DelegatedRecoveryManagerpublic class DelegatedRecoveryManager extends Object This class manages information required for Delegated recovery. This class supports
multiple delegated recoveries at the same time. Functionality is alsomost same as
RecoveryManager.java. This class maintains the map between state and log location
instead of static data incase of Recovery Manager. |
Fields Summary |
---|
private static Hashtable | recoveryStatetable | private static Hashtable | tmoutMgrtable | static Logger | _logger |
Methods Summary |
---|
static boolean | addCoordinator(GlobalTID globalTID, java.lang.Long localTID, CoordinatorImpl coord, int timeout, java.lang.String logPath)
boolean result = true;
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
// Attempt to add the global and local indentifier to
// Coordinator associations to the maps.
state.coordsByGlobalTID.put(globalTID,coord);
state.coordsByLocalTID.put(localTID,coord);
// Set up the timeout for the transaction. When active, the
// timeout thread will periodically examine the map and abort
// any active transactions on it that have gone beyond their
// allocated time.
if (timeout != 0) {
DelegatedTimeoutManager tmoutMgr = getTimeoutManager(logPath);
tmoutMgr.setTimeout(localTID, DelegatedTimeoutManager.ACTIVE_TIMEOUT,
timeout);
}
return result;
| public static boolean | delegated_recover(java.lang.String logPath, javax.transaction.xa.XAResource[] resources)
try {
File recoveryFile = new File(logPath,LogControl.RECOVERY_STRING_FILE_NAME);
RandomAccessFile raf = new RandomAccessFile(recoveryFile,"r");
long length = raf.length();
byte b1[] = new byte[(int)length]; // length is very small
raf.readFully(b1);
String serverName = new String(b1);
raf.close();
return delegated_recover(serverName,logPath,resources);
} catch (IOException ex) {
_logger.log(Level.WARNING,"jts.exception_in_recovery_file_handling",ex);
throw ex;
// return false;
}
| public static boolean | delegated_recover(java.lang.String serverName, java.lang.String logPath, javax.transaction.xa.XAResource[] resources)
if (logPath == null || serverName == null) {
return false;
}
Configuration.setServerName(logPath,serverName);
boolean result = false;
boolean keypointRequired = false;
RecoveryStateHolder state = new RecoveryStateHolder();
recoveryStatetable.put(logPath,state);
Enumeration logRecords = CoordinatorLog.getLogged(logPath);
while (logRecords.hasMoreElements()) {
keypointRequired = true;
try {
(new TopCoordinator()).delegated_reconstruct((CoordinatorLog) logRecords.nextElement(), logPath);
} catch(Exception exc) {
_logger.log(Level.SEVERE,"jts.recovery_in_doubt_exception",exc);
_logger.log(Level.SEVERE,"jts.recovery_in_doubt",exc.toString());
String msg = LogFormatter.getLocalizedMessage(_logger, "jts.recovery_in_doubt",
new java.lang.Object[] {exc.toString()});
throw new org.omg.CORBA.INTERNAL(msg);
}
}
int size = resources.length;
Vector v = new Vector();
for (int i=0; i<size; i++) {
v.addElement(resources[i]);
}
state.uniqueRMSet = getUniqueRMSet(v.elements());
proceedWithXARecovery(logPath);
state.recoveryInProgress.post();
// If resync is not needed, then perform after-resync
// tasks immediately.
result = state.coordsByGlobalTID.size() > 0;
if (!result) {
try {
resyncComplete(false,keypointRequired,logPath);
} catch(Throwable exc) {}
}
if (result)
resync(logPath);
return true;
| static CoordinatorImpl | getCoordinator(GlobalTID globalTID, java.lang.String logPath)
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
CoordinatorImpl result = (CoordinatorImpl)
state.coordsByGlobalTID.get(globalTID);
return result;
| static CoordinatorImpl[] | getCoordinators(java.lang.String logPath)Returns an array of Coordinator objects currently active.
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
int size = state.coordsByGlobalTID.size();
CoordinatorImpl[] result = new CoordinatorImpl[size];
Enumeration coords = state.coordsByGlobalTID.elements();
for(int pos = 0;pos<size;){
result[pos++] = (CoordinatorImpl) coords.nextElement();
}
return result;
| static java.util.Hashtable | getCoordsByGlobalTID(java.lang.String logPath)
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
return state.coordsByGlobalTID;
| static CoordinatorImpl | getLocalCoordinator(java.lang.Long localTID, java.lang.String logPath)Returns a reference to the Coordinator object that corresponds to the
local identifier passed as a parameter.
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
CoordinatorImpl result = (CoordinatorImpl)
state.coordsByLocalTID.get(localTID);
return result;
| static synchronized DelegatedTimeoutManager | getTimeoutManager(java.lang.String logPath)
DelegatedTimeoutManager tmoutMgr = (DelegatedTimeoutManager)tmoutMgrtable.get(logPath);
if (tmoutMgr != null)
return tmoutMgr;
tmoutMgr = new DelegatedTimeoutManager(logPath);
tmoutMgrtable.put(logPath,tmoutMgr);
return tmoutMgr;
| private static java.util.Enumeration | getUniqueRMSet(java.util.Enumeration xaResourceList)Reduce the set of XAResource objects into a unique set such that there
is at most one XAResource object per RM.
