lockfile.java

纯Java的数据库

    public static final LockFile newLockFileLock(final String path)
    throws HsqlException {

        LockFile lockFile = null;

        try {
            lockFile = LockFile.newLockFile(path + ".lck");
        } catch (LockFile.BaseException e) {
            throw Trace.error(Trace.LOCK_FILE_ACQUISITION_FAILURE,
                              e.getMessage());
        }

        boolean locked = false;

        try {
            locked = lockFile.tryLock();
        } catch (LockFile.BaseException e) {
            throw Trace.error(Trace.LOCK_FILE_ACQUISITION_FAILURE,
                              e.getMessage());
        }

        // Paranoia mode: In theory, this case can't happen, given the way
        // tryLock now works; by all current understanding of the involved API
        // contracts, an exception will always be thrown instead by the code
        // above.
        if (!locked) {
            throw Trace.error(Trace.LOCK_FILE_ACQUISITION_FAILURE,
                              lockFile.toString());
        }

        return lockFile;
    }

    /**
     * Checks whether the underlying file is an HSQLDB lock file and, if so,
     * whether its heartbeat timestamp is live (is, as far as can be known,
     * presumably in use by another <tt>LockFile</tt> instance) or stale. <p>
     *
     * The check conforms to the following rules: <p>
     *
     * <ol>
     * <li>If the parameter <tt>withCreateNewFile</tt> is true, {@link
     *     java.io.File#createNewFile()} is available and its invocation
     *     upon this object's <tt>file</tt> object indicates the underlying
     *     file was atomically created if and only if it did not yet exist,
     *     then return immediately (we have won the <em>race</em> to establish
     *     a lock file). <p>
     *
     * <li>Test again if the file exists, returning immediately if it does not
     *     (there's no file and hence no heartbeat to check). <p>
     *
     *     An immediate return can occur here only under pre-JDK 1.2 runtimes;
     *     or when the underlying platform does not correctly support
     *     <tt>File.createNewFile()</tt>; or when the underlying file is deleted
     *     within a very short time after i.), above (typically on the order of
     *     microseconds). <p>
     *
     *     If the underlying platform employs a kernel-enforced mandatory file
     *     locking blanket policy for open files (e.g. <em>Windows</em><sup>tm
     *     </sup>), then this is likely a non-issue. And if this case makes
     *     possible a race condition with another <tt>LockFile</tt> object
     *     (because the test for existence yeilds false and subsequent file
     *     creation is not atomic relative to all other file system actions), it
     *     is still <em>very</em> unlikely that so unfortunate a timing will
     *     occur as to allow simultaneous lock conditions to be established.
     *     Finally, if some non-<tt>LockFile</tt> entity deleted the file, then
     *     there are much worse things to worry about, in particular that the
     *     files this object is supposed to protect are in reality subject to
     *     arbitrary external modification and deletion by some uncooperative
     *     process.<p>
     *
     * <li>If a Java security exception is thrown while testing for existence,
     *     it is rethrown as a <tt>FileSecurityException</tt>.
     *
     * <li>Read the file's length.
     *
     * <li>If a Java security exception is thrown reading length, it is rethrown
     *     as a <tt>FileSecurityException</tt> (it <em>is</em> possible somebody
     *     concurrently refreshed the system Policy in the interim).
     *
     * <li>If the file does not have the expected length, a
     *     <tt>WrongLengthException</tt> is thrown (we're trying to check
     *     something that is not an HSQLDB lock file).
     *
     * <li>Open an input steam to read the file's <tt>MAGIC</tt> and heartbeat
     *     timestamp values.
     *
     * <li>If a file not found exception is thrown above, it is rethrown as an
     *     <tt>UnexpectedFileNotFoundException</tt> (we've already tested for
     *     existence).
     *
     * <li>If a Java security exception is thrown above, it is rethrown as a
     *     <tt>FileSecurityException</tt> (it <em>is</em> possible somebody
     *     concurrently refreshed the system Policy in the interim).
     *
     * <li>Read the <tt>MAGIC</tt> value.
