]>
Commit | Line | Data |
---|---|---|
254912f3 | 1 | /* |
bbc27441 AJ |
2 | * Copyright (C) 1996-2014 The Squid Software Foundation and contributors |
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" |
f5591061 | 12 | #include "base/RunnersRegistry.h" |
254912f3 | 13 | #include "base/TextException.h" |
602d9612 | 14 | #include "disk.h" |
254912f3 AR |
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" |
582c2af2 | 20 | #include "globals.h" |
21d845b1 | 21 | #include "ipc/mem/Pages.h" |
254912f3 AR |
22 | #include "ipc/Messages.h" |
23 | #include "ipc/Port.h" | |
f5591061 | 24 | #include "ipc/Queue.h" |
9a51593d | 25 | #include "ipc/StrandSearch.h" |
254912f3 | 26 | #include "ipc/UdsOp.h" |
7015a149 | 27 | #include "SBuf.h" |
4d5904f7 | 28 | #include "SquidConfig.h" |
21d845b1 FC |
29 | #include "SquidTime.h" |
30 | #include "StatCounters.h" | |
5bed43d6 | 31 | #include "tools.h" |
254912f3 | 32 | |
1a30fdf5 AJ |
33 | #include <cerrno> |
34 | ||
254912f3 AR |
35 | CBDATA_CLASS_INIT(IpcIoFile); |
36 | ||
f5591061 DK |
37 | /// shared memory segment path to use for IpcIoFile maps |
38 | static const char *const ShmLabel = "io_file"; | |
ea2cdeb6 DK |
39 | /// a single worker-to-disker or disker-to-worker queue capacity; up |
40 | /// to 2*QueueCapacity I/O requests queued between a single worker and | |
41 | /// a single disker | |
42 | // TODO: make configurable or compute from squid.conf settings if possible | |
43 | static const int QueueCapacity = 1024; | |
f5591061 | 44 | |
fa61cefe | 45 | const double IpcIoFile::Timeout = 7; // seconds; XXX: ALL,9 may require more |
b2aa0934 DK |
46 | IpcIoFile::IpcIoFileList IpcIoFile::WaitingForOpen; |
47 | IpcIoFile::IpcIoFilesMap IpcIoFile::IpcIoFiles; | |
f5591061 | 48 | std::auto_ptr<IpcIoFile::Queue> IpcIoFile::queue; |
254912f3 | 49 | |
c792f7fc AR |
50 | bool IpcIoFile::DiskerHandleMoreRequestsScheduled = false; |
51 | ||
7015a149 AR |
52 | static bool DiskerOpen(const SBuf &path, int flags, mode_t mode); |
53 | static void DiskerClose(const SBuf &path); | |
254912f3 | 54 | |
fa61cefe AR |
55 | /// IpcIo wrapper for debugs() streams; XXX: find a better class name |
56 | struct SipcIo { | |
57 | SipcIo(int aWorker, const IpcIoMsg &aMsg, int aDisker): | |
f53969cc | 58 | worker(aWorker), msg(aMsg), disker(aDisker) {} |
fa61cefe AR |
59 | |
60 | int worker; | |
61 | const IpcIoMsg &msg; | |
62 | int disker; | |
63 | }; | |
64 | ||
65 | std::ostream & | |
66 | operator <<(std::ostream &os, const SipcIo &sio) | |
67 | { | |
68 | return os << "ipcIo" << sio.worker << '.' << sio.msg.requestId << | |
9199139f | 69 | (sio.msg.command == IpcIo::cmdRead ? 'r' : 'w') << sio.disker; |
fa61cefe AR |
70 | } |
71 | ||
254912f3 | 72 | IpcIoFile::IpcIoFile(char const *aDb): |
f53969cc SM |
73 | dbName(aDb), diskId(-1), error_(false), lastRequestId(0), |
74 | olderRequests(&requestMap1), newerRequests(&requestMap2), | |
75 | timeoutCheckScheduled(false) | |
254912f3 AR |
76 | { |
77 | } | |
78 | ||
79 | IpcIoFile::~IpcIoFile() | |
80 | { | |
b2aa0934 DK |
81 | if (diskId >= 0) { |
82 | const IpcIoFilesMap::iterator i = IpcIoFiles.find(diskId); | |
83 | // XXX: warn and continue? | |
84 | Must(i != IpcIoFiles.end()); | |
85 | Must(i->second == this); | |
86 | IpcIoFiles.erase(i); | |
87 | } | |
254912f3 AR |
88 | } |
89 | ||
43ebbac3 AR |
90 | void |
91 | IpcIoFile::configure(const Config &cfg) | |
92 | { | |
93 | DiskFile::configure(cfg); | |
94 | config = cfg; | |
95 | } | |
96 | ||
254912f3 AR |
97 | void |
98 | IpcIoFile::open(int flags, mode_t mode, RefCount<IORequestor> callback) | |
99 | { | |
100 | ioRequestor = callback; | |
101 | Must(diskId < 0); // we do not know our disker yet | |
f5591061 DK |
102 | |
103 | if (!queue.get()) | |
104 | queue.reset(new Queue(ShmLabel, IamWorkerProcess() ? Queue::groupA : Queue::groupB, KidIdentifier)); | |
254912f3 AR |
105 | |
106 | if (IamDiskProcess()) { | |
7015a149 | 107 | error_ = !DiskerOpen(SBuf(dbName.termedBuf()), flags, mode); |
f7091279 DK |
108 | if (error_) |
109 | return; | |
110 | ||
b2aa0934 DK |
111 | diskId = KidIdentifier; |
112 | const bool inserted = | |
113 | IpcIoFiles.insert(std::make_pair(diskId, this)).second; | |
114 | Must(inserted); | |
9a51593d | 115 | |
df881a0f AR |
116 | queue->localRateLimit() = |
117 | static_cast<Ipc::QueueReader::Rate::Value>(config.ioRate); | |
118 | ||
f7091279 DK |
119 | Ipc::HereIamMessage ann(Ipc::StrandCoord(KidIdentifier, getpid())); |
120 | ann.strand.tag = dbName; | |
121 | Ipc::TypedMsgHdr message; | |
122 | ann.pack(message); | |
1ee292b7 | 123 | SendMessage(Ipc::Port::CoordinatorAddr(), message); |
f7091279 | 124 | |
e528e570 | 125 | ioRequestor->ioCompletedNotification(); |
254912f3 | 126 | return; |
9a51593d | 127 | } |
254912f3 | 128 | |
9a51593d DK |
129 | Ipc::StrandSearchRequest request; |
130 | request.requestorId = KidIdentifier; | |
131 | request.tag = dbName; | |
254912f3 | 132 | |
9a51593d DK |
133 | Ipc::TypedMsgHdr msg; |
134 | request.pack(msg); | |
1ee292b7 | 135 | Ipc::SendMessage(Ipc::Port::CoordinatorAddr(), msg); |
9a51593d | 136 | |
b2aa0934 DK |
