net.java.truevfs.kernel.impl.CacheController.scala Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of truevfs-kernel-impl Show documentation
Show all versions of truevfs-kernel-impl Show documentation
Implements the API for accessing the federated virtual file system space.
This reference implementation can get overridden by providing another
file system manager factory implementation with a higher priority on
the class path.
/*
* Copyright (C) 2005-2012 Schlichtherle IT Services.
* All rights reserved. Use is subject to license terms.
*/
package net.java.truevfs.kernel.impl
import net.java.truecommons.io._
import net.java.truecommons.logging._
import net.java.truecommons.shed._
import java.io._
import java.nio.channels._
import javax.annotation.concurrent._
import net.java.truevfs.kernel.spec._
import net.java.truevfs.kernel.spec.FsAccessOption._
import net.java.truevfs.kernel.spec.FsSyncOption._
import net.java.truevfs.kernel.spec.FsSyncOptions._
import net.java.truevfs.kernel.spec.cio._
import net.java.truevfs.kernel.spec.cio.Entry._;
import net.java.truevfs.kernel.spec.cio.Entry.Type._;
/** A selective cache for file system entries.
* Decorating a file system controller with this class has the following
* effects:
*
* - Caching and buffering for an entry needs to get activated by using the
* methods `input` or `output` with the access option
* [[net.java.truevfs.kernel.impl.FsAccessOption.CACHE]].
* - Unless a write operation succeeds, upon each read operation the entry
* data gets copied from the backing store for buffering purposes only.
* - Upon a successful write operation, the entry data gets cached for
* subsequent read operations until the file system gets `sync`ed again.
* - Entry data written to the cache is not written to the backing store
* until the file system gets `sync`ed - this is a ''write back'' strategy.
* - As a side effect, caching decouples the underlying storage from its
* clients, allowing it to create, read, update or delete the entry data
* while some clients are still busy on reading or writing the copied
* entry data.
*
* '''TO THE FUTURE ME:'''
* FOR TRUEZIP 7.5, IT TOOK ME TWO MONTHS OF CONSECUTIVE CODING, TESTING,
* DEBUGGING, ANALYSIS AND SWEATING TO GET THIS DAMN BEAST WORKING STRAIGHT!
* DON'T EVEN THINK YOU COULD CHANGE A SINGLE CHARACTER IN THIS CODE AND
* EASILY GET AWAY WITH IT!
* '''YOU HAVE BEEN WARNED!'''
*
* Well, if you really feel like changing something, run the integration test
* suite at least ten times to make sure your changes really work - I mean it!
*
* @author Christian Schlichtherle
*/
@NotThreadSafe
private trait CacheController[E <: FsArchiveEntry]
extends ArchiveController[E] {
controller: ArchiveModelAspect[E] =>
import CacheController._
protected def pool: IoBufferPool
private[this] val caches = new java.util.HashMap[FsNodeName, EntryCache]
abstract override def input(options: AccessOptions, name: FsNodeName) = {
/** This class requires ON-DEMAND LOOKUP of its delegate socket! */
class Input extends DelegatingInputSocket[Entry] {
override def socket(): AnyInputSocket = {
assert(writeLockedByCurrentThread)
var cache = caches get name
if (null eq cache) {
if (!(options get CACHE))
return CacheController.super.input(options, name)
cache = new EntryCache(name)
}
cache input options
}
}
new Input
}: AnyInputSocket
abstract override def output(options: AccessOptions, name: FsNodeName, template: Option[Entry]) = {
/** This class requires ON-DEMAND LOOKUP of its delegate socket! */
class Output extends DelegatingOutputSocket[Entry] {
override def socket(): AnyOutputSocket = {
assert(writeLockedByCurrentThread)
var cache = caches get name
if (null eq cache) {
if (!(options get CACHE))
return CacheController.super.output(options, name, template)
cache = new EntryCache(name)
}
cache output (options, template)
}
}
new Output
}: AnyOutputSocket
abstract override def make(options: AccessOptions, name: FsNodeName, tµpe: Type, template: Option[Entry]) {
assert(writeLockedByCurrentThread)
super.make(options, name, tµpe, template)
val cache = caches remove name
if (null ne cache) cache clear ()
}
abstract override def unlink(options: AccessOptions, name: FsNodeName) {
assert(writeLockedByCurrentThread)
super.unlink(options, name)
val cache = caches remove name
if (null ne cache) cache clear()
}
abstract override def sync(options: SyncOptions) {
assert(writeLockedByCurrentThread)
var preSyncEx: NeedsSyncException = null
do {
preSyncEx = null
try {
preSync(options)
} catch {
case invalidState: NeedsSyncException =>
// The target archive controller is in an invalid state because
// it reports to need a sync() while the current thread is
// actually doing a sync().
// This is expected to be a volatile event which may have been
// caused by the following scenario:
// Another thread attempted to sync() the nested target archive
// file but initially failed because the parent file system
// controller has thrown an FsNeedsLockRetryException when
// trying to close() the input or output resources for the
// target archive file.
// The other thread has then released all its file system write
// locks and is now retrying the operation but lost the race
// for the file system write lock against this thread which has
// now detected the invalid state.
// In an attempt to recover from this invalid state, the
// current thread could just step back in order to give the
// other thread a chance to complete its sync().
//throw FsNeedsLockRetryException.get(getModel());
// However, this would unnecessarily defer the current thread
// and might result in yet another thread to discover the
// invalid state, which reduces the overall performance.
