| 1 | /* |
| 2 | |
| 3 | Derby - Class org.apache.derby.impl.store.raw.log.FileLogger |
| 4 | |
| 5 | Copyright 1997, 2004 The Apache Software Foundation or its licensors, as applicable. |
| 6 | |
| 7 | Licensed under the Apache License, Version 2.0 (the "License"); |
| 8 | you may not use this file except in compliance with the License. |
| 9 | You may obtain a copy of the License at |
| 10 | |
| 11 | http://www.apache.org/licenses/LICENSE-2.0 |
| 12 | |
| 13 | Unless required by applicable law or agreed to in writing, software |
| 14 | distributed under the License is distributed on an "AS IS" BASIS, |
| 15 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 16 | See the License for the specific language governing permissions and |
| 17 | limitations under the License. |
| 18 | |
| 19 | */ |
| 20 | |
| 21 | package org.apache.derby.impl.store.raw.log; |
| 22 | |
| 23 | import org.apache.derby.iapi.reference.SQLState; |
| 24 | import org.apache.derby.iapi.reference.MessageId; |
| 25 | |
| 26 | import org.apache.derby.impl.store.raw.log.LogCounter; |
| 27 | import org.apache.derby.impl.store.raw.log.LogRecord; |
| 28 | import org.apache.derby.impl.store.raw.log.StreamLogScan; |
| 29 | |
| 30 | import org.apache.derby.iapi.store.access.TransactionController; |
| 31 | |
| 32 | import org.apache.derby.iapi.store.raw.RawStoreFactory; |
| 33 | |
| 34 | import org.apache.derby.iapi.store.raw.log.LogInstant; |
| 35 | import org.apache.derby.iapi.store.raw.log.Logger; |
| 36 | import org.apache.derby.iapi.store.raw.log.LogScan; |
| 37 | |
| 38 | import org.apache.derby.iapi.store.raw.xact.RawTransaction; |
| 39 | import org.apache.derby.iapi.store.raw.xact.TransactionFactory; |
| 40 | import org.apache.derby.iapi.store.raw.xact.TransactionId; |
| 41 | |
| 42 | import org.apache.derby.iapi.store.raw.Compensation; |
| 43 | import org.apache.derby.iapi.store.raw.ContainerHandle; |
| 44 | import org.apache.derby.iapi.store.raw.LockingPolicy; |
| 45 | import org.apache.derby.iapi.store.raw.Loggable; |
| 46 | import org.apache.derby.iapi.store.raw.Page; |
| 47 | import org.apache.derby.iapi.store.raw.RePreparable; |
| 48 | import org.apache.derby.iapi.store.raw.Undoable; |
| 49 | |
| 50 | import org.apache.derby.iapi.services.io.FormatIdOutputStream; |
| 51 | |
| 52 | import org.apache.derby.iapi.services.sanity.SanityManager; |
| 53 | |
| 54 | import org.apache.derby.iapi.error.StandardException; |
| 55 | import org.apache.derby.iapi.services.i18n.MessageService; |
| 56 | |
| 57 | import org.apache.derby.iapi.services.io.ArrayInputStream; |
| 58 | import org.apache.derby.iapi.services.io.ArrayOutputStream; |
| 59 | import org.apache.derby.iapi.util.ByteArray; |
| 60 | import org.apache.derby.iapi.services.io.DynamicByteArrayOutputStream; |
| 61 | |
| 62 | import org.apache.derby.iapi.services.io.LimitObjectInput; |
| 63 | import java.io.IOException; |
| 64 | |
| 65 | import org.apache.derby.impl.store.raw.data.InitPageOperation; |
| 66 | |
| 67 | /** |
| 68 | Write log records to a log file as a stream |
| 69 | (ie. log records added to the end of the file, no concept of pages). |
| 70 | <P> |
| 71 | The format of a log record that is not a compensation operation is |
| 72 | <PRE> |
| 73 | @format_id no formatId, format is implied by the log file format and the |
| 74 | log record content. |
| 75 | @purpose the log record and optional data |
| 76 | @upgrade |
| 77 | @disk_layout |
| 78 | Log Record |
| 79 | (see org.apache.derby.impl.store.raw.log.LogRecord) |
| 80 | length(int) length of optional data |
| 81 | optionalData(byte[length]) optional data written by the log record |
| 82 | @end_format |
| 83 | </PRE> <HR WIDTH="100%"> |
| 84 | |
| 85 | <P> The form of a log record that is a compensation operation is |
| 86 | <PRE> |
| 87 | @format_id no formatId, format is implied by the log file format and the |
| 88 | log record content. |
| 89 | @purpose undo a previous log record |
| 90 | @upgrade |
| 91 | @disk_layout |
| 92 | Log Record that contains the compenstation operation |
| 93 | (see org.apache.derby.impl.store.raw.log.LogRecord) |
| 94 | undoInstant(long) the log instant of the operation that is to be rolled back |
| 95 | The undo instant is logically part of the LogRecord but is written |
| 96 | by the logger because it is used and controlled by the rollback |
| 97 | code but not by the log operation. |
| 98 | There is no optional data in a compensation operation, all data |
| 99 | necessary for the rollback must be stored in the operation being |
| 100 | undone. |
| 101 | @end_format |
| 102 | </PRE> |
| 103 | |
| 104 | <BR> |
| 105 | |
| 106 | <P>Multithreading considerations:<BR> |
| 107 | Logger must be MT-safe. Each RawTransaction has its own private |
| 108 | FileLogger object. Each logger has a logOutputBuffer and a log input |
| 109 | buffer which are used to read and write to the log. Since multiple |
| 110 | threads can be in the same transaction, fileLogger must be synchronized. |
| 111 | |
| 112 | @see LogRecord |
| 113 | */ |
| 114 | |
| 115 | public class FileLogger implements Logger { |
| 116 | |
| 117 | private LogRecord logRecord; |
| 118 | |
| 119 | protected byte[] encryptionBuffer; |
| 120 | private DynamicByteArrayOutputStream logOutputBuffer; |
| 121 | private FormatIdOutputStream logicalOut; |
| 122 | |
| 123 | private ArrayInputStream logIn; |
| 124 | |
| 125 | private LogToFile logFactory; // actually writes the log records. |
| 126 | |
| 127 | /** |
| 128 | Make a new Logger with its own log record buffers |
| 129 | MT - not needed for constructor |
| 130 | */ |
| 131 | public FileLogger(LogToFile logFactory) { |
| 132 | |
| 133 | this.logFactory = logFactory; |
| 134 | logOutputBuffer = new DynamicByteArrayOutputStream(1024); // init size 1K |
| 135 | logicalOut = new FormatIdOutputStream(logOutputBuffer); |
| 136 | |
| 137 | // logIn and logOutputBuffer must share the same buffer because they |
| 138 | // combined to form an IO stream to access the same log record. |
| 139 | // |
| 140 | // Before each use of logIn, you must reset logIn's data to the |
| 141 | // byte array you want to read from. |
| 142 | // |
| 143 | // To log a record, set logIn's data to point to logOutputBuffer's |
| 144 | // byte array when you know you have everything you need in the output |
| 145 | // buffer, then set limit on logIn and send it to the log operation's |
| 146 | // doMe. |
| 147 | // |
| 148 | // Keep in mind the dynamic nature of the logOutputBuffer which means |
| 149 | // it could switch buffer from underneath the logOutputBuffer on every |
| 150 | // write. |
| 151 | logIn = new ArrayInputStream(); |
| 152 | |
| 153 | logRecord = new LogRecord(); |
| 154 | |
| 155 | } |
| 156 | |
| 157 | /** |
| 158 | Close the logger. |
| 159 | MT - caller provide synchronization |
| 160 | (RESOLVE: not called by anyone ??) |
| 161 | */ |
| 162 | public void close() throws IOException |
| 163 | { |
| 164 | if (logOutputBuffer != null) |
| 165 | { |
| 166 | logOutputBuffer.close(); |
| 167 | logOutputBuffer = null; |
| 168 | } |
| 169 | |
| 170 | logIn = null; |
| 171 | logFactory = null; |