137 | WaitingForOpen.push_back(this); |
138 | ||
139 | eventAdd("IpcIoFile::OpenTimeout", &IpcIoFile::OpenTimeout, | |
140 | this, Timeout, 0, false); // "this" pointer is used as id | |
254912f3 AR |
141 | } |
142 | ||
143 | void | |
9199139f AR |
144 | IpcIoFile::openCompleted(const Ipc::StrandSearchResponse *const response) |
145 | { | |
9a51593d | 146 | Must(diskId < 0); // we do not know our disker yet |
9a51593d DK |
147 | |
148 | if (!response) { | |
82ed74dc AR |
149 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " communication " << |
150 | "channel establishment timeout"); | |
caca86d7 | 151 | error_ = true; |
9a51593d DK |
152 | } else { |
153 | diskId = response->strand.kidId; | |
154 | if (diskId >= 0) { | |
b2aa0934 DK |
155 | const bool inserted = |
156 | IpcIoFiles.insert(std::make_pair(diskId, this)).second; | |
157 | Must(inserted); | |
9a51593d | 158 | } else { |
254912f3 | 159 | error_ = true; |
82ed74dc AR |
160 | debugs(79, DBG_IMPORTANT, "ERROR: no disker claimed " << |
161 | "responsibility for " << dbName); | |
9a51593d DK |
162 | } |
163 | } | |
254912f3 AR |
164 | |
165 | ioRequestor->ioCompletedNotification(); | |
166 | } | |
167 | ||
168 | /** | |
169 | * Alias for IpcIoFile::open(...) | |
170 | \copydoc IpcIoFile::open(int flags, mode_t mode, RefCount<IORequestor> callback) | |
171 | */ | |
172 | void | |
173 | IpcIoFile::create(int flags, mode_t mode, RefCount<IORequestor> callback) | |
174 | { | |
175 | assert(false); // check | |
176 | /* We use the same logic path for open */ | |
177 | open(flags, mode, callback); | |
178 | } | |
179 | ||
180 | void | |
181 | IpcIoFile::close() | |
182 | { | |
183 | assert(ioRequestor != NULL); | |
184 | ||
185 | if (IamDiskProcess()) | |
7015a149 | 186 | DiskerClose(SBuf(dbName.termedBuf())); |
9a51593d | 187 | // XXX: else nothing to do? |
254912f3 AR |
188 | |
189 | ioRequestor->closeCompleted(); | |
190 | } | |
191 | ||
192 | bool | |
193 | IpcIoFile::canRead() const | |
194 | { | |
7e2d5ef6 | 195 | return diskId >= 0 && !error_ && canWait(); |
254912f3 AR |
196 | } |
197 | ||
198 | bool | |
199 | IpcIoFile::canWrite() const | |
200 | { | |
7e2d5ef6 | 201 | return diskId >= 0 && !error_ && canWait(); |
254912f3 AR |
202 | } |
203 | ||
204 | bool | |
205 | IpcIoFile::error() const | |
206 | { | |
207 | return error_; | |
208 | } | |
209 | ||
210 | void | |
211 | IpcIoFile::read(ReadRequest *readRequest) | |
212 | { | |
213 | debugs(79,3, HERE << "(disker" << diskId << ", " << readRequest->len << ", " << | |
9199139f | 214 | readRequest->offset << ")"); |
254912f3 AR |
215 | |
216 | assert(ioRequestor != NULL); | |
254912f3 AR |
217 | assert(readRequest->offset >= 0); |
218 | Must(!error_); | |
219 | ||
220 | //assert(minOffset < 0 || minOffset <= readRequest->offset); | |
221 | //assert(maxOffset < 0 || readRequest->offset + readRequest->len <= (uint64_t)maxOffset); | |
222 | ||
9a51593d | 223 | IpcIoPendingRequest *const pending = new IpcIoPendingRequest(this); |
254912f3 | 224 | pending->readRequest = readRequest; |
7a907247 | 225 | push(pending); |
254912f3 AR |
226 | } |
227 | ||
228 | void | |
229 | IpcIoFile::readCompleted(ReadRequest *readRequest, | |
8ed94021 | 230 | IpcIoMsg *const response) |
254912f3 | 231 | { |
caca86d7 | 232 | bool ioError = false; |
9a51593d | 233 | if (!response) { |
0ef0509e | 234 | debugs(79, 3, HERE << "error: timeout"); |
caca86d7 | 235 | ioError = true; // I/O timeout does not warrant setting error_? |
9a51593d | 236 | } else { |
70e3f706 | 237 | if (response->xerrno) { |
82ed74dc AR |
238 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " read: " << |
239 | xstrerr(response->xerrno)); | |
70e3f706 | 240 | ioError = error_ = true; |
9199139f | 241 | } else if (!response->page) { |
82ed74dc AR |
242 | debugs(79, DBG_IMPORTANT, "ERROR: " << dbName << " read ran " << |
243 | "out of shared memory pages"); | |
70e3f706 DK |
244 | ioError = true; |
245 | } else { | |
246 | const char *const buf = Ipc::Mem::PagePointer(response->page); | |
247 | memcpy(readRequest->buf, buf, response->len); | |
248 | } | |
254912f3 | 249 | |
70e3f706 DK |
250 | Ipc::Mem::PutPage(response->page); |
251 | } | |
8ed94021 | 252 | |
caca86d7 | 253 | const ssize_t rlen = ioError ? -1 : (ssize_t)readRequest->len; |
9199139f | 254 | const int errflag = ioError ? DISK_ERROR :DISK_OK; |
20b0e1fe | 255 | ioRequestor->readCompleted(readRequest->buf, rlen, errflag, readRequest); |
254912f3 AR |
256 | } |
257 | ||
258 | void | |
259 | IpcIoFile::write(WriteRequest *writeRequest) | |
260 | { | |
261 | debugs(79,3, HERE << "(disker" << diskId << ", " << writeRequest->len << ", " << | |
9199139f | 262 | writeRequest->offset << ")"); |
254912f3 AR |
263 | |
264 | assert(ioRequestor != NULL); | |
254912f3 AR |
265 | assert(writeRequest->len > 0); // TODO: work around mmap failures on zero-len? |
266 | assert(writeRequest->offset >= 0); | |
267 | Must(!error_); | |
268 | ||
269 | //assert(minOffset < 0 || minOffset <= writeRequest->offset); | |
270 | //assert(maxOffset < 0 || writeRequest->offset + writeRequest->len <= (uint64_t)maxOffset); | |
271 | ||