// So instead, the current thread will now attempt to resolve
// the invalid state by sync()ing the target archive controller
// before preSync()ing the cache again.
logger debug ("recovering", invalidState)
preSyncEx = invalidState // trigger another iteration
}
super.sync(options.clear(CLEAR_CACHE))
} while (null ne preSyncEx)
if (caches.isEmpty) mounted = false
}
private def preSync(options: SyncOptions) {
if (0 >= caches.size()) return
val flush = !(options get ABORT_CHANGES)
val clear = !flush || (options get CLEAR_CACHE)
assert(flush || clear)
val builder = new FsSyncExceptionBuilder
val i = caches.values.iterator
while (i.hasNext) {
val cache = i next ()
if (flush) {
try {
cache flush ()
} catch {
case ex: IOException =>
throw builder fail new FsSyncException(mountPoint, ex)
}
}
if (clear) {
i remove ()
try {
cache clear ()
} catch {
case ex: IOException =>
builder warn new FsSyncWarningException(mountPoint, ex)
}
}
}
builder check ()
}
/** A cache for the contents of an individual archive entry. */
private final class EntryCache(val name: FsNodeName) {
val cache = CacheEntry.Strategy.WriteBack.newCacheEntry(pool)
def flush() { cache flush () }
def clear() { cache release () }
def register() { caches put (name, this) }
def input(options: AccessOptions) = {
/** This class requires LAZY INITIALIZATION of its channel, but NO
* automatic decoupling on exceptions!
*/
final class Input extends DelegatingInputSocket[Entry] {
private[this] val _options = options clear CACHE // consume
override lazy val socket = CacheController.super.input(_options, name)
override def stream(peer: AnyOutputSocket) = {
assert(writeLockedByCurrentThread)
final class Stream extends DecoratingInputStream(socket stream peer) {
assert(mounted)
override def close() {
assert(writeLockedByCurrentThread)
in close ()
register()
}
}
new Stream
}
override def channel(peer: AnyOutputSocket) = throw new AssertionError
}
cache.configure(new Input).input
}
def output(options: AccessOptions, template: Option[Entry]) = {
/** This class requires LAZY INITIALIZATION of its channel, but NO
* automatic decoupling on exceptions!
*/
final class Output extends DelegatingOutputSocket[Entry] {
private[this] val _options = options clear CACHE // consume
override lazy val socket = cache
.configure(CacheController.super.output(
_options clear EXCLUSIVE, name, template))
.output
override def stream(peer: AnyInputSocket) = {
assert(writeLockedByCurrentThread)
preOutput()
final class Stream extends DecoratingOutputStream(socket stream peer) {
register()
override def close() {
assert(writeLockedByCurrentThread)
out close ()
postOutput()
}
}
new Stream
}
override def channel(peer: AnyInputSocket) = {
assert(writeLockedByCurrentThread)
preOutput()
final class Channel extends DecoratingSeekableChannel(socket channel peer) {
register()
override def close() {
assert(writeLockedByCurrentThread)
channel close ()
postOutput()
}
}
new Channel
}
def preOutput() { make(_options, template) }
def postOutput() {
make(_options clear EXCLUSIVE, template orElse Option(cache))
register()
}
def make(options: AccessOptions, template: Option[Entry]) {
var mknodOpts = options
while (true) {
try {
CacheController.super.make(mknodOpts, name, FILE, template)
return
} catch {
case mknodEx: NeedsSyncException =>
// In this context, this exception means that the entry
// has already been written to the output archive for
// the target archive file.
// Pass on the exception if there is no means to
// resolve the issue locally, that is if we were asked
// to create the entry exclusively or this is a
// non-recursive file system operation.
if (mknodOpts get EXCLUSIVE) throw mknodEx
val syncOpts = SyncController modify SYNC
if (SYNC eq syncOpts) throw mknodEx
// Try to resolve the issue locally.
// Even if we were asked to create the entry
// EXCLUSIVEly, first we must try to get the cache in
// sync() with the virtual file system again and retry
// the make().
try {
CacheController.super.sync(syncOpts)
// sync() succeeded, now repeat the make()
} catch {
case syncEx: FsSyncException =>
syncEx addSuppressed mknodEx
// sync() failed, maybe just because the current
// thread has already acquired some open I/O
// resources for the same target archive file, e.g.
// an input stream for a copy operation and this
// is an artifact of an attempt to acquire the
// output stream for a child file system.
syncEx.getCause match {
case _: FsResourceOpenException =>
// Too bad, sync() failed because of a more
// serious issue than just some open resources.
// Let's rethrow the sync exception.
case _ => throw syncEx
}
// OK, we couldn't sync() because the current
// thread has acquired open I/O resources for the
// same target archive file.
// Normally, we would be expected to rethrow the
// make exception to trigger another sync(), but
// this would fail for the same reason und create
// an endless loop, so we can't do this.
//throw mknodEx;
// Dito for mapping the exception.
//throw FsNeedsLockRetryException.get(getModel());
// Check if we can retry the make with GROW set.
val oldMknodOpts = mknodOpts
mknodOpts = oldMknodOpts set GROW
if (mknodOpts eq oldMknodOpts) {
// Finally, the make failed because the entry
// has already been output to the target archive
// file - so what?!
// This should mark only a volatile issue because
// the next sync() will sort it out once all the
// I/O resources have been closed.
// Let's log the sync exception - mind that it has
// suppressed the make exception - and continue
// anyway...
logger debug ("ignoring", syncEx)
return
}
}
}
}
}
} // Output
new Output
}
} // EntryCache
}
private object CacheController {
private val logger = new LocalizedLogger(classOf[CacheController[_]]);
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy