ipcIoFile = | |
544 | reinterpret_cast<const IpcIoFile *>(param); | |
545 | for (IpcIoFileList::iterator i = WaitingForOpen.begin(); | |
9199139f | 546 | i != WaitingForOpen.end(); ++i) { |
b2aa0934 DK |
547 | if (*i == ipcIoFile) { |
548 | (*i)->openCompleted(NULL); | |
549 | WaitingForOpen.erase(i); | |
550 | break; | |
551 | } | |
552 | } | |
553 | } | |
554 | ||
9a51593d DK |
555 | /// IpcIoFile::checkTimeouts wrapper |
556 | void | |
557 | IpcIoFile::CheckTimeouts(void *const param) | |
558 | { | |
9a51593d | 559 | Must(param); |
b2aa0934 DK |
560 | const int diskId = reinterpret_cast<uintptr_t>(param); |
561 | debugs(47, 7, HERE << "diskId=" << diskId); | |
562 | const IpcIoFilesMap::const_iterator i = IpcIoFiles.find(diskId); | |
563 | if (i != IpcIoFiles.end()) | |
564 | i->second->checkTimeouts(); | |
9a51593d DK |
565 | } |
566 | ||
567 | void | |
568 | IpcIoFile::checkTimeouts() | |
569 | { | |
9a51593d | 570 | timeoutCheckScheduled = false; |
254912f3 | 571 | |
0ef0509e AR |
572 | // last chance to recover in case a notification message was lost, etc. |
573 | const RequestMap::size_type timeoutsBefore = olderRequests->size(); | |
574 | HandleResponses("before timeout"); | |
575 | const RequestMap::size_type timeoutsNow = olderRequests->size(); | |
576 | ||
577 | if (timeoutsBefore > timeoutsNow) { // some requests were rescued | |
578 | // notification message lost or significantly delayed? | |
82ed74dc AR |
579 | debugs(47, DBG_IMPORTANT, "WARNING: communication with " << dbName << |
580 | " may be too slow or disrupted for about " << | |
0ef0509e AR |
581 | Timeout << "s; rescued " << (timeoutsBefore - timeoutsNow) << |
582 | " out of " << timeoutsBefore << " I/Os"); | |
583 | } | |
584 | ||
585 | if (timeoutsNow) { | |
586 | debugs(47, DBG_IMPORTANT, "WARNING: abandoning " << | |
82ed74dc | 587 | timeoutsNow << ' ' << dbName << " I/Os after at least " << |
0ef0509e AR |
588 | Timeout << "s timeout"); |
589 | } | |
590 | ||
caca86d7 AR |
591 | // any old request would have timed out by now |
592 | typedef RequestMap::const_iterator RMCI; | |
9a51593d DK |
593 | for (RMCI i = olderRequests->begin(); i != olderRequests->end(); ++i) { |
594 | IpcIoPendingRequest *const pending = i->second; | |
ccfbe8f4 AR |
595 | CallBack(pending->codeContext, [&] { |
596 | const auto requestId = i->first; | |
597 | debugs(47, 7, "disker timeout; ipcIo" << KidIdentifier << '.' << requestId); | |
598 | pending->completeIo(nullptr); // no response | |
599 | delete pending; // XXX: leaking if throwing | |
600 | }); | |
9a51593d DK |
601 | } |
602 | olderRequests->clear(); | |
caca86d7 | 603 | |
9a51593d | 604 | swap(olderRequests, newerRequests); // switches pointers around |
0ef0509e | 605 | if (!olderRequests->empty() && !timeoutCheckScheduled) |
9a51593d | 606 | scheduleTimeoutCheck(); |
254912f3 AR |
607 | } |
608 | ||
caca86d7 | 609 | /// prepare to check for timeouts in a little while |
254912f3 | 610 | void |
9a51593d | 611 | IpcIoFile::scheduleTimeoutCheck() |