| 172 | |
| 173 | logicalOut = null; |
| 174 | |
| 175 | logRecord = null; |
| 176 | } |
| 177 | |
| 178 | /* |
| 179 | ** Methods of Logger |
| 180 | */ |
| 181 | |
| 182 | /** |
| 183 | Writes out a log record to the log stream, and call its doMe method to |
| 184 | apply the change to the rawStore. |
| 185 | <BR>Any optional data the doMe method need is first written to the log |
| 186 | stream using operation.writeOptionalData, then whatever is written to |
| 187 | the log stream is passed back to the operation for the doMe method. |
| 188 | |
| 189 | <P>MT - there could be multiple threads running in the same raw |
| 190 | transactions and they can be calling the same logger to log different |
| 191 | log operations. This whole method is synchronized to make sure log |
| 192 | records are logged one at a time. |
| 193 | |
| 194 | @param xact the transaction logging the change |
| 195 | @param operation the log operation |
| 196 | @return the instant in the log that can be used to identify the log |
| 197 | record |
| 198 | |
| 199 | @exception StandardException Cloudscape Standard error policy |
| 200 | */ |
| 201 | public synchronized LogInstant logAndDo(RawTransaction xact, Loggable operation) |
| 202 | throws StandardException |
| 203 | { |
| 204 | boolean isLogPrepared = false; |
| 205 | |
| 206 | boolean inUserCode = false; |
| 207 | byte[] preparedLog; |
| 208 | |
| 209 | try { |
| 210 | |
| 211 | logOutputBuffer.reset(); |
| 212 | |
| 213 | // always use the short Id, only the BeginXact log record contains |
| 214 | // the XactId (long form) |
| 215 | TransactionId transactionId = xact.getId(); |
| 216 | |
| 217 | // write out the log header with the operation embedded |
| 218 | // this is by definition not a compensation log record, |
| 219 | // those are called thru the logAndUndo interface |
| 220 | logRecord.setValue(transactionId, operation); |
| 221 | |
| 222 | inUserCode = true; |
| 223 | logicalOut.writeObject(logRecord); |
| 224 | inUserCode = false; |
| 225 | |
| 226 | int optionalDataLength = 0; |
| 227 | int optionalDataOffset = 0; |
| 228 | int completeLength = 0; |
| 229 | |
| 230 | ByteArray preparedLogArray = operation.getPreparedLog(); |
| 231 | if (preparedLogArray != null) { |
| 232 | |
| 233 | preparedLog = preparedLogArray.getArray(); |
| 234 | optionalDataLength = preparedLogArray.getLength(); |
| 235 | optionalDataOffset = preparedLogArray.getOffset(); |
| 236 | |
| 237 | // There is a race condition if the operation is a begin tran in |
| 238 | // that between the time the beginXact log record is written to |
| 239 | // disk and the time the transaction object is updated in the |
| 240 | // beginXact.doMe method, other log records may be written. |
| 241 | // This will render the transaction table in an inconsistent state |
| 242 | // since it may think a later transaction is the earliest |
| 243 | // transaction or it may think that there is no active transactions |
| 244 | // where there is a bunch of them sitting on the log. |
| 245 | // |
| 246 | // Similarly, there is a race condition for endXact, i.e., |
| 247 | // 1) endXact is written to the log, |
| 248 | // 2) checkpoint gets that (committed) transaction as the |
| 249 | // firstUpdateTransaction |
| 250 | // 3) the transaction calls postComplete, nulling out itself |
| 251 | // 4) checkpoint tries to access a closed transaction object |
| 252 | // |
| 253 | // The solution is to sync between the time a begin tran or end |
| 254 | // tran log record is sent to the log stream and its doMe method is |
| 255 | // called to update the transaction table and in memory state |
| 256 | // |
| 257 | // We only need to serialized the begin and end Xact log records |
| 258 | // because once a transaction has been started and in the |
| 259 | // transaction table, its order and transaction state does not |
| 260 | // change. |
| 261 | // |
| 262 | // Use the logFactory as the sync object so that a checkpoint can |
| 263 | // take its snap shot of the undoLWM before or after a transaction |
| 264 | // is started, but not in the middle. (see LogToFile.checkpoint) |
| 265 | // |
| 266 | |
| 267 | // now set the input limit to be the optional data. |
| 268 | // This limits amount of data availiable to logIn that doMe can |
| 269 | // use |
| 270 | logIn.setData(preparedLog); |
| 271 | logIn.setPosition(optionalDataOffset); |
| 272 | logIn.setLimit(optionalDataLength); |
| 273 | |
| 274 | if (SanityManager.DEBUG) |
| 275 | { |
| 276 | if ((optionalDataLength) != logIn.available()) |
| 277 | SanityManager.THROWASSERT( |
| 278 | " stream not set correctly " + |
| 279 | optionalDataLength + " != " + |
| 280 | logIn.available()); |
| 281 | } |
| 282 | |
| 283 | } else { |
| 284 | preparedLog = null; |
| 285 | optionalDataLength = 0; |
| 286 | } |
| 287 | |
| 288 | logicalOut.writeInt(optionalDataLength); |
| 289 | completeLength = logOutputBuffer.getPosition() + optionalDataLength; |
| 290 | |
| 291 | |
| 292 | LogInstant logInstant = null; |
| 293 | int encryptedLength = 0; // in case of encryption, we need to pad |
| 294 | |
| 295 | try |
| 296 | { |
| 297 | if (logFactory.databaseEncrypted()) |
| 298 | { |
| 299 | // we must pad the encryption data to be multiple of block |
| 300 | // size, which is logFactory.getEncryptionBlockSize() |
| 301 | encryptedLength = completeLength; |
| 302 | if ((encryptedLength % logFactory.getEncryptionBlockSize()) != 0) |
| 303 | encryptedLength = encryptedLength + logFactory.getEncryptionBlockSize() - (encryptedLength % logFactory.getEncryptionBlockSize()); |
| 304 | |
| 305 | if (encryptionBuffer == null || |
| 306 | encryptionBuffer.length < encryptedLength) |
| 307 | encryptionBuffer = new byte[encryptedLength]; |
| 308 | |
| 309 | System.arraycopy(logOutputBuffer.getByteArray(), 0, |
| 310 | encryptionBuffer, 0, completeLength-optionalDataLength); |
| 311 | |
| 312 | if (optionalDataLength > 0) |
| 313 | System.arraycopy(preparedLog, optionalDataOffset, |
| 314 | encryptionBuffer, |
| 315 | completeLength-optionalDataLength, optionalDataLength); |
| 316 | |
| 317 | // do not bother to clear out the padding area |
| 318 | int len = |
| 319 | logFactory.encrypt(encryptionBuffer, 0, encryptedLength, |
| 320 | encryptionBuffer, 0); |
| 321 | |
| 322 | if (SanityManager.DEBUG) |
| 323 | SanityManager.ASSERT(len == encryptedLength, |
| 324 | "encrypted log buffer length != log buffer len"); |
| 325 | } |
| 326 | |
| 327 | if ((operation.group() & (Loggable.FIRST | Loggable.LAST)) != 0) |
| 328 | { |
| 329 | synchronized (logFactory) |
| 330 | { |
| 331 | long instant = 0; |
| 332 | |
| 333 | if (logFactory.databaseEncrypted()) |
| 334 | { |
| 335 | // encryption has completely drained both the the |
| 336 | // logOuputBuffer array and the preparedLog array |
| 337 | instant = logFactory. |
| 338 | appendLogRecord(encryptionBuffer, 0, |
| 339 | encryptedLength, null, |
| 340 | -1, 0); |
| 341 | } |
| 342 | else |
| 343 | { |
| 344 | instant = logFactory. |
| 345 | appendLogRecord(logOutputBuffer.getByteArray(), |
| 346 | 0, completeLength, preparedLog, |
| 347 | optionalDataOffset, |
