SQLite: Check-in [bb0b6021]

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview

Comment:	Instead of transitioning to RECOVER state from CHECKPOINT when a recovery is required, perform the recovery while holding the CHECKPOINT lock.
Downloads:	Tarball \| ZIP archive \| SQL archive
Timelines:	family \| ancestors \| descendants \| both \| trunk
Files:	files \| file ages \| folders
SHA1:	bb0b6021e721c2e4f1533ab18381604c6b7b31d5
User & Date:	dan 2010-05-06 18:48:28

Context

2010-05-06
19:04		Updates to the VFS SHM locking documentation. check-in: 9927ce42 user: drh tags: trunk
18:48		Instead of transitioning to RECOVER state from CHECKPOINT when a recovery is required, perform the recovery while holding the CHECKPOINT lock. check-in: bb0b6021 user: dan tags: trunk
18:27		If recovery is run before a checkpoint, change back to a CHECKPOINT lock before performing the actual checkpoint. check-in: dc98ee16 user: dan tags: trunk

Changes

Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to doc/vfs-shm.txt.

(11h)  READ_FULL          UNLOCK          UNLOCK
(11i)  READ_FULL          WRITE           WRITE
(11j)  READ_FULL          RECOVER         RECOVER
(11k)  WRITE              READ            READ
(11l)  PENDING            UNLOCK          UNLOCK
(11m)  PENDING            CHECKPOINT      CHECKPOINT
(11n)  CHECKPOINT         UNLOCK          UNLOCK
(11o)  CHECKPOINT         RECOVER         RECOVER
(11p)  RECOVER            READ            READ
(11q)  RECOVER            CHECKPOINT      CHECKPOINT

These 17 transitions are all that needs to be supported.  The lock
manager implementation can assert that fact.  The other 25 possible
transitions among the 7 locking states will never occur.

The rules above are sufficient for correctness.  For maximum concurrency,
the following additional considerations apply:








<
|
<

(11h)  READ_FULL          UNLOCK          UNLOCK
(11i)  READ_FULL          WRITE           WRITE
(11j)  READ_FULL          RECOVER         RECOVER
(11k)  WRITE              READ            READ
(11l)  PENDING            UNLOCK          UNLOCK
(11m)  PENDING            CHECKPOINT      CHECKPOINT
(11n)  CHECKPOINT         UNLOCK          UNLOCK

(11o)  RECOVER            READ            READ


These 17 transitions are all that needs to be supported.  The lock
manager implementation can assert that fact.  The other 25 possible
transitions among the 7 locking states will never occur.

The rules above are sufficient for correctness.  For maximum concurrency,
the following additional considerations apply:

Changes to src/os_unix.c.

        p->lockState = SQLITE_SHM_WRITE;
      }
      break;
    }
    case SQLITE_SHM_CHECKPOINT: {
      assert( p->lockState==SQLITE_SHM_UNLOCK
           || p->lockState==SQLITE_SHM_PENDING
           || p->lockState==SQLITE_SHM_RECOVER );
      if( p->lockState==SQLITE_SHM_RECOVER ){
        unixShmUnlock(pFile, p, UNIX_SHM_C);
        p->lockState = SQLITE_SHM_CHECKPOINT;
        rc = SQLITE_OK;
      }
      if( p->lockState==SQLITE_SHM_UNLOCK ){
        rc = unixShmExclusiveLock(pFile, p, UNIX_SHM_B|UNIX_SHM_C);
        if( rc==SQLITE_OK ){
          p->lockState = SQLITE_SHM_PENDING;
        }
      }
      if( p->lockState==SQLITE_SHM_PENDING ){
................................................................................
      }
      break;
    }
    default: {
      assert( desiredLock==SQLITE_SHM_RECOVER );
      assert( p->lockState==SQLITE_SHM_READ
           || p->lockState==SQLITE_SHM_READ_FULL
           || p->lockState==SQLITE_SHM_CHECKPOINT );
      assert( sqlite3_mutex_held(pFile->mutexBuf) );
      rc = unixShmExclusiveLock(pFile, p, UNIX_SHM_C);
      if( rc==SQLITE_OK ){
        p->lockState = SQLITE_SHM_RECOVER;
      }
      break;
    }








|
<
<
<
<
<







 







|

        p->lockState = SQLITE_SHM_WRITE;
      }
      break;
    }
    case SQLITE_SHM_CHECKPOINT: {
      assert( p->lockState==SQLITE_SHM_UNLOCK
           || p->lockState==SQLITE_SHM_PENDING
      );





      if( p->lockState==SQLITE_SHM_UNLOCK ){
        rc = unixShmExclusiveLock(pFile, p, UNIX_SHM_B|UNIX_SHM_C);
        if( rc==SQLITE_OK ){
          p->lockState = SQLITE_SHM_PENDING;
        }
      }
      if( p->lockState==SQLITE_SHM_PENDING ){
................................................................................
      }
      break;
    }
    default: {
      assert( desiredLock==SQLITE_SHM_RECOVER );
      assert( p->lockState==SQLITE_SHM_READ
           || p->lockState==SQLITE_SHM_READ_FULL
      );
      assert( sqlite3_mutex_held(pFile->mutexBuf) );
      rc = unixShmExclusiveLock(pFile, p, UNIX_SHM_C);
      if( rc==SQLITE_OK ){
        p->lockState = SQLITE_SHM_RECOVER;
      }
      break;
    }

Changes to src/wal.c.