254912f3 | 612 | { |
ccfbe8f4 AR |
613 | // We may be running in an I/O requestor CodeContext, but are scheduling |
614 | // one-for-all CheckTimeouts() that is not specific to any request. | |
615 | CallService(nullptr, [&] { | |
616 | // we check all older requests at once so some may be wait for 2*Timeout | |
617 | eventAdd("IpcIoFile::CheckTimeouts", &IpcIoFile::CheckTimeouts, | |
618 | reinterpret_cast<void *>(diskId), Timeout, 0, false); | |
619 | timeoutCheckScheduled = true; | |
620 | }); | |
254912f3 AR |
621 | } |
622 | ||
623 | /// returns and forgets the right IpcIoFile pending request | |
624 | IpcIoPendingRequest * | |
9a51593d | 625 | IpcIoFile::dequeueRequest(const unsigned int requestId) |
254912f3 | 626 | { |
254912f3 | 627 | Must(requestId != 0); |
caca86d7 | 628 | |
9a51593d DK |
629 | RequestMap *map = NULL; |
630 | RequestMap::iterator i = requestMap1.find(requestId); | |
caca86d7 | 631 | |
9a51593d DK |
632 | if (i != requestMap1.end()) |
633 | map = &requestMap1; | |
caca86d7 | 634 | else { |
9a51593d DK |
635 | i = requestMap2.find(requestId); |
636 | if (i != requestMap2.end()) | |
637 | map = &requestMap2; | |
caca86d7 AR |
638 | } |
639 | ||
640 | if (!map) // not found in both maps | |
641 | return NULL; | |
642 | ||
643 | IpcIoPendingRequest *pending = i->second; | |
644 | map->erase(i); | |
645 | return pending; | |
254912f3 AR |
646 | } |
647 | ||
648 | int | |
9a51593d | 649 | IpcIoFile::getFD() const |
254912f3 AR |
650 | { |
651 | assert(false); // not supported; TODO: remove this method from API | |
652 | return -1; | |
653 | } | |
654 | ||
9a51593d | 655 | /* IpcIoMsg */ |
254912f3 | 656 | |
9a51593d | 657 | IpcIoMsg::IpcIoMsg(): |
f53969cc SM |
658 | requestId(0), |
659 | offset(0), | |
660 | len(0), | |
30fe12bc | 661 | workerPid(-1), // Unix-like systems use process IDs starting from 0 |
f53969cc SM |
662 | command(IpcIo::cmdNone), |
663 | xerrno(0) | |
254912f3 | 664 | { |
0a11e039 | 665 | start.tv_sec = 0; |
e19994df | 666 | start.tv_usec = 0; |
254912f3 AR |
667 | } |
668 | ||
d8ee9e8d EB |
669 | void |
670 | IpcIoMsg::stat(std::ostream &os) | |
671 | { | |
672 | timeval elapsedTime; | |
673 | tvSub(elapsedTime, start, current_time); | |
674 | os << "id: " << requestId << | |
70ac5b29 | 675 | ", offset: " << offset << |
676 | ", size: " << len << | |
677 | ", workerPid: " << workerPid << | |
678 | ", page: " << page << | |
679 | ", command: " << command << | |
680 | ", start: " << start << | |
681 | ", elapsed: " << elapsedTime << | |
682 | ", errno: " << xerrno; | |
d8ee9e8d EB |
683 | } |
684 | ||
caca86d7 | 685 | /* IpcIoPendingRequest */ |
254912f3 AR |
686 | |
687 | IpcIoPendingRequest::IpcIoPendingRequest(const IpcIoFile::Pointer &aFile): | |
ccfbe8f4 AR |
688 | file(aFile), |
689 | readRequest(nullptr), | |
690 | writeRequest(nullptr), | |
691 | codeContext(CodeContext::Current()) | |
254912f3 AR |
692 | { |
693 | } | |
694 | ||
caca86d7 | 695 | void |
8ed94021 | 696 | IpcIoPendingRequest::completeIo(IpcIoMsg *const response) |