| 348 | optionalDataLength); |
| 349 | } |
| 350 | logInstant = new LogCounter(instant); |
| 351 | |
| 352 | operation.doMe(xact, logInstant, logIn); |
| 353 | } |
| 354 | } |
| 355 | else |
| 356 | { |
| 357 | long instant = 0; |
| 358 | |
| 359 | if (logFactory.databaseEncrypted()) |
| 360 | { |
| 361 | // encryption has completely drained both the the |
| 362 | // logOuputBuffer array and the preparedLog array |
| 363 | instant = logFactory. |
| 364 | appendLogRecord(encryptionBuffer, 0, |
| 365 | encryptedLength, null, -1, 0); |
| 366 | } |
| 367 | else |
| 368 | { |
| 369 | instant = logFactory. |
| 370 | appendLogRecord(logOutputBuffer.getByteArray(), 0, |
| 371 | completeLength, preparedLog, |
| 372 | optionalDataOffset, |
| 373 | optionalDataLength); |
| 374 | } |
| 375 | |
| 376 | logInstant = new LogCounter(instant); |
| 377 | |
| 378 | operation.doMe(xact, logInstant, logIn); |
| 379 | } |
| 380 | |
| 381 | } |
| 382 | catch (StandardException se) |
| 383 | { |
| 384 | throw logFactory.markCorrupt( |
| 385 | StandardException.newException( |
| 386 | SQLState.LOG_DO_ME_FAIL, se, operation)); |
| 387 | } |
| 388 | catch (IOException ioe) |
| 389 | { |
| 390 | throw logFactory.markCorrupt( |
| 391 | StandardException.newException( |
| 392 | SQLState.LOG_DO_ME_FAIL, ioe, operation)); |
| 393 | } |
| 394 | finally |
| 395 | { |
| 396 | logIn.clearLimit(); |
| 397 | } |
| 398 | |
| 399 | if (SanityManager.DEBUG) |
| 400 | { |
| 401 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 402 | { |
| 403 | SanityManager.DEBUG( |
| 404 | LogToFile.DBG_FLAG, |
| 405 | "Write log record: tranId=" + transactionId.toString() + |
| 406 | " instant: " + logInstant.toString() + " length: " + |
| 407 | completeLength + "\n" + operation + "\n"); |
| 408 | } |
| 409 | } |
| 410 | return logInstant; |
| 411 | } |
| 412 | |
| 413 | catch (IOException ioe) |
| 414 | { |
| 415 | // error writing to the log buffer |
| 416 | if (inUserCode) |
| 417 | { |
| 418 | throw StandardException.newException( |
| 419 | SQLState.LOG_WRITE_LOG_RECORD, ioe, operation); |
| 420 | } |
| 421 | else |
| 422 | { |
| 423 | throw StandardException.newException( |
| 424 | SQLState.LOG_BUFFER_FULL, ioe, operation); |
| 425 | } |
| 426 | } |
| 427 | |
| 428 | } |
| 429 | |
| 430 | /** |
| 431 | Writes out a compensation log record to the log stream, and call its |
| 432 | doMe method to undo the change of a previous log operation. |
| 433 | |
| 434 | <P>MT - Not needed. A transaction must be single threaded thru undo, each |
| 435 | RawTransaction has its own logger, therefore no need to synchronize. |
| 436 | The RawTransaction must handle synchronizing with multiple threads |
| 437 | during rollback. |
| 438 | |
| 439 | @param xact the transaction logging the change |
| 440 | @param compensation the compensation log operation |
| 441 | @param undoInstant the log instant of the operation that is to be |
| 442 | rolled back |
| 443 | @param in optional data input for the compenastion doMe method |
| 444 | |
| 445 | @return the instant in the log that can be used to identify the log |
| 446 | record |
| 447 | |
| 448 | @exception StandardException Cloudscape Standard error policy |
| 449 | */ |
| 450 | public LogInstant logAndUndo(RawTransaction xact, |
| 451 | Compensation compensation, |
| 452 | LogInstant undoInstant, |
| 453 | LimitObjectInput in) |
| 454 | throws StandardException |
| 455 | { |
| 456 | boolean inUserCode = false; |
| 457 | |
| 458 | try { |
| 459 | logOutputBuffer.reset(); |
| 460 | |
| 461 | TransactionId transactionId = xact.getId(); |
| 462 | |
| 463 | // write out the log header with the operation embedded |
| 464 | logRecord.setValue(transactionId, compensation); |
| 465 | |
| 466 | inUserCode = true; |
| 467 | logicalOut.writeObject(logRecord); |
| 468 | inUserCode = false; |
| 469 | |
| 470 | // write out the undoInstant |
| 471 | logicalOut.writeLong(((LogCounter)undoInstant).getValueAsLong()); |
| 472 | |
| 473 | // in this implemetaion, there is no optional data for the |
| 474 | // compensation operation. Optional data for the rollback comes |
| 475 | // from the undoable operation - and is passed into this call. |
| 476 | int completeLength = logOutputBuffer.getPosition(); |
| 477 | long instant = 0; |
| 478 | |
| 479 | if (logFactory.databaseEncrypted()) |
| 480 | { |
| 481 | // we must pad the encryption data to be multiple of block |
| 482 | // size, which is logFactory.getEncryptionBlockSize() |
| 483 | int encryptedLength = completeLength; |
| 484 | if ((encryptedLength % logFactory.getEncryptionBlockSize()) != 0) |
| 485 | encryptedLength = encryptedLength + logFactory.getEncryptionBlockSize() - (encryptedLength % logFactory.getEncryptionBlockSize()); |
| 486 | |
| 487 | if (encryptionBuffer == null || |
| 488 | encryptionBuffer.length < encryptedLength) |
| 489 | encryptionBuffer = new byte[encryptedLength]; |
| 490 | |
| 491 | System.arraycopy(logOutputBuffer.getByteArray(), 0, |
| 492 | encryptionBuffer, 0, completeLength); |
| 493 | |
| 494 | // do not bother to clear out the padding area |
| 495 | int len = |
| 496 | logFactory.encrypt(encryptionBuffer, 0, encryptedLength, |
| 497 | encryptionBuffer, 0); |
| 498 | |
| 499 | if (SanityManager.DEBUG) |
| 500 | SanityManager.ASSERT(len == encryptedLength, |
| 501 | "encrypted log buffer length != log buffer len"); |
| 502 | |
| 503 | instant = logFactory. |
| 504 | appendLogRecord(encryptionBuffer, |
| 505 | 0, encryptedLength, null, 0, 0); |
| 506 | } |
| 507 | else |
| 508 | { |
| 509 | instant = logFactory. |
| 510 | appendLogRecord(logOutputBuffer.getByteArray(), |
| 511 | 0, completeLength, null, 0, 0); |
| 512 | } |
| 513 | |
| 514 | LogInstant logInstant = new LogCounter(instant); |
| 515 | |
| 516 | if (SanityManager.DEBUG) |
| 517 | { |
| 518 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 519 | { |
| 520 | SanityManager.DEBUG( |
| 521 | LogToFile.DBG_FLAG, |
| 522 | "Write CLR: Xact: " + transactionId.toString() + |
| 523 | "clrinstant: " + logInstant.toString() + |
| 524 | " undoinstant " + undoInstant + "\n"); |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | try |
| 529 | { |
| 530 | // in and dataLength contains optional data that was written |
| 531 | // to the log during a previous call to logAndDo. |
| 532 | compensation.doMe(xact, logInstant, in); |
| 533 | } |
| 534 | catch (StandardException se) |
| 535 | { |
| 536 | throw logFactory.markCorrupt( |
| 537 | StandardException.newException( |
| 538 | SQLState.LOG_DO_ME_FAIL, se, compensation)); |
| 539 | } |
| 540 | catch (IOException ioe) |
| 541 | { |
| 542 | throw logFactory.markCorrupt( |
| 543 | StandardException.newException( |
| 544 | SQLState.LOG_DO_ME_FAIL, ioe, compensation)); |
| 545 | } |
| 546 | |
| 547 | return logInstant; |
| 548 | |
| 549 | } |
| 550 | catch (IOException ioe) |
| 551 | { |
| 552 | if (inUserCode) |
| 553 | { |
| 554 | throw StandardException.newException( |
| 555 | SQLState.LOG_WRITE_LOG_RECORD, ioe, compensation); |
| 556 | } |
| 557 | else |
| 558 | { |
| 559 | throw StandardException.newException( |
| 560 | SQLState.LOG_BUFFER_FULL, ioe, compensation); |