     *
     * <li>If an end of file exepction is thrown above, it is rethrown as an
     *     <tt>UnexpectedEndOfFileException</tt> (we've already tested the
     *     length... did someone truncate the file in the interim?).
     *
     * <li>If an I/O exception is thrown, it is rethrown as an
     *     <tt>UnexpectedFileIOException</tt> (we've already tested for
     *     existence, length and successfully opened a stream...did someone,
     *     for example, force unmount or physically remove the underlying device
     *     in the interim?)
     *
     * <li>If the value read in does not match the expected <tt>MAGIC</tt> value,
     *     a <tt>WrongMagicException</tt> is thrown (we're trying to check
     *     something that is not an HSQLDB lock file).
     *
     * <li>Read the heartbeat timestamp.
     *
     * <li>If a Java security exception is thrown above, it is rethrown as a
     *     <tt>FileSecurityException</tt> (it <em>is</em> possible somebody
     *     concurrently refreshed the system Policy in the interim).
     *
     * <li>If an end of file exection is thrown above, it is rethrown as an
     *     <tt>UnexpectedEndOfFileException</tt> (we've already tested the
     *     length... did someone truncate the file in the interim?).
     *
     * <li>If an I/O exception is thrown, it is rethrown as an
     *     <tt>UnexpectedFileIOException</tt> (we've already tested for
     *     existence, length and successfully opened a stream...did someone,
     *     for example, force unmount or physically remove the underlying device
     *     in the interim?)
     *
     * <li>If the timestamp read in is less than or equal to
     *     {@link #HEARTBEAT_INTERVAL_PADDED} milliseconds into the past or
     *     future, then a <tt>LockHeldExternallyException</tt> is thrown.
     *
     * <li>Otherwise, this method simply returns.
     * </ol>
     *
     * @param withCreateNewFile if <tt>true</tt>, attempt to employ
     *      <tt>File.createNewFile()</tt> as part of the check so as to
     *      eliminate potential race conditions when establising a new
     *      lock file
     * @throws FileSecurityException if the check fails due to a Java
     *      security permission check failure
     * @throws LockHeldExternallyException if it is determined that the
     *      file's heartbeat timestamp is less than
     *      <tt>HEARTBEAT_INTERVAL_PADDED</tt> into the past (or future)
     * @throws UnexpectedEndOfFileException if an <tt>EOFExceoption</tt> is
     *      thrown while reading either the magic or heartbeat timestamp values
     * @throws UnexpectedFileIOException if an <tt>IOException</tt> other than
     *      <tt>EOFException</tt> is thrown while reading either the magic or
     *      heartbeat timestamp values
     * @throws UnexpectedFileNotFoundException if a
     *      <tt>FileNotFoundException</tt> is thrown while attempting to open a
     *      stream to read the underlying file's magic and heartbeat timestamp
     *      values
     * @throws WrongLengthException if it is determined that the length
     *      of the file does not equal {@link #USED_REGION}
     * @throws WrongMagicException if it is determined that the file's
     *      content does not start with {@link #MAGIC}.
     */
    private final void checkHeartbeat(boolean withCreateNewFile)
    throws LockFile.FileSecurityException,
           LockFile.LockHeldExternallyException,
           LockFile.UnexpectedEndOfFileException,
           LockFile.UnexpectedFileIOException,
           LockFile.UnexpectedFileNotFoundException,
           LockFile.WrongLengthException, LockFile.WrongMagicException {

        long now;
        long lastHeartbeat;
        long length = 0;

//#ifdef JAVA2FULL
        try {
            if (withCreateNewFile) {
                try {
                    file.createNewFile();

                    return;
                } catch (IOException ioe) {}
            }

            if (!file.exists()) {
                return;
            }

            length = file.length();
        } catch (SecurityException se) {
            throw new FileSecurityException(this, "checkHeartbeat", se);
        }

//#else
/*
        if (!file.exists()) {
            return;
        }

        length = file.length();
*/

//#endif JAVA2
        if (length != USED_REGION) {
            throw new WrongLengthException(this, "checkHeartbeat", length);
        }