caca86d7 | 697 | { |
caca86d7 AR |
698 | if (readRequest) |
699 | file->readCompleted(readRequest, response); | |
9199139f | 700 | else if (writeRequest) |
caca86d7 | 701 | file->writeCompleted(writeRequest, response); |
9a51593d DK |
702 | else { |
703 | Must(!response); // only timeouts are handled here | |
704 | file->openCompleted(NULL); | |
705 | } | |
caca86d7 AR |
706 | } |
707 | ||
254912f3 AR |
708 | /* XXX: disker code that should probably be moved elsewhere */ |
709 | ||
7015a149 | 710 | static SBuf DbName; ///< full db file name |
254912f3 AR |
711 | static int TheFile = -1; ///< db file descriptor |
712 | ||
9a51593d DK |
713 | static void |
714 | diskerRead(IpcIoMsg &ipcIo) | |
254912f3 | 715 | { |
551f8a18 | 716 | if (!Ipc::Mem::GetPage(Ipc::Mem::PageId::ioPage, ipcIo.page)) { |
8ed94021 | 717 | ipcIo.len = 0; |
c792f7fc | 718 | debugs(47,2, HERE << "run out of shared memory pages for IPC I/O"); |
8ed94021 DK |
719 | return; |
720 | } | |
721 | ||
722 | char *const buf = Ipc::Mem::PagePointer(ipcIo.page); | |
723 | const ssize_t read = pread(TheFile, buf, min(ipcIo.len, Ipc::Mem::PageSize()), ipcIo.offset); | |
e4f1fdae | 724 | ++statCounter.syscalls.disk.reads; |
2d338731 AR |
725 | fd_bytes(TheFile, read, FD_READ); |
726 | ||
254912f3 | 727 | if (read >= 0) { |
9a51593d DK |
728 | ipcIo.xerrno = 0; |
729 | const size_t len = static_cast<size_t>(read); // safe because read > 0 | |
254912f3 | 730 | debugs(47,8, HERE << "disker" << KidIdentifier << " read " << |
9199139f | 731 | (len == ipcIo.len ? "all " : "just ") << read); |
9a51593d DK |
732 | ipcIo.len = len; |
733 | } else { | |
734 | ipcIo.xerrno = errno; | |
735 | ipcIo.len = 0; | |
254912f3 | 736 | debugs(47,5, HERE << "disker" << KidIdentifier << " read error: " << |
9199139f | 737 | ipcIo.xerrno); |
9a51593d | 738 | } |
254912f3 AR |
739 | } |
740 | ||
9d4e9cfb | 741 | /// Tries to write buffer to disk (a few times if needed); |
4e4f7fd9 | 742 | /// sets ipcIo results, but does no cleanup. The caller must cleanup. |
9a51593d | 743 | static void |
4e4f7fd9 AR |
744 | diskerWriteAttempts(IpcIoMsg &ipcIo) |
745 | { | |
746 | const char *buf = Ipc::Mem::PagePointer(ipcIo.page); | |
747 | size_t toWrite = min(ipcIo.len, Ipc::Mem::PageSize()); | |
748 | size_t wroteSoFar = 0; | |
749 | off_t offset = ipcIo.offset; | |
750 | // Partial writes to disk do happen. It is unlikely that the caller can | |
751 | // handle partial writes by doing something other than writing leftovers | |
752 | // again, so we try to write them ourselves to minimize overheads. | |
753 | const int attemptLimit = 10; | |
754 | for (int attempts = 1; attempts <= attemptLimit; ++attempts) { | |
755 | const ssize_t result = pwrite(TheFile, buf, toWrite, offset); | |
756 | ++statCounter.syscalls.disk.writes; | |
757 | fd_bytes(TheFile, result, FD_WRITE); | |
758 | ||
759 | if (result < 0) { | |
760 | ipcIo.xerrno = errno; | |
761 | assert(ipcIo.xerrno); | |
82ed74dc AR |