| 561 | } |
| 562 | } |
| 563 | } |
| 564 | |
| 565 | /** |
| 566 | Flush the log up to the given log instant. |
| 567 | |
| 568 | <P>MT - not needed, wrapper method |
| 569 | |
| 570 | @exception StandardException cannot sync log file |
| 571 | */ |
| 572 | public void flush(LogInstant where) |
| 573 | throws StandardException |
| 574 | { |
| 575 | if (SanityManager.DEBUG) |
| 576 | { |
| 577 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 578 | { |
| 579 | SanityManager.DEBUG( |
| 580 | LogToFile.DBG_FLAG, "Flush log to " + where.toString()); |
| 581 | } |
| 582 | } |
| 583 | |
| 584 | logFactory.flush(where); |
| 585 | } |
| 586 | |
| 587 | /** |
| 588 | Flush all outstanding log to disk. |
| 589 | |
| 590 | <P>MT - not needed, wrapper method |
| 591 | |
| 592 | @exception StandardException cannot sync log file |
| 593 | */ |
| 594 | public void flushAll () throws StandardException |
| 595 | { |
| 596 | logFactory.flushAll(); |
| 597 | } |
| 598 | |
| 599 | /** |
| 600 | * During recovery re-prepare a transaction. |
| 601 | * <p> |
| 602 | * After redo() and undo(), this routine is called on all outstanding |
| 603 | * in-doubt (prepared) transactions. This routine re-acquires all |
| 604 | * logical write locks for operations in the xact, and then modifies |
| 605 | * the transaction table entry to make the transaction look as if it |
| 606 | * had just been prepared following startup after recovery. |
| 607 | * <p> |
| 608 | * |
| 609 | * @param t is the transaction performing the re-prepare |
| 610 | * @param prepareId is the transaction ID to be re-prepared |
| 611 | * @param prepareStopAt is where the log instant (inclusive) where the |
| 612 | * re-prepare should stop. |
| 613 | * @param prepareStartAt is the log instant (inclusive) where re-prepare |
| 614 | * should begin, this is normally the log instant |
| 615 | * of the last log record of the transaction that |
| 616 | * is to be re-prepare. If null, then re-prepare |
| 617 | * starts from the end of the log. |
| 618 | * |
| 619 | * @exception StandardException Standard exception policy. |
| 620 | **/ |
| 621 | public void reprepare( |
| 622 | RawTransaction t, |
| 623 | TransactionId prepareId, |
| 624 | LogInstant prepareStopAt, |
| 625 | LogInstant prepareStartAt) |
| 626 | throws StandardException |
| 627 | { |
| 628 | |
| 629 | if (SanityManager.DEBUG) |
| 630 | { |
| 631 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 632 | { |
| 633 | if (prepareStartAt != null) |
| 634 | { |
| 635 | SanityManager.DEBUG( |
| 636 | LogToFile.DBG_FLAG, |
| 637 | "----------------------------------------------------\n" + |
| 638 | "\nBegin of RePrepare : " + prepareId.toString() + |
| 639 | "start at " + prepareStartAt.toString() + |
| 640 | " stop at " + prepareStopAt.toString() + |
| 641 | "\n----------------------------------------------------\n"); |
| 642 | } |
| 643 | else |
| 644 | { |
| 645 | SanityManager.DEBUG( |
| 646 | LogToFile.DBG_FLAG, |
| 647 | "----------------------------------------------------\n" + |
| 648 | "\nBegin of Reprepare: " + prepareId.toString() + |
| 649 | "start at end of log stop at " + |
| 650 | prepareStopAt.toString() + |
| 651 | "\n----------------------------------------------------\n"); |
| 652 | } |
| 653 | } |
| 654 | } |
| 655 | |
| 656 | // statistics |
| 657 | int clrskipped = 0; |
| 658 | int logrecordseen = 0; |
| 659 | |
| 660 | RePreparable lop = null; |
| 661 | |
| 662 | // stream to read the log record - initial size 4096, scanLog needs |
| 663 | // to resize if the log record is larger than that. |
| 664 | ArrayInputStream rawInput = null; |
| 665 | |
| 666 | try |
| 667 | { |
| 668 | StreamLogScan scanLog; |
| 669 | |
| 670 | if (prepareStartAt == null) |
| 671 | { |
| 672 | // don't know where to start, scan from end of log |
| 673 | scanLog = |
| 674 | (StreamLogScan) logFactory.openBackwardsScan(prepareStopAt); |
| 675 | } |
| 676 | else |
| 677 | { |
| 678 | if (prepareStartAt.lessThan(prepareStopAt)) |
| 679 | { |
| 680 | // nothing to prepare! |
| 681 | return; |
| 682 | } |
| 683 | |
| 684 | scanLog = (StreamLogScan) |
| 685 | logFactory.openBackwardsScan( |
| 686 | ((LogCounter) prepareStartAt).getValueAsLong(), |
| 687 | prepareStopAt); |
| 688 | } |
| 689 | |
| 690 | if (SanityManager.DEBUG) |
| 691 | SanityManager.ASSERT( |
| 692 | scanLog != null, "cannot open log for prepare"); |
| 693 | |
| 694 | rawInput = new ArrayInputStream(new byte[4096]); |
| 695 | |
| 696 | LogRecord record; |
| 697 | |
| 698 | while ((record = |
| 699 | scanLog.getNextRecord(rawInput, prepareId, 0)) |
| 700 | != null) |
| 701 | { |
| 702 | if (SanityManager.DEBUG) |
| 703 | { |
| 704 | SanityManager.ASSERT( |
| 705 | record.getTransactionId().equals(prepareId), |
| 706 | "getNextRecord return unqualified log rec for prepare"); |
| 707 | } |
| 708 | |
| 709 | logrecordseen++; |
| 710 | |
| 711 | if (record.isCLR()) |
| 712 | { |
| 713 | clrskipped++; |
| 714 | |
| 715 | // the loggable is still in the input stream, get rid of it |
| 716 | record.skipLoggable(); |
| 717 | |
| 718 | // read the prepareInstant |
| 719 | long prepareInstant = rawInput.readLong(); |
| 720 | |
| 721 | if (SanityManager.DEBUG) |
| 722 | { |
| 723 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 724 | { |
| 725 | SanityManager.DEBUG( |
| 726 | LogToFile.DBG_FLAG, |
| 727 | "Skipping over CLRs, reset scan to " + |
| 728 | LogCounter.toDebugString(prepareInstant)); |
| 729 | } |
| 730 | } |
| 731 | |
| 732 | scanLog.resetPosition(new LogCounter(prepareInstant)); |
| 733 | // scanLog now positioned at the beginning of the log |
| 734 | // record that was rolled back by this CLR. |
| 735 | // The scan is a backward one so getNextRecord will skip |
| 736 | // over the record that was rolled back and go to the one |
| 737 | // previous to it |
| 738 | |
| 739 | continue; |
| 740 | } |
| 741 | |
| 742 | if (record.requiresPrepareLocks()) |
| 743 | { |
| 744 | lop = record.getRePreparable(); |
| 745 | } |
| 746 | else |
| 747 | { |
| 748 | continue; |
| 749 | } |
| 750 | |
| 751 | if (lop != null) |
| 752 | { |
| 753 | // Reget locks based on log record. reclaim all locks with |
| 754 | // a serializable locking policy, since we are only |
| 755 | // reclaiming write locks, isolation level does not matter |
| 756 | // much. |
| 757 | |
| 758 | lop.reclaimPrepareLocks( |
| 759 | t, |
| 760 | t.newLockingPolicy( |
| 761 | LockingPolicy.MODE_RECORD, |
| 762 | TransactionController.ISOLATION_REPEATABLE_READ, |
| 763 | true)); |
| 764 | |
| 765 | if (SanityManager.DEBUG) |
| 766 | { |
| 767 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 768 | { |
| 769 | SanityManager.DEBUG( |
| 770 | LogToFile.DBG_FLAG, |
| 771 | "Reprepare log record at instant " + |
| 772 | scanLog.getInstant() + " : " + lop); |
| 773 | } |
| 774 | } |
| 775 | |
| 776 | } |
| 777 | } |
| 778 | |
| 779 | } |
| 780 | catch (ClassNotFoundException cnfe) |
| 781 | { |
| 782 | throw logFactory.markCorrupt( |
| 783 | StandardException.newException(SQLState.LOG_CORRUPTED, cnfe)); |
| 784 | } |
| 785 | catch (IOException ioe) |
| 786 | { |