9a51593d | 272 | IpcIoPendingRequest *const pending = new IpcIoPendingRequest(this); |
254912f3 | 273 | pending->writeRequest = writeRequest; |
7a907247 | 274 | push(pending); |
254912f3 AR |
275 | } |
276 | ||
277 | void | |
278 | IpcIoFile::writeCompleted(WriteRequest *writeRequest, | |
9a51593d | 279 | const IpcIoMsg *const response) |
254912f3 | 280 | { |
caca86d7 | 281 | bool ioError = false; |
9a51593d | 282 | if (!response) { |
4e4f7fd9 | 283 | debugs(79, 3, "disker " << diskId << " timeout"); |
caca86d7 | 284 | ioError = true; // I/O timeout does not warrant setting error_? |
9199139f | 285 | } else if (response->xerrno) { |
4e4f7fd9 AR |
286 | debugs(79, DBG_IMPORTANT, "ERROR: disker " << diskId << |
287 | " error writing " << writeRequest->len << " bytes at " << | |
288 | writeRequest->offset << ": " << xstrerr(response->xerrno) << | |
289 | "; this worker will stop using " << dbName); | |
9a51593d | 290 | ioError = error_ = true; |
9199139f | 291 | } else if (response->len != writeRequest->len) { |
4e4f7fd9 AR |
292 | debugs(79, DBG_IMPORTANT, "ERROR: disker " << diskId << " wrote " << |
293 | response->len << " instead of " << writeRequest->len << | |
294 | " bytes (offset " << writeRequest->offset << "); " << | |
295 | "this worker will stop using " << dbName); | |
254912f3 AR |
296 | error_ = true; |
297 | } | |
298 | ||
299 | if (writeRequest->free_func) | |
300 | (writeRequest->free_func)(const_cast<char*>(writeRequest->buf)); // broken API? | |
301 | ||
caca86d7 | 302 | if (!ioError) { |
254912f3 | 303 | debugs(79,5, HERE << "wrote " << writeRequest->len << " to disker" << |
9199139f AR |
304 | diskId << " at " << writeRequest->offset); |
305 | } | |
254912f3 | 306 | |
caca86d7 | 307 | const ssize_t rlen = ioError ? 0 : (ssize_t)writeRequest->len; |
9199139f | 308 | const int errflag = ioError ? DISK_ERROR :DISK_OK; |
254912f3 AR |
309 | ioRequestor->writeCompleted(errflag, rlen, writeRequest); |
310 | } | |
311 | ||
312 | bool | |
313 | IpcIoFile::ioInProgress() const | |
314 | { | |
9a51593d | 315 | return !olderRequests->empty() || !newerRequests->empty(); |
254912f3 AR |
316 | } |
317 | ||
9a51593d | 318 | /// track a new pending request |
254912f3 | 319 | void |
28bd45ba | 320 | IpcIoFile::trackPendingRequest(const unsigned int id, IpcIoPendingRequest *const pending) |
254912f3 | 321 | { |
28bd45ba AR |
322 | const std::pair<RequestMap::iterator,bool> result = |
323 | newerRequests->insert(std::make_pair(id, pending)); | |
324 | Must(result.second); // failures means that id was not unique | |
9a51593d DK |
325 | if (!timeoutCheckScheduled) |
326 | scheduleTimeoutCheck(); | |
327 | } | |
254912f3 | 328 | |
9a51593d DK |
329 | /// push an I/O request to disker |
330 | void | |
7a907247 | 331 | IpcIoFile::push(IpcIoPendingRequest *const pending) |
9a51593d | 332 | { |
fa61cefe | 333 | // prevent queue overflows: check for responses to earlier requests |
28bd45ba | 334 | // warning: this call may result in indirect push() recursion |
f5591061 | 335 | HandleResponses("before push"); |
fa61cefe | 336 | |
a1c98830 | 337 | debugs(47, 7, HERE); |
9a51593d | 338 | Must(diskId >= 0); |
7a907247 DK |
339 | Must(pending); |
340 | Must(pending->readRequest || pending->writeRequest); | |
9a51593d DK |
341 | |
342 | IpcIoMsg ipcIo; | |
8ed94021 | 343 | try { |
28bd45ba AR |
344 | if (++lastRequestId == 0) // don't use zero value as requestId |
345 | ++lastRequestId; | |
8ed94021 | 346 | ipcIo.requestId = lastRequestId; |
0a11e039 | 347 | ipcIo.start = current_time; |
8ed94021 DK |
348 | if (pending->readRequest) { |
349 | ipcIo.command = IpcIo::cmdRead; | |
350 | ipcIo.offset = pending->readRequest->offset; | |
351 | ipcIo.len = pending->readRequest->len; | |
352 | } else { // pending->writeRequest | |
353 | Must(pending->writeRequest->len <= Ipc::Mem::PageSize()); | |
551f8a18 | 354 | if (!Ipc::Mem::GetPage(Ipc::Mem::PageId::ioPage, ipcIo.page)) { |
8ed94021 | 355 | ipcIo.len = 0; |
551f8a18 | 356 | throw TexcHere("run out of shared memory pages for IPC I/O"); |
8ed94021 DK |
357 | } |
358 | ipcIo.command = IpcIo::cmdWrite; | |
359 | ipcIo.offset = pending->writeRequest->offset; | |
360 | ipcIo.len = pending->writeRequest->len; | |
361 | char *const buf = Ipc::Mem::PagePointer(ipcIo.page); | |
362 | memcpy(buf, pending->writeRequest->buf, ipcIo.len); // optimize away | |
363 | } | |
254912f3 | 364 | |
f5591061 | 365 | debugs(47, 7, HERE << "pushing " << SipcIo(KidIdentifier, ipcIo, diskId)); |
fa61cefe | 366 | |
f5591061 | 367 | if (queue->push(diskId, ipcIo)) |
fa61cefe | 368 | Notify(diskId); // must notify disker |
28bd45ba | 369 | trackPendingRequest(ipcIo.requestId, pending); |
f5591061 | 370 | } catch (const Queue::Full &) { |
82ed74dc AR |
371 | debugs(47, DBG_IMPORTANT, "ERROR: worker I/O push queue for " << |
372 | dbName << " overflow: " << | |
fa61cefe AR |
373 | SipcIo(KidIdentifier, ipcIo, diskId)); // TODO: report queue len |
374 | // TODO: grow queue size | |
9199139f | 375 | |
0ef0509e | 376 | pending->completeIo(NULL); |
8ed94021 DK |
377 | delete pending; |
378 | } catch (const TextException &e) { | |