** The caller must hold RECOVER lock on the wal-index file.
*/
static int walIndexRecover(Wal *pWal){
  int rc;                         /* Return Code */
  i64 nSize;                      /* Size of log file */
  WalIndexHdr hdr;              /* Recovered wal-index header */

  assert( pWal->lockState==SQLITE_SHM_RECOVER );
  memset(&hdr, 0, sizeof(hdr));

  rc = sqlite3OsFileSize(pWal->pFd, &nSize);
  if( rc!=SQLITE_OK ){
    return rc;
  }

................................................................................
  }

  /* If the first attempt to read the header failed, lock the wal-index
  ** file and try again. If the header checksum verification fails this
  ** time as well, run log recovery.
  */
  lockState = pWal->lockState;

  if( SQLITE_OK==(rc = walSetLock(pWal, SQLITE_SHM_RECOVER)) ){

    /* This call to walIndexTryHdr() may not return an error code, as the
    ** wal-index is already mapped. It may find that the header is invalid,
    ** but there is no chance of hitting an actual error.  */
    assert( pWal->pWiData );
    rc = walIndexTryHdr(pWal, &isValid, pChanged);
    assert( rc==SQLITE_OK );
    if( isValid==0 ){
      *pChanged = 1;
      rc = walIndexRecover(pWal);
    }

    walSetLock(pWal, lockState);

  }

  return rc;
}

/*
** Lock a snapshot.

** The caller must hold RECOVER lock on the wal-index file.
*/
static int walIndexRecover(Wal *pWal){
  int rc;                         /* Return Code */
  i64 nSize;                      /* Size of log file */
  WalIndexHdr hdr;              /* Recovered wal-index header */

  assert( pWal->lockState>SQLITE_SHM_READ );
  memset(&hdr, 0, sizeof(hdr));

  rc = sqlite3OsFileSize(pWal->pFd, &nSize);
  if( rc!=SQLITE_OK ){
    return rc;
  }

................................................................................
  }

  /* If the first attempt to read the header failed, lock the wal-index
  ** file and try again. If the header checksum verification fails this
  ** time as well, run log recovery.
  */
  lockState = pWal->lockState;
  if( lockState>SQLITE_SHM_READ
   || SQLITE_OK==(rc = walSetLock(pWal, SQLITE_SHM_RECOVER)) 
  ){
    /* This call to walIndexTryHdr() may not return an error code, as the
    ** wal-index is already mapped. It may find that the header is invalid,
    ** but there is no chance of hitting an actual error.  */
    assert( pWal->pWiData );
    rc = walIndexTryHdr(pWal, &isValid, pChanged);
    assert( rc==SQLITE_OK );
    if( isValid==0 ){
      *pChanged = 1;
      rc = walIndexRecover(pWal);
    }
    if( lockState==SQLITE_SHM_READ ){
      walSetLock(pWal, SQLITE_SHM_READ);
    }
  }

  return rc;
}

/*
** Lock a snapshot.

Changes to test/wal2.test.

  catchsql { SELECT * FROM data }
} {0 {{need xShmOpen to see this}}}
db close
tvfs delete

#-------------------------------------------------------------------------
# Test that if a database connection is forced to run recovery before it
# can perform a checkpoint, it transistions from RECOVERY->CHECKPOINT
# before doing so.
#
do_test wal2-5.1 {
  proc tvfs_cb {method args} {
    set ::shm_file [lindex $args 0]
    if {$method == "xShmLock"} { lappend ::locks [lindex $args 2] }
    return $::tvfs_cb_return
  }
................................................................................
    INSERT INTO x VALUES(1);
  }

  incr_tvfs_hdr $::shm_file 1 1
  set ::locks [list]
  execsql { PRAGMA wal_checkpoint }
  set ::locks
} {CHECKPOINT RECOVER CHECKPOINT UNLOCK}





finish_test








|
<







 







|

>
>

<

  catchsql { SELECT * FROM data }
} {0 {{need xShmOpen to see this}}}
db close
tvfs delete

#-------------------------------------------------------------------------
# Test that if a database connection is forced to run recovery before it
# can perform a checkpoint, it does not transition into RECOVER state.

#
do_test wal2-5.1 {
  proc tvfs_cb {method args} {
    set ::shm_file [lindex $args 0]
    if {$method == "xShmLock"} { lappend ::locks [lindex $args 2] }
    return $::tvfs_cb_return
  }
................................................................................
    INSERT INTO x VALUES(1);
  }

  incr_tvfs_hdr $::shm_file 1 1
  set ::locks [list]
  execsql { PRAGMA wal_checkpoint }
  set ::locks
} {CHECKPOINT UNLOCK}

db close
tvfs delete


finish_test