| 787 | throw logFactory.markCorrupt( |
| 788 | StandardException.newException( |
| 789 | SQLState.LOG_READ_LOG_FOR_UNDO, ioe)); |
| 790 | } |
| 791 | catch (StandardException se) |
| 792 | { |
| 793 | throw |
| 794 | logFactory.markCorrupt( |
| 795 | StandardException.newException( |
| 796 | SQLState.LOG_UNDO_FAILED, se, |
| 797 | prepareId, lop, (Object) null)); |
| 798 | } |
| 799 | finally |
| 800 | { |
| 801 | if (rawInput != null) |
| 802 | { |
| 803 | try |
| 804 | { |
| 805 | rawInput.close(); |
| 806 | } |
| 807 | catch (IOException ioe) |
| 808 | { |
| 809 | throw logFactory.markCorrupt( |
| 810 | StandardException.newException( |
| 811 | SQLState.LOG_READ_LOG_FOR_UNDO, ioe, prepareId)); |
| 812 | } |
| 813 | } |
| 814 | } |
| 815 | |
| 816 | if (SanityManager.DEBUG) |
| 817 | { |
| 818 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 819 | { |
| 820 | SanityManager.DEBUG(LogToFile.DBG_FLAG, "Finish prepare" + |
| 821 | ", clr skipped = " + clrskipped + |
| 822 | ", record seen = " + logrecordseen + "\n"); |
| 823 | } |
| 824 | } |
| 825 | |
| 826 | if (SanityManager.DEBUG) |
| 827 | { |
| 828 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 829 | { |
| 830 | SanityManager.DEBUG( |
| 831 | LogToFile.DBG_FLAG, |
| 832 | "----------------------------------------------------\n" + |
| 833 | "End of recovery rePrepare\n" + |
| 834 | ", clr skipped = " + clrskipped + |
| 835 | ", record seen = " + logrecordseen + |
| 836 | "\n----------------------------------------------------\n"); |
| 837 | } |
| 838 | } |
| 839 | } |
| 840 | |
| 841 | |
| 842 | /** |
| 843 | Undo a part of or the entire transaction. Begin rolling back the log |
| 844 | record at undoStartAt and stopping at (inclusive) the log record at |
| 845 | undoStopAt. |
| 846 | |
| 847 | <P>MT - Not needed. A transaction must be single threaded thru undo, |
| 848 | each RawTransaction has its own logger, therefore no need to |
| 849 | synchronize. The RawTransaction must handle synchronizing with |
| 850 | multiple threads during rollback. |
| 851 | |
| 852 | @param t the transaction that needs to be rolled back |
| 853 | @param undoId the transaction ID |
| 854 | @param undoStopAt the last log record that should be rolled back |
| 855 | @param undoStartAt the first log record that should be rolled back |
| 856 | |
| 857 | @exception StandardException Standard Cloudscape error policy |
| 858 | |
| 859 | @see Logger#undo |
| 860 | */ |
| 861 | public void undo( |
| 862 | RawTransaction t, |
| 863 | TransactionId undoId, |
| 864 | LogInstant undoStopAt, |
| 865 | LogInstant undoStartAt) |
| 866 | throws StandardException |
| 867 | { |
| 868 | |
| 869 | if (SanityManager.DEBUG) |
| 870 | { |
| 871 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 872 | { |
| 873 | if (undoStartAt != null) |
| 874 | { |
| 875 | SanityManager.DEBUG( |
| 876 | LogToFile.DBG_FLAG, |
| 877 | "\nUndo transaction: " + undoId.toString() + |
| 878 | "start at " + undoStartAt.toString() + |
| 879 | " stop at " + undoStopAt.toString() ); |
| 880 | } |
| 881 | else |
| 882 | { |
| 883 | SanityManager.DEBUG( |
| 884 | LogToFile.DBG_FLAG, |
| 885 | "\nUndo transaction: " + undoId.toString() + |
| 886 | "start at end of log stop at " + undoStopAt.toString()); |
| 887 | } |
| 888 | } |
| 889 | } |
| 890 | |
| 891 | // statistics |
| 892 | int clrgenerated = 0; |
| 893 | int clrskipped = 0; |
| 894 | int logrecordseen = 0; |
| 895 | |
| 896 | StreamLogScan scanLog; |
| 897 | Compensation compensation = null; |
| 898 | Undoable lop = null; |
| 899 | |
| 900 | // stream to read the log record - initial size 4096, scanLog needs |
| 901 | // to resize if the log record is larget than that. |
| 902 | ArrayInputStream rawInput = null; |
| 903 | |
| 904 | try |
| 905 | { |
| 906 | if (undoStartAt == null) |
| 907 | { |
| 908 | // don't know where to start, rollback from end of log |
| 909 | |
| 910 | scanLog = (StreamLogScan) |
| 911 | logFactory.openBackwardsScan(undoStopAt); |
| 912 | } |
| 913 | else |
| 914 | { |
| 915 | if (undoStartAt.lessThan(undoStopAt)) |
| 916 | { |
| 917 | // nothing to undo! |
| 918 | return; |
| 919 | } |
| 920 | |
| 921 | long undoStartInstant = |
| 922 | ((LogCounter) undoStartAt).getValueAsLong(); |
| 923 | |
| 924 | scanLog = (StreamLogScan) |
| 925 | logFactory.openBackwardsScan(undoStartInstant, undoStopAt); |
| 926 | } |
| 927 | |
| 928 | if (SanityManager.DEBUG) |
| 929 | SanityManager.ASSERT( |
| 930 | scanLog != null, "cannot open log for undo"); |
| 931 | |
| 932 | rawInput = new ArrayInputStream(new byte[4096]); |
| 933 | |
| 934 | LogRecord record; |
| 935 | |
| 936 | while ((record = |
| 937 | scanLog.getNextRecord(rawInput, undoId, 0)) |
| 938 | != null) |
| 939 | { |
| 940 | if (SanityManager.DEBUG) |
| 941 | { |
| 942 | SanityManager.ASSERT( |
| 943 | record.getTransactionId().equals(undoId), |
| 944 | "getNextRecord return unqualified log record for undo"); |
| 945 | } |
| 946 | |
| 947 | logrecordseen++; |
| 948 | |
| 949 | if (record.isCLR()) |
| 950 | { |
| 951 | clrskipped++; |
| 952 | |
| 953 | // the loggable is still in the input stream, get rid of it |
| 954 | record.skipLoggable(); |
| 955 | |
| 956 | // read the undoInstant |
| 957 | long undoInstant = rawInput.readLong(); |
| 958 | |
| 959 | if (SanityManager.DEBUG) |
| 960 | { |
| 961 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 962 | { |
| 963 | SanityManager.DEBUG( |
| 964 | LogToFile.DBG_FLAG, |
| 965 | "Skipping over CLRs, reset scan to " + |
| 966 | LogCounter.toDebugString(undoInstant)); |
| 967 | } |
| 968 | } |
| 969 | |
| 970 | |
| 971 | scanLog.resetPosition(new LogCounter(undoInstant)); |
| 972 | |
| 973 | // scanLog now positioned at the beginning of the log |
| 974 | // record that was rolled back by this CLR. |
| 975 | // The scan is a backward one so getNextRecord will skip |
| 976 | // over the record that was rolled back and go to the one |
| 977 | // previous to it |
| 978 | |
| 979 | continue; |
| 980 | } |
| 981 | |
| 982 | lop = record.getUndoable(); |
| 983 | |
| 984 | if (lop != null) |
| 985 | { |
| 986 | int optionalDataLength = rawInput.readInt(); |
| 987 | int savePosition = rawInput.getPosition(); |
| 988 | rawInput.setLimit(savePosition, optionalDataLength); |
| 989 | |
| 990 | compensation = lop.generateUndo(t, rawInput); |
| 991 | |
| 992 | if (SanityManager.DEBUG) |
| 993 | { |
| 994 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 995 | { |
| 996 | SanityManager.DEBUG( |
| 997 | LogToFile.DBG_FLAG, |
| 998 | "Rollback log record at instant " + |
| 999 | LogCounter.toDebugString(scanLog.getInstant()) + |
| 1000 | " : " + lop); |
| 1001 | } |
| 1002 | } |
| 1003 | |
| 1004 | clrgenerated++; |
| 1005 | |
| 1006 | if (compensation != null) |
| 1007 | { |
| 1008 | // generateUndo may have read stuff off the |
| 1009 | // stream, reset it for the undo operation. |
| 1010 | rawInput.setLimit(savePosition, optionalDataLength); |
| 1011 | |
| 1012 | // log the compensation op that rolls back the |
| 1013 | // operation at this instant |
| 1014 | t.logAndUndo( |
| 1015 | compensation, new LogCounter(scanLog.getInstant()), |