82ed74dc | 379 | debugs(47, DBG_IMPORTANT, "ERROR: " << dbName << " exception: " << e.what()); |
0ef0509e | 380 | pending->completeIo(NULL); |
7a907247 | 381 | delete pending; |
9a51593d | 382 | } |
254912f3 AR |
383 | } |
384 | ||
0a11e039 AR |
385 | /// whether we think there is enough time to complete the I/O |
386 | bool | |
9199139f AR |
387 | IpcIoFile::canWait() const |
388 | { | |
43ebbac3 | 389 | if (!config.ioTimeout) |
0a11e039 AR |
390 | return true; // no timeout specified |
391 | ||
392 | IpcIoMsg oldestIo; | |
5aac671b | 393 | if (!queue->findOldest(diskId, oldestIo) || oldestIo.start.tv_sec <= 0) |
0a11e039 AR |
394 | return true; // we cannot estimate expected wait time; assume it is OK |
395 | ||
55939a01 AR |
396 | const int oldestWait = tvSubMsec(oldestIo.start, current_time); |
397 | ||
398 | int rateWait = -1; // time in millisecons | |
399 | const Ipc::QueueReader::Rate::Value ioRate = queue->rateLimit(diskId); | |
400 | if (ioRate > 0) { | |
401 | // if there are N requests pending, the new one will wait at | |
402 | // least N/max-swap-rate seconds | |
75017bc9 | 403 | rateWait = static_cast<int>(1e3 * queue->outSize(diskId) / ioRate); |
55939a01 AR |
404 | // adjust N/max-swap-rate value based on the queue "balance" |
405 | // member, in case we have been borrowing time against future | |
406 | // I/O already | |
407 | rateWait += queue->balance(diskId); | |
408 | } | |
409 | ||
410 | const int expectedWait = max(oldestWait, rateWait); | |
0a11e039 | 411 | if (expectedWait < 0 || |
43ebbac3 | 412 | static_cast<time_msec_t>(expectedWait) < config.ioTimeout) |
0a11e039 AR |
413 | return true; // expected wait time is acceptible |
414 | ||
c792f7fc AR |
415 | debugs(47,2, HERE << "cannot wait: " << expectedWait << |
416 | " oldest: " << SipcIo(KidIdentifier, oldestIo, diskId)); | |
0a11e039 AR |
417 | return false; // do not want to wait that long |
418 | } | |
419 | ||
9a51593d | 420 | /// called when coordinator responds to worker open request |
254912f3 | 421 | void |
9a51593d | 422 | IpcIoFile::HandleOpenResponse(const Ipc::StrandSearchResponse &response) |
254912f3 | 423 | { |
9a51593d | 424 | debugs(47, 7, HERE << "coordinator response to open request"); |
b2aa0934 | 425 | for (IpcIoFileList::iterator i = WaitingForOpen.begin(); |
9199139f | 426 | i != WaitingForOpen.end(); ++i) { |
b2aa0934 DK |
427 | if (response.strand.tag == (*i)->dbName) { |
428 | (*i)->openCompleted(&response); | |
429 | WaitingForOpen.erase(i); | |
430 | return; | |
431 | } | |
432 | } | |
433 | ||
434 | debugs(47, 4, HERE << "LATE disker response to open for " << | |
435 | response.strand.tag); | |
436 | // nothing we can do about it; completeIo() has been called already | |
9a51593d | 437 | } |
254912f3 | 438 | |
9a51593d | 439 | void |
f5591061 | 440 | IpcIoFile::HandleResponses(const char *const when) |
fa61cefe AR |
441 | { |
442 | debugs(47, 4, HERE << "popping all " << when); | |
fa61cefe AR |
443 | IpcIoMsg ipcIo; |
444 | // get all responses we can: since we are not pushing, this will stop | |
f5591061 DK |
445 | int diskId; |
446 | while (queue->pop(diskId, ipcIo)) { | |
447 | const IpcIoFilesMap::const_iterator i = IpcIoFiles.find(diskId); | |
448 | Must(i != IpcIoFiles.end()); // TODO: warn but continue | |
449 | i->second->handleResponse(ipcIo); | |
450 | } | |
9a51593d | 451 | } |
254912f3 | 452 | |
9a51593d | 453 | void |
8ed94021 | 454 | IpcIoFile::handleResponse(IpcIoMsg &ipcIo) |
9a51593d DK |
455 | { |
456 | const int requestId = ipcIo.requestId; | |
fa61cefe | 457 | debugs(47, 7, HERE << "popped disker response: " << |
9199139f | 458 | SipcIo(KidIdentifier, ipcIo, diskId)); |
fa61cefe | 459 | |
9a51593d DK |
460 | Must(requestId); |
461 | if (IpcIoPendingRequest *const pending = dequeueRequest(requestId)) { | |
462 | pending->completeIo(&ipcIo); | |
caca86d7 AR |
463 | delete pending; // XXX: leaking if throwing |
464 | } else { | |
9a51593d DK |
465 | debugs(47, 4, HERE << "LATE disker response to " << ipcIo.command << |
466 | "; ipcIo" << KidIdentifier << '.' << requestId); | |
caca86d7 AR |
467 | // nothing we can do about it; completeIo() has been called already |
468 | } | |
254912f3 AR |
469 | } |
470 | ||
254912f3 | 471 | void |
9a51593d | 472 | IpcIoFile::Notify(const int peerId) |
254912f3 | 473 | { |
fa61cefe | 474 | // TODO: Count and report the total number of notifications, pops, pushes. |
a1c98830 | 475 | debugs(47, 7, HERE << "kid" << peerId); |
9a51593d DK |
476 | Ipc::TypedMsgHdr msg; |
477 | msg.setType(Ipc::mtIpcIoNotification); // TODO: add proper message type? | |
478 | msg.putInt(KidIdentifier); | |
1ee292b7 | 479 | const String addr = Ipc::Port::MakeAddr(Ipc::strandAddrLabel, peerId); |
9a51593d DK |
480 | Ipc::SendMessage(addr, msg); |
481 | } | |
482 | ||
483 | void | |
484 | IpcIoFile::HandleNotification(const Ipc::TypedMsgHdr &msg) | |
485 | { | |
fa61cefe AR |
486 | const int from = msg.getInt(); |
487 | debugs(47, 7, HERE << "from " << from); | |
f5591061 DK |
488 | queue->clearReaderSignal(from); |
489 | if (IamDiskProcess()) | |
490 | DiskerHandleRequests(); | |
491 | else | |