762 | debugs(47, DBG_IMPORTANT, "ERROR: " << DbName << " failure" << |
763 | " writing " << toWrite << '/' << ipcIo.len << | |
4e4f7fd9 AR |
764 | " at " << ipcIo.offset << '+' << wroteSoFar << |
765 | " on " << attempts << " try: " << xstrerr(ipcIo.xerrno)); | |
766 | ipcIo.len = wroteSoFar; | |
767 | return; // bail on error | |
768 | } | |
2d338731 | 769 | |
4e4f7fd9 | 770 | const size_t wroteNow = static_cast<size_t>(result); // result >= 0 |
9a51593d | 771 | ipcIo.xerrno = 0; |
4e4f7fd9 AR |
772 | |
773 | debugs(47,3, "disker" << KidIdentifier << " wrote " << | |
774 | (wroteNow >= toWrite ? "all " : "just ") << wroteNow << | |
775 | " out of " << toWrite << '/' << ipcIo.len << " at " << | |
776 | ipcIo.offset << '+' << wroteSoFar << " on " << attempts << | |
777 | " try"); | |
778 | ||
779 | wroteSoFar += wroteNow; | |
780 | ||
781 | if (wroteNow >= toWrite) { | |
782 | ipcIo.xerrno = 0; | |
783 | ipcIo.len = wroteSoFar; | |
784 | return; // wrote everything there was to write | |
785 | } | |
786 | ||
787 | buf += wroteNow; | |
788 | offset += wroteNow; | |
789 | toWrite -= wroteNow; | |
9a51593d | 790 | } |
8ed94021 | 791 | |
82ed74dc AR |
792 | debugs(47, DBG_IMPORTANT, "ERROR: " << DbName << " exhausted all " << |
793 | attemptLimit << " attempts while writing " << | |
4e4f7fd9 AR |
794 | toWrite << '/' << ipcIo.len << " at " << ipcIo.offset << '+' << |
795 | wroteSoFar); | |
796 | return; // not a fatal I/O error, unless the caller treats it as such | |
797 | } | |
798 | ||
799 | static void | |
800 | diskerWrite(IpcIoMsg &ipcIo) | |
801 | { | |
802 | diskerWriteAttempts(ipcIo); // may fail | |
8ed94021 | 803 | Ipc::Mem::PutPage(ipcIo.page); |
254912f3 AR |
804 | } |
805 | ||
c792f7fc | 806 | void |
df881a0f | 807 | IpcIoFile::DiskerHandleMoreRequests(void *source) |
c792f7fc | 808 | { |
df881a0f AR |
809 | debugs(47, 7, HERE << "resuming handling requests after " << |
810 | static_cast<const char *>(source)); | |
c792f7fc AR |
811 | DiskerHandleMoreRequestsScheduled = false; |
812 | IpcIoFile::DiskerHandleRequests(); | |
813 | } | |
814 | ||
df881a0f AR |
815 | bool |
816 | IpcIoFile::WaitBeforePop() | |
817 | { | |
6c6a656a | 818 | const int ioRate = queue->localRateLimit().load(); |
df881a0f AR |
819 | const double maxRate = ioRate/1e3; // req/ms |
820 | ||
821 | // do we need to enforce configured I/O rate? | |
822 | if (maxRate <= 0) | |
823 | return false; | |
824 | ||
825 | // is there an I/O request we could potentially delay? | |
30fe12bc | 826 | int kidId; |
1e614370 | 827 | IpcIoMsg ipcIo; |
30fe12bc | 828 | if (!queue->peek(kidId, ipcIo)) { |
1e614370 | 829 | // unlike pop(), peek() is not reliable and does not block reader |
df881a0f AR |
830 | // so we must proceed with pop() even if it is likely to fail |
831 | return false; | |
832 | } | |
833 | ||
9a585a30 | 834 | static timeval LastIo = current_time; |
df881a0f AR |
835 | |
836 | const double ioDuration = 1.0 / maxRate; // ideal distance between two I/Os | |
837 | // do not accumulate more than 100ms or 100 I/Os, whichever is smaller | |