| 1016 | rawInput); |
| 1017 | |
| 1018 | compensation.releaseResource(t); |
| 1019 | compensation = null; |
| 1020 | } |
| 1021 | |
| 1022 | // if compensation is null, log operation is redo only |
| 1023 | } |
| 1024 | // if this is not an undoable operation, continue with next log |
| 1025 | // record |
| 1026 | } |
| 1027 | } |
| 1028 | catch (ClassNotFoundException cnfe) |
| 1029 | { |
| 1030 | throw logFactory.markCorrupt( |
| 1031 | StandardException.newException(SQLState.LOG_CORRUPTED, cnfe)); |
| 1032 | } |
| 1033 | catch (IOException ioe) |
| 1034 | { |
| 1035 | throw logFactory.markCorrupt( |
| 1036 | StandardException.newException( |
| 1037 | SQLState.LOG_READ_LOG_FOR_UNDO, ioe)); |
| 1038 | } |
| 1039 | catch (StandardException se) |
| 1040 | { |
| 1041 | // TODO (4327) - exceptions caught here are nested in the exception |
| 1042 | // below but for some reason the nested exceptions are not logged |
| 1043 | // or reported in any way. |
| 1044 | |
| 1045 | throw logFactory.markCorrupt( |
| 1046 | StandardException.newException( |
| 1047 | SQLState.LOG_UNDO_FAILED, se, undoId, lop, compensation)); |
| 1048 | } |
| 1049 | finally |
| 1050 | { |
| 1051 | if (compensation != null) |
| 1052 | { |
| 1053 | // errored out |
| 1054 | compensation.releaseResource(t); |
| 1055 | } |
| 1056 | |
| 1057 | if (rawInput != null) |
| 1058 | { |
| 1059 | try |
| 1060 | { |
| 1061 | rawInput.close(); |
| 1062 | } |
| 1063 | catch (IOException ioe) |
| 1064 | { |
| 1065 | throw logFactory.markCorrupt( |
| 1066 | StandardException.newException( |
| 1067 | SQLState.LOG_READ_LOG_FOR_UNDO, ioe, undoId)); |
| 1068 | } |
| 1069 | } |
| 1070 | } |
| 1071 | |
| 1072 | if (SanityManager.DEBUG) |
| 1073 | { |
| 1074 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 1075 | { |
| 1076 | SanityManager.DEBUG( |
| 1077 | LogToFile.DBG_FLAG, |
| 1078 | "Finish undo" + |
| 1079 | ", clr generated = " + clrgenerated + |
| 1080 | ", clr skipped = " + clrskipped + |
| 1081 | ", record seen = " + logrecordseen + "\n"); |
| 1082 | } |
| 1083 | } |
| 1084 | } |
| 1085 | |
| 1086 | |
| 1087 | /** |
| 1088 | Recovery Redo loop. |
| 1089 | |
| 1090 | <P> The log stream is scanned from the beginning (or |
| 1091 | from the undo low water mark of a checkpoint) forward until the end. |
| 1092 | The purpose of the redo pass is to repeat history, i.e, to repeat |
| 1093 | exactly the same set of changes the rawStore went thru right before it |
| 1094 | stopped. With each log record that is encountered in the redo pass: |
| 1095 | <OL> |
| 1096 | <LI>if it isFirst(), then the transaction factory is called upon to |
| 1097 | create a new transaction object. |
| 1098 | <LI>if it needsRedo(), its doMe() is called (if it is a compensation |
| 1099 | operation, then the undoable operation needs to be created first |
| 1100 | before the doMe is called). |
| 1101 | <LI>if it isComplete(), then the transaction object is closed. |
| 1102 | </OL> |
| 1103 | |
| 1104 | <P> MT - caller provides synchronization |
| 1105 | |
| 1106 | @param transFactory - the transaction factory |
| 1107 | @param redoLWM - if checkpoint seen, starting from this point |
| 1108 | on, apply redo if necessary |
| 1109 | |
| 1110 | @return the log instant of the next log record (or the instant just |
| 1111 | after the last log record). This is used to determine where the log |
| 1112 | truly ends |
| 1113 | |
| 1114 | @exception StandardException Standard Cloudscape error policy |
| 1115 | @exception IOException error reading log file |
| 1116 | @exception ClassNotFoundException log file corrupted |
| 1117 | |
| 1118 | @see LogToFile#recover |
| 1119 | */ |
| 1120 | protected long redo( |
| 1121 | RawTransaction recoveryTransaction, |
| 1122 | TransactionFactory transFactory, |
| 1123 | StreamLogScan redoScan, |
| 1124 | long redoLWM, |
| 1125 | long ttabInstant) |
| 1126 | throws IOException, StandardException, ClassNotFoundException |
| 1127 | { |
| 1128 | // begin debug info |
| 1129 | if (SanityManager.DEBUG) |
| 1130 | { |
| 1131 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 1132 | { |
| 1133 | SanityManager.DEBUG( |
| 1134 | LogToFile.DBG_FLAG, |
| 1135 | "In recovery redo, redoLWM = " + redoLWM); |
| 1136 | } |
| 1137 | } |
| 1138 | |
| 1139 | int scanCount = 0; |
| 1140 | int redoCount = 0; |
| 1141 | int prepareCount = 0; |
| 1142 | int clrCount = 0; |
| 1143 | int btranCount = 0; |
| 1144 | int etranCount = 0; |
| 1145 | |
| 1146 | // end debug info |
| 1147 | |
| 1148 | TransactionId tranId = null; |
| 1149 | |
| 1150 | // the current log instant |
| 1151 | long instant = LogCounter.INVALID_LOG_INSTANT; |
| 1152 | |
| 1153 | ////////////////////////////////////////////////////////////////////// |
| 1154 | // During redo time, the byte array in the logOutputBuffer is not used. |
| 1155 | // Use it to read the log record - if it is not big enough, scan |
| 1156 | // will resize it. We could create a brand new log input stream that |
| 1157 | // has nothing to do with logIn or logOutputBuffer but that seem like |
| 1158 | // a waste of memory. |
| 1159 | ////////////////////////////////////////////////////////////////////// |
| 1160 | logIn.setData(logOutputBuffer.getByteArray()); |
| 1161 | |
| 1162 | // use this scan to reconstitute operation to be undone |
| 1163 | // when we see a CLR in the redo scan |
| 1164 | StreamLogScan undoScan = null; |
| 1165 | Loggable op = null; |
| 1166 | long logEnd = 0; // we need to determine the log's true end |
| 1167 | |
| 1168 | try |
| 1169 | { |
| 1170 | |
| 1171 | // scan the log forward in redo pass and go to the end |
| 1172 | LogRecord record; |
| 1173 | while((record = |
| 1174 | redoScan.getNextRecord(logIn, null, 0)) |
| 1175 | != null) |
| 1176 | { |
| 1177 | scanCount++; |
| 1178 | long undoInstant = 0; |
| 1179 | |
| 1180 | // last known good instant |
| 1181 | instant = redoScan.getInstant(); |
| 1182 | |
| 1183 | // last known good log end |
| 1184 | logEnd = redoScan.getLogRecordEnd(); |
| 1185 | |
| 1186 | |
| 1187 | // NOTE NOTE -- be very careful about the undoInstant, it is |
| 1188 | // read off the input stream in this debug section. |
| 1189 | // if we change the log format we will need to change the way |
| 1190 | // the undo instant is gotten. Also, once it is read off, it |
| 1191 | // should not be read from the stream any more |
| 1192 | // NOTE NOTE |
| 1193 | if (SanityManager.DEBUG) |
| 1194 | { |
| 1195 | if (SanityManager.DEBUG_ON(LogToFile.DUMP_LOG_ONLY) || |
| 1196 | SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 1197 | |
| 1198 | { |
| 1199 | if (SanityManager.DEBUG_ON(LogToFile.DUMP_LOG_ONLY)) |
| 1200 | SanityManager.DEBUG_SET(LogToFile.DBG_FLAG); |
| 1201 | |
| 1202 | op = record.getLoggable(); |
| 1203 | tranId = record.getTransactionId(); |
| 1204 | if (record.isCLR()) |
| 1205 | { |
| 1206 | // !!!!!!! this moves the file pointer |
| 1207 | undoInstant = logIn.readLong(); |
| 1208 | |
| 1209 | SanityManager.DEBUG( |
| 1210 | LogToFile.DBG_FLAG, |
| 1211 | "scanned " + tranId + " : " + op + |
| 1212 | " instant = " + |
| 1213 | LogCounter.toDebugString(instant) + |