492 | HandleResponses("after notification"); | |
9a51593d DK |
493 | } |
494 | ||
b2aa0934 DK |
495 | /// handles open request timeout |
496 | void | |
497 | IpcIoFile::OpenTimeout(void *const param) | |
498 | { | |
499 | Must(param); | |
500 | // the pointer is used for comparison only and not dereferenced | |
501 | const IpcIoFile *const ipcIoFile = | |
502 | reinterpret_cast<const IpcIoFile *>(param); | |
503 | for (IpcIoFileList::iterator i = WaitingForOpen.begin(); | |
9199139f | 504 | i != WaitingForOpen.end(); ++i) { |
b2aa0934 DK |
505 | if (*i == ipcIoFile) { |
506 | (*i)->openCompleted(NULL); | |
507 | WaitingForOpen.erase(i); | |
508 | break; | |
509 | } | |
510 | } | |
511 | } | |
512 | ||
9a51593d DK |
513 | /// IpcIoFile::checkTimeouts wrapper |
514 | void | |
515 | IpcIoFile::CheckTimeouts(void *const param) | |
516 | { | |
9a51593d | 517 | Must(param); |
b2aa0934 DK |
518 | const int diskId = reinterpret_cast<uintptr_t>(param); |
519 | debugs(47, 7, HERE << "diskId=" << diskId); | |
520 | const IpcIoFilesMap::const_iterator i = IpcIoFiles.find(diskId); | |
521 | if (i != IpcIoFiles.end()) | |
522 | i->second->checkTimeouts(); | |
9a51593d DK |
523 | } |
524 | ||
525 | void | |
526 | IpcIoFile::checkTimeouts() | |
527 | { | |
9a51593d | 528 | timeoutCheckScheduled = false; |
254912f3 | 529 | |
0ef0509e AR |
530 | // last chance to recover in case a notification message was lost, etc. |
531 | const RequestMap::size_type timeoutsBefore = olderRequests->size(); | |
532 | HandleResponses("before timeout"); | |
533 | const RequestMap::size_type timeoutsNow = olderRequests->size(); | |
534 | ||
535 | if (timeoutsBefore > timeoutsNow) { // some requests were rescued | |
536 | // notification message lost or significantly delayed? | |
82ed74dc AR |
537 | debugs(47, DBG_IMPORTANT, "WARNING: communication with " << dbName << |
538 | " may be too slow or disrupted for about " << | |
0ef0509e AR |
539 | Timeout << "s; rescued " << (timeoutsBefore - timeoutsNow) << |
540 | " out of " << timeoutsBefore << " I/Os"); | |
541 | } | |
542 | ||
543 | if (timeoutsNow) { | |
544 | debugs(47, DBG_IMPORTANT, "WARNING: abandoning " << | |
82ed74dc | 545 | timeoutsNow << ' ' << dbName << " I/Os after at least " << |
0ef0509e AR |
546 | Timeout << "s timeout"); |
547 | } | |
548 | ||
caca86d7 AR |
549 | // any old request would have timed out by now |
550 | typedef RequestMap::const_iterator RMCI; | |
9a51593d DK |
551 | for (RMCI i = olderRequests->begin(); i != olderRequests->end(); ++i) { |
552 | IpcIoPendingRequest *const pending = i->second; | |
254912f3 | 553 | |
b2aa0934 | 554 | const unsigned int requestId = i->first; |
caca86d7 | 555 | debugs(47, 7, HERE << "disker timeout; ipcIo" << |
b2aa0934 | 556 | KidIdentifier << '.' << requestId); |
254912f3 | 557 | |
caca86d7 AR |
558 | pending->completeIo(NULL); // no response |
559 | delete pending; // XXX: leaking if throwing | |
9a51593d DK |
560 | } |
561 | olderRequests->clear(); | |
caca86d7 | 562 | |
9a51593d | 563 | swap(olderRequests, newerRequests); // switches pointers around |
0ef0509e | 564 | if (!olderRequests->empty() && !timeoutCheckScheduled) |
9a51593d | 565 | scheduleTimeoutCheck(); |
254912f3 AR |
566 | } |
567 | ||
caca86d7 | 568 | /// prepare to check for timeouts in a little while |
254912f3 | 569 | void |
9a51593d | 570 | IpcIoFile::scheduleTimeoutCheck() |
254912f3 | 571 | { |
b2aa0934 | 572 | // we check all older requests at once so some may be wait for 2*Timeout |
caca86d7 | 573 | eventAdd("IpcIoFile::CheckTimeouts", &IpcIoFile::CheckTimeouts, |
b2aa0934 | 574 | reinterpret_cast<void *>(diskId), Timeout, 0, false); |
9a51593d | 575 | timeoutCheckScheduled = true; |
254912f3 AR |
576 | } |
577 | ||
578 | /// returns and forgets the right IpcIoFile pending request | |
579 | IpcIoPendingRequest * | |
9a51593d | 580 | IpcIoFile::dequeueRequest(const unsigned int requestId) |
254912f3 | 581 | { |
254912f3 | 582 | Must(requestId != 0); |
caca86d7 | 583 | |
9a51593d DK |
584 | RequestMap *map = NULL; |
585 | RequestMap::iterator i = requestMap1.find(requestId); | |
caca86d7 | 586 | |
9a51593d DK |
587 | if (i != requestMap1.end()) |
588 | map = &requestMap1; | |
caca86d7 | 589 | else { |
9a51593d DK |
590 | i = requestMap2.find(requestId); |
591 | if (i != requestMap2.end()) | |
592 | map = &requestMap2; | |
caca86d7 AR |
593 | } |
594 | ||
595 | if (!map) // not found in both maps | |
596 | return NULL; | |
597 | ||
598 | IpcIoPendingRequest *pending = i->second; | |
599 | map->erase(i); | |
600 | return pending; | |
254912f3 AR |
601 | } |
602 | ||
603 | int | |
9a51593d | 604 | IpcIoFile::getFD() const |
254912f3 AR |
605 | { |
606 | assert(false); // not supported; TODO: remove this method from API | |
607 | return -1; | |
608 | } | |
609 | ||
9a51593d | 610 | /* IpcIoMsg */ |
254912f3 | 611 | |
9a51593d | 612 | IpcIoMsg::IpcIoMsg(): |
f53969cc SM |
613 | requestId(0), |
614 | offset(0), | |
615 | len(0), | |
616 | command(IpcIo::cmdNone), | |
617 | xerrno(0) | |
254912f3 | 618 | { |
0a11e039 | 619 | start.tv_sec = 0; |
e19994df | 620 | start.tv_usec = 0; |