838 | const int64_t maxImbalance = min(static_cast<int64_t>(100), static_cast<int64_t>(100 * ioDuration)); | |
839 | ||
840 | const double credit = ioDuration; // what the last I/O should have cost us | |
841 | const double debit = tvSubMsec(LastIo, current_time); // actual distance from the last I/O | |
842 | LastIo = current_time; | |
843 | ||
844 | Ipc::QueueReader::Balance &balance = queue->localBalance(); | |
845 | balance += static_cast<int64_t>(credit - debit); | |
846 | ||
847 | debugs(47, 7, HERE << "rate limiting balance: " << balance << " after +" << credit << " -" << debit); | |
848 | ||
1e614370 DK |
849 | if (ipcIo.command == IpcIo::cmdWrite && balance > maxImbalance) { |
850 | // if the next request is (likely) write and we accumulated | |
851 | // too much time for future slow I/Os, then shed accumulated | |
852 | // time to keep just half of the excess | |
df881a0f | 853 | const int64_t toSpend = balance - maxImbalance/2; |
69321ae9 AR |
854 | |
855 | if (toSpend/1e3 > Timeout) | |
82ed74dc AR |
856 | debugs(47, DBG_IMPORTANT, "WARNING: " << DbName << " delays " << |
857 | "I/O requests for " << (toSpend/1e3) << " seconds " << | |
858 | "to obey " << ioRate << "/sec rate limit"); | |
69321ae9 | 859 | |
df881a0f AR |
860 | debugs(47, 3, HERE << "rate limiting by " << toSpend << " ms to get" << |
861 | (1e3*maxRate) << "/sec rate"); | |
862 | eventAdd("IpcIoFile::DiskerHandleMoreRequests", | |
863 | &IpcIoFile::DiskerHandleMoreRequests, | |
864 | const_cast<char*>("rate limiting"), | |
865 | toSpend/1e3, 0, false); | |
866 | DiskerHandleMoreRequestsScheduled = true; | |
867 | return true; | |
e29ccb57 | 868 | } else if (balance < -maxImbalance) { |
df881a0f AR |
869 | // do not owe "too much" to avoid "too large" bursts of I/O |
870 | balance = -maxImbalance; | |
871 | } | |
872 | ||
873 | return false; | |
874 | } | |
875 | ||
254912f3 | 876 | void |
f5591061 | 877 | IpcIoFile::DiskerHandleRequests() |
254912f3 | 878 | { |
c792f7fc | 879 | // Balance our desire to maximize the number of concurrent I/O requests |
fdb3059b | 880 | // (reordred by OS to minimize seek time) with a requirement to |
c792f7fc | 881 | // send 1st-I/O notification messages, process Coordinator events, etc. |
fdb3059b AR |
882 | const int maxSpentMsec = 10; // keep small: most RAM I/Os are under 1ms |
883 | const timeval loopStart = current_time; | |
884 | ||
c792f7fc | 885 | int popped = 0; |
fa61cefe AR |
886 | int workerId = 0; |
887 | IpcIoMsg ipcIo; | |
df881a0f | 888 | while (!WaitBeforePop() && queue->pop(workerId, ipcIo)) { |
fdb3059b AR |
889 | ++popped; |
890 | ||
891 | // at least one I/O per call is guaranteed if the queue is not empty | |
fa61cefe AR |
892 | DiskerHandleRequest(workerId, ipcIo); |
893 | ||
fdb3059b AR |
894 | getCurrentTime(); |
895 | const double elapsedMsec = tvSubMsec(loopStart, current_time); | |
896 | if (elapsedMsec > maxSpentMsec || elapsedMsec < 0) { | |
897 | if (!DiskerHandleMoreRequestsScheduled) { | |