| 1214 | " undoInstant : " + |
| 1215 | LogCounter.toDebugString(undoInstant)); |
| 1216 | } |
| 1217 | else |
| 1218 | { |
| 1219 | SanityManager.DEBUG( |
| 1220 | LogToFile.DBG_FLAG, |
| 1221 | "scanned " + tranId + " : " + op + |
| 1222 | " instant = " + |
| 1223 | LogCounter.toDebugString(instant) |
| 1224 | + " logEnd = " + |
| 1225 | LogCounter.toDebugString(logEnd) |
| 1226 | + " logIn at " + logIn.getPosition() |
| 1227 | + " available " + logIn.available()); |
| 1228 | } |
| 1229 | |
| 1230 | // we only want to dump the log, don't touch it |
| 1231 | if (SanityManager.DEBUG_ON(LogToFile.DUMP_LOG_ONLY)) |
| 1232 | continue; |
| 1233 | } |
| 1234 | } |
| 1235 | |
| 1236 | // if the redo scan is between the undoLWM and redoLWM, we only |
| 1237 | // need to redo begin and end tran. Everything else has |
| 1238 | // already been flushed by checkpoint |
| 1239 | if (redoLWM != |
| 1240 | LogCounter.INVALID_LOG_INSTANT && instant < redoLWM) |
| 1241 | { |
| 1242 | if (!(record.isFirst() || |
| 1243 | record.isComplete() || |
| 1244 | record.isPrepare())) |
| 1245 | { |
| 1246 | continue; |
| 1247 | } |
| 1248 | } |
| 1249 | |
| 1250 | // get the transaction |
| 1251 | tranId = record.getTransactionId(); |
| 1252 | |
| 1253 | // if this transaction is known to the transaction factory, make |
| 1254 | // the recoveryTransaction assume its identitiy and properties |
| 1255 | // otherwise, make it known to the transaction factory |
| 1256 | if (!transFactory.findTransaction(tranId, recoveryTransaction)) |
| 1257 | { |
| 1258 | // transaction not found |
| 1259 | |
| 1260 | if (redoLWM != LogCounter.INVALID_LOG_INSTANT && |
| 1261 | instant < redoLWM && |
| 1262 | (record.isPrepare() || record.isComplete())) |
| 1263 | { |
| 1264 | // What is happening here is that a transaction that |
| 1265 | // started before the undoLWM has commited by the time |
| 1266 | // the checkpoint undoLWM was taken. Hence, we only |
| 1267 | // see the tail end of its log record and its endXact |
| 1268 | // record. |
| 1269 | // |
| 1270 | // NOTE: |
| 1271 | // Since we didn't see its beginXact, we cannot do the |
| 1272 | // endXact's doMe either. Also if the endXact, is |
| 1273 | // actually just a prepare, we don't need to do |
| 1274 | // anything as the transaction will commit or abort |
| 1275 | // prior to point we are recovering to. |
| 1276 | // If it is deemed necessary to do the endXact's doMe, |
| 1277 | // then we should start the transaction right here. |
| 1278 | // For now, just completely ignore this transaction |
| 1279 | // |
| 1280 | etranCount++; |
| 1281 | |
| 1282 | continue; |
| 1283 | } |
| 1284 | |
| 1285 | if ((ttabInstant == LogCounter.INVALID_LOG_INSTANT) && |
| 1286 | !record.isFirst()) |
| 1287 | { |
| 1288 | throw StandardException.newException( |
| 1289 | SQLState.LOG_UNEXPECTED_RECOVERY_PROBLEM, |
| 1290 | MessageService.getTextMessage(MessageId.LOG_RECORD_NOT_FIRST,tranId)); |
| 1291 | |
| 1292 | } |
| 1293 | |
| 1294 | if (SanityManager.DEBUG) |
| 1295 | { |
| 1296 | // if we dumped the transaction table but see a non |
| 1297 | // BeginXact record after the transaction table dump |
| 1298 | // instant, error. |
| 1299 | if (ttabInstant != LogCounter.INVALID_LOG_INSTANT) |
| 1300 | { |
| 1301 | if (instant > ttabInstant && !record.isFirst()) |
| 1302 | { |
| 1303 | SanityManager.THROWASSERT( |
| 1304 | "log record is Not first but transaction " + |
| 1305 | "is not in transaction table (2) : " + tranId); |
| 1306 | } |
| 1307 | |
| 1308 | // If we dump the transaction table and the table |
| 1309 | // does not have the transaction, and we see this |
| 1310 | // beginXact before the ttab instant, we could have |
| 1311 | // igored it because we "know" that we should see |
| 1312 | // the endXact before the ttab instant also. |
| 1313 | // Leave it in just in case. |
| 1314 | } |
| 1315 | } |
| 1316 | btranCount++; |
| 1317 | |
| 1318 | // the long transaction ID is embedded in the beginXact log |
| 1319 | // record. The short ID is stored in the log record. |
| 1320 | recoveryTransaction.setTransactionId( |
| 1321 | record.getLoggable(), tranId); |
| 1322 | |
| 1323 | } |
| 1324 | else |
| 1325 | { |
| 1326 | // recoveryTransaction found |
| 1327 | |
| 1328 | if ((ttabInstant == LogCounter.INVALID_LOG_INSTANT) && |
| 1329 | record.isFirst()) |
| 1330 | { |
| 1331 | throw StandardException.newException( |
| 1332 | SQLState.LOG_UNEXPECTED_RECOVERY_PROBLEM, |
| 1333 | MessageService.getTextMessage(MessageId.LOG_RECORD_FIRST, |
| 1334 | tranId)); |
| 1335 | |
| 1336 | |
| 1337 | } |
| 1338 | |
| 1339 | if (SanityManager.DEBUG) |
| 1340 | { |
| 1341 | if (ttabInstant != LogCounter.INVALID_LOG_INSTANT && |
| 1342 | instant > ttabInstant && |
| 1343 | record.isFirst()) |
| 1344 | { |
| 1345 | SanityManager.THROWASSERT( |
| 1346 | "log record is first but transaction is " + |
| 1347 | "already in transaction table (3): " + tranId); |
| 1348 | } |
| 1349 | |
| 1350 | if (record.isPrepare()) |
| 1351 | prepareCount++; |
| 1352 | } |
| 1353 | |
| 1354 | // if we have a transaction table dumped with the |
| 1355 | // checkpoint log record, then during the redo scan we may |
| 1356 | // see the beginXact of a transaction which is already in |
| 1357 | // the transaction table, just ignore it if it is after the |
| 1358 | // redoLWM but before the transaction table instant. We |
| 1359 | // still need to redo any database changes but since the |
| 1360 | // transaction is already recorded in the transaction |
| 1361 | // table, ignore it. |
| 1362 | // |
| 1363 | if (record.isFirst()) |
| 1364 | { |
| 1365 | btranCount++; |
| 1366 | continue; |
| 1367 | } |
| 1368 | } |
| 1369 | |
| 1370 | op = record.getLoggable(); |
| 1371 | |
| 1372 | if (SanityManager.DEBUG) |
| 1373 | { |
| 1374 | if (!record.isCLR()) |
| 1375 | { |
| 1376 | if (logIn.available() < 4) |
| 1377 | { |
| 1378 | SanityManager.THROWASSERT( |
| 1379 | "not enough bytes read in : " + |
| 1380 | logIn.available() + |
| 1381 | " for " + op + " instant " + |
| 1382 | LogCounter.toDebugString(instant)); |
| 1383 | } |
| 1384 | } |
| 1385 | } |
| 1386 | |
| 1387 | if (SanityManager.DEBUG) |
| 1388 | { |
| 1389 | SanityManager.ASSERT( |
| 1390 | !recoveryTransaction.handlesPostTerminationWork(), |
| 1391 | "recovery transaction handles post termination work"); |
| 1392 | } |
| 1393 | |
| 1394 | if (op.needsRedo(recoveryTransaction)) |
| 1395 | { |
| 1396 | redoCount++; |
| 1397 | |
| 1398 | if (record.isCLR()) |
| 1399 | { |
| 1400 | clrCount++; |
| 1401 | |
| 1402 | // the log operation is not complete, the operation to |
| 1403 | // undo is stashed away at the undoInstant. |
| 1404 | // Reconstitute that first. |
| 1405 | |
| 1406 | if (SanityManager.DEBUG) |
| 1407 | SanityManager.ASSERT(op instanceof Compensation); |
| 1408 | |
| 1409 | |
| 1410 | // this value may be set by sanity xxxx |
| 1411 | if (undoInstant == 0) |
| 1412 | undoInstant = logIn.readLong(); |
| 1413 | |
| 1414 | if (undoScan == null) |
| 1415 | { |
| 1416 | undoScan = (StreamLogScan) |
| 1417 | logFactory.openForwardsScan( |
| 1418 | undoInstant,(LogInstant)null); |