254912f3 AR |
621 | } |
622 | ||
caca86d7 | 623 | /* IpcIoPendingRequest */ |
254912f3 AR |
624 | |
625 | IpcIoPendingRequest::IpcIoPendingRequest(const IpcIoFile::Pointer &aFile): | |
f53969cc | 626 | file(aFile), readRequest(NULL), writeRequest(NULL) |
254912f3 AR |
627 | { |
628 | } | |
629 | ||
caca86d7 | 630 | void |
8ed94021 | 631 | IpcIoPendingRequest::completeIo(IpcIoMsg *const response) |
caca86d7 | 632 | { |
caca86d7 AR |
633 | if (readRequest) |
634 | file->readCompleted(readRequest, response); | |
9199139f | 635 | else if (writeRequest) |
caca86d7 | 636 | file->writeCompleted(writeRequest, response); |
9a51593d DK |
637 | else { |
638 | Must(!response); // only timeouts are handled here | |
639 | file->openCompleted(NULL); | |
640 | } | |
caca86d7 AR |
641 | } |
642 | ||
254912f3 AR |
643 | /* XXX: disker code that should probably be moved elsewhere */ |
644 | ||
7015a149 | 645 | static SBuf DbName; ///< full db file name |
254912f3 AR |
646 | static int TheFile = -1; ///< db file descriptor |
647 | ||
9a51593d DK |
648 | static void |
649 | diskerRead(IpcIoMsg &ipcIo) | |
254912f3 | 650 | { |
551f8a18 | 651 | if (!Ipc::Mem::GetPage(Ipc::Mem::PageId::ioPage, ipcIo.page)) { |
8ed94021 | 652 | ipcIo.len = 0; |
c792f7fc | 653 | debugs(47,2, HERE << "run out of shared memory pages for IPC I/O"); |
8ed94021 DK |
654 | return; |
655 | } | |
656 | ||
657 | char *const buf = Ipc::Mem::PagePointer(ipcIo.page); | |
658 | const ssize_t read = pread(TheFile, buf, min(ipcIo.len, Ipc::Mem::PageSize()), ipcIo.offset); | |
e4f1fdae | 659 | ++statCounter.syscalls.disk.reads; |
2d338731 AR |
660 | fd_bytes(TheFile, read, FD_READ); |
661 | ||
254912f3 | 662 | if (read >= 0) { |
9a51593d DK |
663 | ipcIo.xerrno = 0; |
664 | const size_t len = static_cast<size_t>(read); // safe because read > 0 | |
254912f3 | 665 | debugs(47,8, HERE << "disker" << KidIdentifier << " read " << |
9199139f | 666 | (len == ipcIo.len ? "all " : "just ") << read); |
9a51593d DK |
667 | ipcIo.len = len; |
668 | } else { | |
669 | ipcIo.xerrno = errno; | |
670 | ipcIo.len = 0; | |
254912f3 | 671 | debugs(47,5, HERE << "disker" << KidIdentifier << " read error: " << |
9199139f | 672 | ipcIo.xerrno); |
9a51593d | 673 | } |
254912f3 AR |
674 | } |
675 | ||
9d4e9cfb | 676 | /// Tries to write buffer to disk (a few times if needed); |
4e4f7fd9 | 677 | /// sets ipcIo results, but does no cleanup. The caller must cleanup. |
9a51593d | 678 | static void |
4e4f7fd9 AR |
679 | diskerWriteAttempts(IpcIoMsg &ipcIo) |
680 | { | |
681 | const char *buf = Ipc::Mem::PagePointer(ipcIo.page); | |
682 | size_t toWrite = min(ipcIo.len, Ipc::Mem::PageSize()); | |
683 | size_t wroteSoFar = 0; | |
684 | off_t offset = ipcIo.offset; | |
685 | // Partial writes to disk do happen. It is unlikely that the caller can | |
686 | // handle partial writes by doing something other than writing leftovers | |
687 | // again, so we try to write them ourselves to minimize overheads. | |
688 | const int attemptLimit = 10; | |
689 | for (int attempts = 1; attempts <= attemptLimit; ++attempts) { | |
690 | const ssize_t result = pwrite(TheFile, buf, toWrite, offset); | |
691 | ++statCounter.syscalls.disk.writes; | |
692 | fd_bytes(TheFile, result, FD_WRITE); | |
693 | ||
694 | if (result < 0) { | |
695 | ipcIo.xerrno = errno; | |
696 | assert(ipcIo.xerrno); | |
82ed74dc AR |
697 | debugs(47, DBG_IMPORTANT, "ERROR: " << DbName << " failure" << |
698 | " writing " << toWrite << '/' << ipcIo.len << | |
4e4f7fd9 AR |
699 | " at " << ipcIo.offset << '+' << wroteSoFar << |
700 | " on " << attempts << " try: " << xstrerr(ipcIo.xerrno)); | |
701 | ipcIo.len = wroteSoFar; | |
702 | return; // bail on error | |
703 | } | |
2d338731 | 704 | |
4e4f7fd9 | 705 | const size_t wroteNow = static_cast<size_t>(result); // result >= 0 |
9a51593d | 706 | ipcIo.xerrno = 0; |
4e4f7fd9 AR |
707 | |
708 | debugs(47,3, "disker" << KidIdentifier << " wrote " << | |
709 | (wroteNow >= toWrite ? "all " : "just ") << wroteNow << | |
710 | " out of " << toWrite << '/' << ipcIo.len << " at " << | |
711 | ipcIo.offset << '+' << wroteSoFar << " on " << attempts << | |
712 | " try"); | |
713 | ||
714 | wroteSoFar += wroteNow; | |
715 | ||
716 | if (wroteNow >= toWrite) { | |
717 | ipcIo.xerrno = 0; | |
718 | ipcIo.len = wroteSoFar; | |
719 | return; // wrote everything there was to write | |
720 | } | |
721 | ||
722 | buf += wroteNow; | |
723 | offset += wroteNow; | |
724 | toWrite -= wroteNow; | |
9a51593d | 725 | } |
8ed94021 | 726 | |
82ed74dc AR |
727 | debugs(47, DBG_IMPORTANT, "ERROR: " << DbName << " exhausted all " << |
728 | attemptLimit << " attempts while writing " << | |
4e4f7fd9 AR |
729 | toWrite << '/' << ipcIo.len << " at " << ipcIo.offset << '+' << |
730 | wroteSoFar); | |
731 | return; // not a fatal I/O error, unless the caller treats it as such | |
732 | } | |
733 | ||
734 | static void | |
735 | diskerWrite(IpcIoMsg &ipcIo) | |
736 | { | |
737 | diskerWriteAttempts(ipcIo); // may fail | |
8ed94021 | 738 | Ipc::Mem::PutPage(ipcIo.page); |
254912f3 AR |
739 | } |
740 | ||
c792f7fc | 741 | void |