898 | // the gap must be positive for select(2) to be given a chance | |
899 | const double minBreakSecs = 0.001; | |
900 | eventAdd("IpcIoFile::DiskerHandleMoreRequests", | |
9199139f | 901 | &IpcIoFile::DiskerHandleMoreRequests, |
df881a0f AR |
902 | const_cast<char*>("long I/O loop"), |
903 | minBreakSecs, 0, false); | |
fdb3059b AR |
904 | DiskerHandleMoreRequestsScheduled = true; |
905 | } | |
906 | debugs(47, 3, HERE << "pausing after " << popped << " I/Os in " << | |
9199139f | 907 | elapsedMsec << "ms; " << (elapsedMsec/popped) << "ms per I/O"); |
fdb3059b AR |
908 | break; |
909 | } | |
c792f7fc AR |
910 | } |
911 | ||
fdb3059b | 912 | // TODO: consider using O_DIRECT with "elevator" optimization where we pop |
fa61cefe AR |
913 | // requests first, then reorder the popped requests to optimize seek time, |
914 | // then do I/O, then take a break, and come back for the next set of I/O | |
915 | // requests. | |
9a51593d | 916 | } |
254912f3 | 917 | |
9a51593d DK |
918 | /// called when disker receives an I/O request |
919 | void | |
920 | IpcIoFile::DiskerHandleRequest(const int workerId, IpcIoMsg &ipcIo) | |
921 | { | |
9a51593d | 922 | if (ipcIo.command != IpcIo::cmdRead && ipcIo.command != IpcIo::cmdWrite) { |
82ed74dc | 923 | debugs(0, DBG_CRITICAL, "ERROR: " << DbName << |
9a51593d DK |
924 | " should not receive " << ipcIo.command << |
925 | " ipcIo" << workerId << '.' << ipcIo.requestId); | |
926 | return; | |
927 | } | |
928 | ||
929 | debugs(47,5, HERE << "disker" << KidIdentifier << | |
930 | (ipcIo.command == IpcIo::cmdRead ? " reads " : " writes ") << | |
931 | ipcIo.len << " at " << ipcIo.offset << | |
932 | " ipcIo" << workerId << '.' << ipcIo.requestId); | |
933 | ||
30fe12bc EB |
934 | const auto workerPid = ipcIo.workerPid; |
935 | assert(workerPid >= 0); | |
936 | ||
9a51593d DK |
937 | if (ipcIo.command == IpcIo::cmdRead) |
938 | diskerRead(ipcIo); | |
939 | else // ipcIo.command == IpcIo::cmdWrite | |
940 | diskerWrite(ipcIo); | |
941 | ||
30fe12bc EB |
942 | assert(ipcIo.workerPid == workerPid); |
943 | ||
f5591061 | 944 | debugs(47, 7, HERE << "pushing " << SipcIo(workerId, ipcIo, KidIdentifier)); |
fa61cefe | 945 | |
7a907247 | 946 | try { |
f5591061 | 947 | if (queue->push(workerId, ipcIo)) |
5e44782e | 948 | Notify(workerId); // must notify worker |
f5591061 | 949 | } catch (const Queue::Full &) { |
387e1d45 EB |
950 | // The worker pop queue should not overflow because the worker can |
951 | // push only if pendingRequests() is less than QueueCapacity. | |
82ed74dc AR |
952 | debugs(47, DBG_IMPORTANT, "BUG: Worker I/O pop queue for " << |
953 | DbName << " overflow: " << | |
5e44782e | 954 | SipcIo(workerId, ipcIo, KidIdentifier)); // TODO: report queue len |
fa61cefe AR |
955 | |
956 | // the I/O request we could not push will timeout | |
7a907247 | 957 | } |
254912f3 AR |
958 | } |
959 | ||
960 | static bool | |
ced8def3 | 961 | DiskerOpen(const SBuf &path, int flags, mode_t) |
254912f3 AR |
962 | { |