        // Compute the current wall clock time *first* to reduce possibility
        // of unwanted time dilation effects introduced, for example,
        // by intervening thread or process context switches under CPU
        // bursts.
        //
        // Example:
        //
        // Say currentTimeMillis is actually somewhere in (-0.5 and 0.5]
        // and another LockFile concurrently writes a 0-valued heartbeat
        // timestamp.
        //
        // Then, if readHeartbeat comes first here, happens to 'win the race
        // condition' (reads the previous heartbeat: -10,000) and an intervening
        // switch causes greater than ~0.5 millisecond elapsed time to
        // be experienced between readHeartbeat and currentTimeMillis, then
        // currentTimeMillis will be computed as n (n > 0), and (now -
        // lastHearbeat) will be HEARTBEAT_INTERVAL + n, instead of
        // HEARTBEAT_INTERVAL.
        //
        // Now, let n be greater than (HEARTBEAT_INTERVAL_PADDED -
        // HEARTBEAT_INTERVAL).
        //
        // Then the check will succeed, although it should fail.
        //
        // On the other hand, if currentTimeMillis is computed first, the
        // worst than can happen is a false positive indication that
        // the read heartbeat timestamp value was written by a live LockFile
        // instance.
        //
        now           = System.currentTimeMillis();
        lastHeartbeat = readHeartbeat();

        // Using padded interval to further reduce corner case effects,
        // now that heartbeat polling is in effect.
        //
        // Basically, it is absolutely essential to fail when a lock really is
        // still held elsewhere, so it is OK to fail on corner cases where
        // the last written heartbeat is very close to HEARTBEAT_INTERVAL
        // in the past and it is possible that timing jitters make it uncertain
        // whether the lock really is still held.
        if (Math.abs(now - lastHeartbeat) <= (HEARTBEAT_INTERVAL_PADDED)) {
            throw new LockHeldExternallyException(this, "checkHeartbeat", now,
                                                  lastHeartbeat);
        }
    }

    /**
     * Closes this object's {@link #raf RandomAccessFile}. <p>
     *
     * As a side-effect, the associated <tt>FileChannel</tt> object, if any,
     * is closed as well.
     *
     * @throws UnexpectedFileIOException if an <tt>IOException</tt> is thrown
     */
    private final void closeRAF() throws LockFile.UnexpectedFileIOException {

        if (raf != null) {
            try {
                raf.close();
            } catch (IOException ex) {
                throw new UnexpectedFileIOException(this, "closeRAF", ex);
            } finally {
                raf = null;
            }
        }
    }

    /**
     * Provides any optional locking actions for the {@link #tryLock()
     * tryLock()} template method. <p>
     *
     * Descendents are free to provide additional functionality here,
     * using the following rules: <p>
     *
     * <b>PRE:</b><p>
     *
     * This method is called only from <tt>tryLock()</tt> and it is called if
     * and only if <tt>tryLock()</tt> successfully invokes
     * <tt>pollHeartbeat()</tt> and <tt>openRAF()</tt> first. <p>
     *
     * From this, it can be inferred that upon entry: <p>
     *
     * <ol>
     * <li><tt>locked == false</tt>.
     * <li><tt>raf</tt> is a non-null instance that can be used to get a
     *     <tt>FileChannel</tt> instance, if desired.
     * <li>the underlying file either did not exist before invoking
     *     <tt>openRAF()</tt> or it was a valid but stale HSQLDB lock file
     *     because it:
     *
     *     <ol style="list-style-type: lower-roman">
     *     <li>did exist,
     *     <li>was readable on <tt>USED_REGION</tt>,
     *     <li>had the expected length and <tt>MAGIC</tt> value and
     *     <li>had a stale heartbeat timestamp value.
     *     </ol>
     * </ol> <p>
     *
     * Further, it can be assumed that this object's heatbeat task is definitely
     * cancelled and/or has never been scheduled at this point, so whatever
     * timestamp is recorded in the lock file, if it did pre-exist, was written
     * by a different <tt>LockFile</tt> instance or as the result of a previous,
     * successful <tt>tryLock()</tt> invocation upon this <tt>LockFile</tt>
     * instance. <p>
     *
     * Finally, it is important that this method does not rethrow any exceptions
     * it encounters as unchecked exceptions to the calling context. <p>
     *
     * <b>POST:</b><p>
     *