df881a0f | 742 | IpcIoFile::DiskerHandleMoreRequests(void *source) |
c792f7fc | 743 | { |
df881a0f AR |
744 | debugs(47, 7, HERE << "resuming handling requests after " << |
745 | static_cast<const char *>(source)); | |
c792f7fc AR |
746 | DiskerHandleMoreRequestsScheduled = false; |
747 | IpcIoFile::DiskerHandleRequests(); | |
748 | } | |
749 | ||
df881a0f AR |
750 | bool |
751 | IpcIoFile::WaitBeforePop() | |
752 | { | |
753 | const Ipc::QueueReader::Rate::Value ioRate = queue->localRateLimit(); | |
754 | const double maxRate = ioRate/1e3; // req/ms | |
755 | ||
756 | // do we need to enforce configured I/O rate? | |
757 | if (maxRate <= 0) | |
758 | return false; | |
759 | ||
760 | // is there an I/O request we could potentially delay? | |
1e614370 DK |
761 | int processId; |
762 | IpcIoMsg ipcIo; | |
763 | if (!queue->peek(processId, ipcIo)) { | |
764 | // unlike pop(), peek() is not reliable and does not block reader | |
df881a0f AR |
765 | // so we must proceed with pop() even if it is likely to fail |
766 | return false; | |
767 | } | |
768 | ||
9a585a30 | 769 | static timeval LastIo = current_time; |
df881a0f AR |
770 | |
771 | const double ioDuration = 1.0 / maxRate; // ideal distance between two I/Os | |
772 | // do not accumulate more than 100ms or 100 I/Os, whichever is smaller | |
773 | const int64_t maxImbalance = min(static_cast<int64_t>(100), static_cast<int64_t>(100 * ioDuration)); | |
774 | ||
775 | const double credit = ioDuration; // what the last I/O should have cost us | |
776 | const double debit = tvSubMsec(LastIo, current_time); // actual distance from the last I/O | |
777 | LastIo = current_time; | |
778 | ||
779 | Ipc::QueueReader::Balance &balance = queue->localBalance(); | |
780 | balance += static_cast<int64_t>(credit - debit); | |
781 | ||
782 | debugs(47, 7, HERE << "rate limiting balance: " << balance << " after +" << credit << " -" << debit); | |
783 | ||
1e614370 DK |
784 | if (ipcIo.command == IpcIo::cmdWrite && balance > maxImbalance) { |
785 | // if the next request is (likely) write and we accumulated | |
786 | // too much time for future slow I/Os, then shed accumulated | |
787 | // time to keep just half of the excess | |
df881a0f | 788 | const int64_t toSpend = balance - maxImbalance/2; |
69321ae9 AR |
789 | |
790 | if (toSpend/1e3 > Timeout) | |
82ed74dc AR |
791 | debugs(47, DBG_IMPORTANT, "WARNING: " << DbName << " delays " << |
792 | "I/O requests for " << (toSpend/1e3) << " seconds " << | |
793 | "to obey " << ioRate << "/sec rate limit"); | |
69321ae9 | 794 | |
df881a0f AR |
795 | debugs(47, 3, HERE << "rate limiting by " << toSpend << " ms to get" << |
796 | (1e3*maxRate) << "/sec rate"); | |
797 | eventAdd("IpcIoFile::DiskerHandleMoreRequests", | |
798 | &IpcIoFile::DiskerHandleMoreRequests, | |
799 | const_cast<char*>("rate limiting"), | |
800 | toSpend/1e3, 0, false); | |
801 | DiskerHandleMoreRequestsScheduled = true; | |
802 | return true; | |
e29ccb57 | 803 | } else if (balance < -maxImbalance) { |
df881a0f AR |
804 | // do not owe "too much" to avoid "too large" bursts of I/O |
805 | balance = -maxImbalance; | |
806 | } | |
807 | ||
808 | return false; | |
809 | } | |
810 | ||
254912f3 | 811 | void |
f5591061 | 812 | IpcIoFile::DiskerHandleRequests() |
254912f3 | 813 | { |
c792f7fc | 814 | // Balance our desire to maximize the number of concurrent I/O requests |
fdb3059b | 815 | // (reordred by OS to minimize seek time) with a requirement to |
c792f7fc | 816 | // send 1st-I/O notification messages, process Coordinator events, etc. |
fdb3059b AR |
817 | const int maxSpentMsec = 10; // keep small: most RAM I/Os are under 1ms |
818 | const timeval loopStart = current_time; | |
819 | ||
c792f7fc | 820 | int popped = 0; |
fa61cefe AR |
821 | int workerId = 0; |
822 | IpcIoMsg ipcIo; | |
df881a0f | 823 | while (!WaitBeforePop() && queue->pop(workerId, ipcIo)) { |
fdb3059b AR |
824 | ++popped; |
825 | ||
826 | // at least one I/O per call is guaranteed if the queue is not empty | |
fa61cefe AR |
827 | DiskerHandleRequest(workerId, ipcIo); |
828 | ||
fdb3059b AR |
829 | getCurrentTime(); |
830 | const double elapsedMsec = tvSubMsec(loopStart, current_time); | |
831 | if (elapsedMsec > maxSpentMsec || elapsedMsec < 0) { | |
832 | if (!DiskerHandleMoreRequestsScheduled) { | |
833 | // the gap must be positive for select(2) to be given a chance | |
834 | const double minBreakSecs = 0.001; | |
835 | eventAdd("IpcIoFile::DiskerHandleMoreRequests", | |
9199139f | 836 | &IpcIoFile::DiskerHandleMoreRequests, |
df881a0f AR |
837 | const_cast<char*>("long I/O loop"), |
838 | minBreakSecs, 0, false); | |
fdb3059b AR |
839 | DiskerHandleMoreRequestsScheduled = true; |
840 | } | |
841 | debugs(47, 3, HERE << "pausing after " << popped << " I/Os in " << | |
9199139f | 842 | elapsedMsec << "ms; " << (elapsedMsec/popped) << "ms per I/O"); |
fdb3059b AR |
843 | break; |
844 | } | |
c792f7fc AR |
845 | } |
846 | ||
fdb3059b | 847 | // TODO: consider using O_DIRECT with "elevator" optimization where we pop |
fa61cefe AR |
848 | // requests first, then reorder the popped requests to optimize seek time, |