| 1419 | } |
| 1420 | else |
| 1421 | { |
| 1422 | undoScan.resetPosition(new LogCounter(undoInstant)); |
| 1423 | } |
| 1424 | |
| 1425 | // undoScan now positioned at the beginning of the log |
| 1426 | // record was rolled back by this CLR. |
| 1427 | // The scan is a forward one so getNextRecord will get |
| 1428 | // the log record that needs to be rolled back. |
| 1429 | |
| 1430 | // reuse the buffer in logIn and logIn since CLR |
| 1431 | // has no optional data and has no use for them anymore |
| 1432 | logIn.clearLimit(); |
| 1433 | LogRecord undoRecord = |
| 1434 | undoScan.getNextRecord(logIn, null, 0); |
| 1435 | |
| 1436 | Undoable undoOp = undoRecord.getUndoable(); |
| 1437 | |
| 1438 | if (SanityManager.DEBUG) |
| 1439 | { |
| 1440 | SanityManager.DEBUG( |
| 1441 | LogToFile.DBG_FLAG, |
| 1442 | "Redoing CLR: undoInstant = " + |
| 1443 | LogCounter.toDebugString(undoInstant) + |
| 1444 | " clrinstant = " + |
| 1445 | LogCounter.toDebugString(instant)); |
| 1446 | |
| 1447 | SanityManager.ASSERT( |
| 1448 | undoRecord.getTransactionId().equals(tranId)); |
| 1449 | |
| 1450 | SanityManager.ASSERT(undoOp != null); |
| 1451 | } |
| 1452 | |
| 1453 | ((Compensation)op).setUndoOp(undoOp); |
| 1454 | } |
| 1455 | |
| 1456 | // at this point, logIn points to the optional |
| 1457 | // data of the loggable that is to be redone or to be |
| 1458 | // rolled back |
| 1459 | |
| 1460 | if (SanityManager.DEBUG) |
| 1461 | { |
| 1462 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 1463 | { |
| 1464 | SanityManager.DEBUG( |
| 1465 | LogToFile.DBG_FLAG, |
| 1466 | "redoing " + op + |
| 1467 | " instant = " + |
| 1468 | LogCounter.toDebugString(instant)); |
| 1469 | } |
| 1470 | } |
| 1471 | |
| 1472 | int dataLength = logIn.readInt(); |
| 1473 | logIn.setLimit(logIn.getPosition(), dataLength); |
| 1474 | |
| 1475 | // even though the log has already been written, we need to |
| 1476 | // tie the page to the log stream so that if redo failed |
| 1477 | // for some reasons, the log factory's corruption will stop |
| 1478 | // the corrupt page from flushing to disk. |
| 1479 | |
| 1480 | op.doMe( |
| 1481 | recoveryTransaction, |
| 1482 | new LogCounter(instant), logIn); |
| 1483 | |
| 1484 | op.releaseResource(recoveryTransaction); |
| 1485 | |
| 1486 | op = null; |
| 1487 | } |
| 1488 | |
| 1489 | // RESOLVE: to speed up undo, may want to update the |
| 1490 | // LastLogInstant in the transaction table. |
| 1491 | // Right now, undo always start from the end of the log. |
| 1492 | |
| 1493 | // one last thing, if this is the last log record of the |
| 1494 | // transaction, then commit the transaction and clean up |
| 1495 | // |
| 1496 | // 'commit' even though the transaction maybe a rollback |
| 1497 | // because we already did all the rollback work when redoing |
| 1498 | // the CLRs. Commit will only flush the log if this session |
| 1499 | // has written any transaction, so in this case, it is a noop. |
| 1500 | if (record.isComplete()) |
| 1501 | { |
| 1502 | etranCount++; |
| 1503 | |
| 1504 | if (SanityManager.DEBUG) |
| 1505 | SanityManager.ASSERT( |
| 1506 | !recoveryTransaction.handlesPostTerminationWork(), |
| 1507 | "recovery xact handles post termination work"); |
| 1508 | |
| 1509 | recoveryTransaction.commit(); |
| 1510 | } |
| 1511 | } // while redoScan.getNextRecord() != null |
| 1512 | |
| 1513 | // If the scan ended in an empty file, update logEnd to reflect that |
| 1514 | // in order to avoid to continue logging to an older file |
| 1515 | long end = redoScan.getLogRecordEnd(); |
| 1516 | if (end != LogCounter.INVALID_LOG_INSTANT |
| 1517 | && (LogCounter.getLogFileNumber(logEnd) |
| 1518 | < LogCounter.getLogFileNumber(end))) { |
| 1519 | logEnd = end; |
| 1520 | } |
| 1521 | } |
| 1522 | catch (StandardException se) |
| 1523 | { |
| 1524 | throw StandardException.newException( |
| 1525 | SQLState.LOG_REDO_FAILED, se, op); |
| 1526 | } |
| 1527 | finally |
| 1528 | { |
| 1529 | // close all the io streams |
| 1530 | redoScan.close(); |
| 1531 | redoScan = null; |
| 1532 | |
| 1533 | if (undoScan != null) |
| 1534 | { |
| 1535 | undoScan.close(); |
| 1536 | undoScan = null; |
| 1537 | } |
| 1538 | |
| 1539 | if (op != null) |
| 1540 | op.releaseResource(recoveryTransaction); |
| 1541 | |
| 1542 | } |
| 1543 | |
| 1544 | if (SanityManager.DEBUG) |
| 1545 | { |
| 1546 | if (SanityManager.DEBUG_ON(LogToFile.DBG_FLAG)) |
| 1547 | { |
| 1548 | SanityManager.DEBUG( |
| 1549 | LogToFile.DBG_FLAG, |
| 1550 | "----------------------------------------------------\n" + |
| 1551 | "End of recovery redo\n" + |
| 1552 | "Scanned = " + scanCount + " log records" + |
| 1553 | ", redid = " + redoCount + |
| 1554 | " ( clr = " + clrCount + " )" + |
| 1555 | " begintran = " + btranCount + |
| 1556 | " endtran = " + etranCount + |
| 1557 | " preparetran = " + prepareCount + |
| 1558 | "\n log ends at " + LogCounter.toDebugString(logEnd) + |
| 1559 | "\n----------------------------------------------------\n"); |
| 1560 | } |
| 1561 | } |
| 1562 | |
| 1563 | if (SanityManager.DEBUG) |
| 1564 | { |
| 1565 | // make sure logEnd and instant is consistent |
| 1566 | if (instant != LogCounter.INVALID_LOG_INSTANT) |
| 1567 | { |
| 1568 | SanityManager.ASSERT( |
| 1569 | LogCounter.getLogFileNumber(instant) < |
| 1570 | LogCounter.getLogFileNumber(logEnd) || |
| 1571 | (LogCounter.getLogFileNumber(instant) == |
| 1572 | LogCounter.getLogFileNumber(logEnd) && |
| 1573 | LogCounter.getLogFilePosition(instant) <= |
| 1574 | LogCounter.getLogFilePosition(logEnd))); |
| 1575 | } |
| 1576 | else |
| 1577 | { |
| 1578 | SanityManager.ASSERT(logEnd == LogCounter.INVALID_LOG_INSTANT); |
| 1579 | } |
| 1580 | } |
| 1581 | |
| 1582 | // logEnd is the last good log record position in the log |
| 1583 | return logEnd; |
| 1584 | } |
| 1585 | |
| 1586 | |
| 1587 | /** |
| 1588 | Read the next log record from the scan. |
| 1589 | |
| 1590 | <P>MT - caller must provide synchronization (right now, it is only |
| 1591 | called in recovery to find the checkpoint log record. When this method |
| 1592 | is called by a more general audience, MT must be revisited). |
| 1593 | |
| 1594 | @param scan an opened log scan |
| 1595 | @param size estimated size of the log record |
| 1596 | |
| 1597 | @return the log operation that is the next in the scan, or null if no |
| 1598 | more log operation in the log scan |
| 1599 | |
| 1600 | @exception IOException Error reading the log file |
| 1601 | @exception StandardException Standard Cloudscape error policy |
| 1602 | @exception ClassNotFoundException log corrupted |
| 1603 | */ |
| 1604 | protected Loggable readLogRecord(StreamLogScan scan, int size) |
| 1605 | throws IOException, StandardException, ClassNotFoundException |
| 1606 | { |
| 1607 | Loggable lop = null; |
| 1608 | |
| 1609 | ArrayInputStream logInputBuffer = new ArrayInputStream(new byte[size]); |
| 1610 | |
| 1611 | LogRecord record = scan.getNextRecord(logInputBuffer, null, 0); |
| 1612 | if (record != null) |
| 1613 | lop = record.getLoggable(); |
| 1614 | return lop; |
| 1615 | } |
| 1616 | |
| 1617 | } |