963 | assert(TheFile < 0); | |
964 | ||
82ed74dc | 965 | DbName = path; |
7015a149 | 966 | TheFile = file_open(DbName.c_str(), flags); |
254912f3 AR |
967 | |
968 | if (TheFile < 0) { | |
969 | const int xerrno = errno; | |
82ed74dc | 970 | debugs(47, DBG_CRITICAL, "ERROR: cannot open " << DbName << ": " << |
254912f3 AR |
971 | xstrerr(xerrno)); |
972 | return false; | |
9a51593d | 973 | } |
254912f3 | 974 | |
cb4185f1 | 975 | ++store_open_disk_fd; |
7015a149 | 976 | debugs(79,3, "rock db opened " << DbName << ": FD " << TheFile); |
9a51593d | 977 | return true; |
254912f3 AR |
978 | } |
979 | ||
980 | static void | |
7015a149 | 981 | DiskerClose(const SBuf &path) |
254912f3 AR |
982 | { |
983 | if (TheFile >= 0) { | |
984 | file_close(TheFile); | |
985 | debugs(79,3, HERE << "rock db closed " << path << ": FD " << TheFile); | |
986 | TheFile = -1; | |
5e263176 | 987 | --store_open_disk_fd; |
a1c98830 | 988 | } |
7015a149 | 989 | DbName.clear(); |
254912f3 | 990 | } |
f5591061 | 991 | |
ea2cdeb6 | 992 | /// reports our needs for shared memory pages to Ipc::Mem::Pages |
21b7990f AR |
993 | /// and initializes shared memory segments used by IpcIoFile |
994 | class IpcIoRr: public Ipc::Mem::RegisteredRunner | |
ea2cdeb6 DK |
995 | { |
996 | public: | |
997 | /* RegisteredRunner API */ | |
21b7990f AR |
998 | IpcIoRr(): owner(NULL) {} |
999 | virtual ~IpcIoRr(); | |
1000 | virtual void claimMemoryNeeds(); | |
1001 | ||
1002 | protected: | |
1003 | /* Ipc::Mem::RegisteredRunner API */ | |
1004 | virtual void create(); | |
1005 | ||
1006 | private: | |
1007 | Ipc::FewToFewBiQueue::Owner *owner; | |
ea2cdeb6 DK |
1008 | }; |
1009 | ||
21b7990f | 1010 | RunnerRegistrationEntry(IpcIoRr); |
ea2cdeb6 | 1011 | |
ea2cdeb6 | 1012 | void |
21b7990f | 1013 | IpcIoRr::claimMemoryNeeds() |
ea2cdeb6 DK |
1014 | { |
1015 | const int itemsCount = Ipc::FewToFewBiQueue::MaxItemsCount( | |
c11211d9 | 1016 | ::Config.workers, ::Config.cacheSwap.n_strands, QueueCapacity); |
ea2cdeb6 DK |
1017 | // the maximum number of shared I/O pages is approximately the |
1018 | // number of queue slots, we add a fudge factor to that to account | |
1019 | // for corner cases where I/O pages are created before queue | |
1020 | // limits are checked or destroyed long after the I/O is dequeued | |
29a9ff4e DK |
1021 | Ipc::Mem::NotePageNeed(Ipc::Mem::PageId::ioPage, |
1022 | static_cast<int>(itemsCount * 1.1)); | |
ea2cdeb6 DK |
1023 | } |
1024 | ||
21b7990f AR |
1025 | void |
1026 | IpcIoRr::create() | |
f5591061 | 1027 | { |
e9f68413 | 1028 | if (Config.cacheSwap.n_strands <= 0) |
f5591061 DK |
1029 | return; |
1030 | ||
4404f1c5 | 1031 | Must(!owner); |
4404f1c5 | 1032 | owner = Ipc::FewToFewBiQueue::Init(ShmLabel, Config.workers, 1, |
3b581957 | 1033 | Config.cacheSwap.n_strands, |
4404f1c5 | 1034 | 1 + Config.workers, sizeof(IpcIoMsg), |
ea2cdeb6 | 1035 | QueueCapacity); |
f5591061 DK |
1036 | } |
1037 | ||
1038 | IpcIoRr::~IpcIoRr() | |
1039 | { | |
1040 | delete owner; | |
1041 | } | |
f53969cc | 1042 |