849 | // then do I/O, then take a break, and come back for the next set of I/O | |
850 | // requests. | |
9a51593d | 851 | } |
254912f3 | 852 | |
9a51593d DK |
853 | /// called when disker receives an I/O request |
854 | void | |
855 | IpcIoFile::DiskerHandleRequest(const int workerId, IpcIoMsg &ipcIo) | |
856 | { | |
9a51593d | 857 | if (ipcIo.command != IpcIo::cmdRead && ipcIo.command != IpcIo::cmdWrite) { |
82ed74dc | 858 | debugs(0, DBG_CRITICAL, "ERROR: " << DbName << |
9a51593d DK |
859 | " should not receive " << ipcIo.command << |
860 | " ipcIo" << workerId << '.' << ipcIo.requestId); | |
861 | return; | |
862 | } | |
863 | ||
864 | debugs(47,5, HERE << "disker" << KidIdentifier << | |
865 | (ipcIo.command == IpcIo::cmdRead ? " reads " : " writes ") << | |
866 | ipcIo.len << " at " << ipcIo.offset << | |
867 | " ipcIo" << workerId << '.' << ipcIo.requestId); | |
868 | ||
869 | if (ipcIo.command == IpcIo::cmdRead) | |
870 | diskerRead(ipcIo); | |
871 | else // ipcIo.command == IpcIo::cmdWrite | |
872 | diskerWrite(ipcIo); | |
873 | ||
f5591061 | 874 | debugs(47, 7, HERE << "pushing " << SipcIo(workerId, ipcIo, KidIdentifier)); |
fa61cefe | 875 | |
7a907247 | 876 | try { |
f5591061 | 877 | if (queue->push(workerId, ipcIo)) |
5e44782e | 878 | Notify(workerId); // must notify worker |
f5591061 | 879 | } catch (const Queue::Full &) { |
fa61cefe AR |
880 | // The worker queue should not overflow because the worker should pop() |
881 | // before push()ing and because if disker pops N requests at a time, | |
882 | // we should make sure the worker pop() queue length is the worker | |
883 | // push queue length plus N+1. XXX: implement the N+1 difference. | |
82ed74dc AR |
884 | debugs(47, DBG_IMPORTANT, "BUG: Worker I/O pop queue for " << |
885 | DbName << " overflow: " << | |
5e44782e | 886 | SipcIo(workerId, ipcIo, KidIdentifier)); // TODO: report queue len |
fa61cefe AR |
887 | |
888 | // the I/O request we could not push will timeout | |
7a907247 | 889 | } |
254912f3 AR |
890 | } |
891 | ||
892 | static bool | |
7015a149 | 893 | DiskerOpen(const SBuf &path, int flags, mode_t mode) |
254912f3 AR |
894 | { |
895 | assert(TheFile < 0); | |
896 | ||
82ed74dc | 897 | DbName = path; |
7015a149 | 898 | TheFile = file_open(DbName.c_str(), flags); |
254912f3 AR |
899 | |
900 | if (TheFile < 0) { | |
901 | const int xerrno = errno; | |
82ed74dc | 902 | debugs(47, DBG_CRITICAL, "ERROR: cannot open " << DbName << ": " << |
254912f3 AR |
903 | xstrerr(xerrno)); |
904 | return false; | |
9a51593d | 905 | } |
254912f3 | 906 | |
cb4185f1 | 907 | ++store_open_disk_fd; |
7015a149 | 908 | debugs(79,3, "rock db opened " << DbName << ": FD " << TheFile); |
9a51593d | 909 | return true; |
254912f3 AR |
910 | } |
911 | ||
912 | static void | |
7015a149 | 913 | DiskerClose(const SBuf &path) |
254912f3 AR |
914 | { |
915 | if (TheFile >= 0) { | |
916 | file_close(TheFile); | |
917 | debugs(79,3, HERE << "rock db closed " << path << ": FD " << TheFile); | |
918 | TheFile = -1; | |
5e263176 | 919 | --store_open_disk_fd; |
a1c98830 | 920 | } |
7015a149 | 921 | DbName.clear(); |
254912f3 | 922 | } |
f5591061 | 923 | |
ea2cdeb6 | 924 | /// reports our needs for shared memory pages to Ipc::Mem::Pages |
21b7990f AR |
925 | /// and initializes shared memory segments used by IpcIoFile |
926 | class IpcIoRr: public Ipc::Mem::RegisteredRunner | |
ea2cdeb6 DK |
927 | { |
928 | public: | |
929 | /* RegisteredRunner API */ | |
21b7990f AR |
930 | IpcIoRr(): owner(NULL) {} |
931 | virtual ~IpcIoRr(); | |
932 | virtual void claimMemoryNeeds(); | |
933 | ||
934 | protected: | |
935 | /* Ipc::Mem::RegisteredRunner API */ | |
936 | virtual void create(); | |
937 | ||
938 | private: | |
939 | Ipc::FewToFewBiQueue::Owner *owner; | |
ea2cdeb6 DK |
940 | }; |
941 | ||
21b7990f | 942 | RunnerRegistrationEntry(IpcIoRr); |
ea2cdeb6 | 943 | |
ea2cdeb6 | 944 | void |
21b7990f | 945 | IpcIoRr::claimMemoryNeeds() |
ea2cdeb6 DK |
946 | { |
947 | const int itemsCount = Ipc::FewToFewBiQueue::MaxItemsCount( | |
c11211d9 | 948 | ::Config.workers, ::Config.cacheSwap.n_strands, QueueCapacity); |
ea2cdeb6 DK |
949 | // the maximum number of shared I/O pages is approximately the |
950 | // number of queue slots, we add a fudge factor to that to account | |
951 | // for corner cases where I/O pages are created before queue | |
952 | // limits are checked or destroyed long after the I/O is dequeued | |
29a9ff4e DK |
953 | Ipc::Mem::NotePageNeed(Ipc::Mem::PageId::ioPage, |
954 | static_cast<int>(itemsCount * 1.1)); | |
ea2cdeb6 DK |
955 | } |
956 | ||
21b7990f AR |
957 | void |
958 | IpcIoRr::create() | |
f5591061 | 959 | { |
e9f68413 | 960 | if (Config.cacheSwap.n_strands <= 0) |
f5591061 DK |
961 | return; |
962 | ||
4404f1c5 | 963 | Must(!owner); |
4404f1c5 | 964 | owner = Ipc::FewToFewBiQueue::Init(ShmLabel, Config.workers, 1, |
3b581957 | 965 | Config.cacheSwap.n_strands, |
4404f1c5 | 966 | 1 + Config.workers, sizeof(IpcIoMsg), |
ea2cdeb6 | 967 | QueueCapacity); |
f5591061 DK |
968 | } |
969 | ||
970 | IpcIoRr::~IpcIoRr() | |
971 | { | |
972 | delete owner; | |
973 | } | |
f53969cc | 974 |