Vector uniqueRMList = new Vector();
while (xaResourceList.hasMoreElements()) {
XAResource xaRes = (XAResource) xaResourceList.nextElement();
int size = uniqueRMList.size();
boolean match = false;
for (int i = 0; i < size; i++) { // compare and eliminate duplicates
XAResource uniqueXaRes = (XAResource) uniqueRMList.elementAt(i);
try {
if (xaRes.isSameRM(uniqueXaRes)) {
match = true;
break;
}
} catch (XAException xe) {}
}
if (!match) {
uniqueRMList.add(xaRes);
}
}
return uniqueRMList.elements();
| private static void | proceedWithXARecovery(java.lang.String logPath)This method is used to recontruct and register the Resource objects
corresponding to in-doubt transactions in the RMs. It is assumed
that the XAResource list has already been provided to the
Recovery Manager. This method can be called by Recovery Thread as
well as any other thread driving recovery of XA Resources.
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
/* This method has been newly added - Ram Jeyaraman */
Enumeration xaResources = state.uniqueRMSet;
// sanity check
if (xaResources == null) {
return;
}
Vector otsResources = new Vector();
// Map uniqueXids = new Hashtable();
Set uniqueXids = new HashSet();
while (xaResources.hasMoreElements()) {
XAResource xaResource = (XAResource) xaResources.nextElement();
// Get the list of XIDs which represent in-doubt transactions
// for the database.
Xid[] inDoubtXids = RecoveryManager.getInDoubtXids(xaResource);
// uniqueXids.clear();
if (inDoubtXids == null || inDoubtXids.length == 0) {
break; // all in-doubt xids have been obtained.
}
for (int i = 0; i < inDoubtXids.length; i++) {
// check to see if the xid belongs to this server.
String branchQualifier =
new String(inDoubtXids[i].getBranchQualifier());
String serverName = Configuration.getServerName(logPath);
if (branchQualifier.startsWith(serverName)) {
// check if the xid is a duplicate. i.e., Xids
// which have same globalId and branchId are
// considered duplicates. Note that the
// branchId format is (serverId, rmId). This is
// to make sure that at most one OTSResource object
// is registered with the coordinator per transaction
// per RM.
String xidStr = stringifyXid(inDoubtXids[i]);
if (!uniqueXids.contains(inDoubtXids[i])) { // unique xid
if(_logger.isLoggable(Level.FINE))
{
_logger.logp(Level.FINE,"DelegatedRecoveryManager",
"proceedWithXARecovery",
" This xid is UNIQUE " +
inDoubtXids[i]);
}
uniqueXids.add(inDoubtXids[i]); // add to uniqueList
// Create an OTSResource for the in-doubt
// transaction and add it to the list. Each
// OTSResource represents a RM per transaction.
otsResources.addElement(
new OTSResourceImpl(inDoubtXids[i],
xaResource, null
).getCORBAObjReference());
} else {
if(_logger.isLoggable(Level.FINE))
{
_logger.logp(Level.FINE,"DelegatedRecoveryManager",
"proceedWithXARecovery",
" This xid is NOTUNIQUE " +
inDoubtXids[i]);
}
}
} else {
if(_logger.isLoggable(Level.FINE))
{
_logger.logp(Level.FINE,"DelegatedRecoveryManager",
"proceedWithXARecovery",
" This xid doesn't belong to me " +
inDoubtXids[i]);
}
}
}
}
// For each OTSResource, determine whether the transaction is known,
// and if so, register it, otherwise roll it back.
for (int i = 0; i < otsResources.size(); i++) {
OTSResource otsResource = (OTSResource) otsResources.elementAt(i);
GlobalTID globalTID = new GlobalTID(otsResource.getGlobalTID());
TopCoordinator coord =
(TopCoordinator) state.coordsByGlobalTID.get(globalTID);
if (coord == null) {
// Roll the OTSResource back if the transaction is not
// recognised. This happens when the RM has recorded its
// prepare vote, but the JTS has not recorded its prepare vote.
if(_logger.isLoggable(Level.FINE)) {
_logger.logp(Level.FINE,"DelegatedRecoveryManager","proceedWithXARecovery()",
"Could not recognize OTSResource: "+otsResource +
" with tid: " +
LogFormatter.convertToString(globalTID.realTID.tid)+
";Hence rolling this resource back...");
}
boolean infiniteRetry = true;
int commitRetries = Configuration.getRetries();
if (commitRetries >= 0)
infiniteRetry = false;
int commitRetriesLeft = commitRetries;
boolean exceptionisThrown = true;
while (exceptionisThrown) {
try {
otsResource.rollback();
exceptionisThrown = false;
} catch (Throwable exc) {
if ((exc instanceof COMM_FAILURE) || (exc instanceof TRANSIENT)) {
if (commitRetriesLeft > 0 || infiniteRetry) {
// For TRANSIENT or COMM_FAILURE, wait
// for a while, then retry the commit.
if (!infiniteRetry) {
commitRetriesLeft--;
}
try {
Thread.sleep(Configuration.COMMIT_RETRY_WAIT);
} catch( Throwable e ) {}
}
else {
_logger.log(Level.WARNING,"jts.exception_during_resync",
new java.lang.Object[] {exc.toString(),"OTSResource rollback"});
exceptionisThrown = false;
}
}
else {
_logger.log(Level.WARNING,"jts.exception_during_resync",
new java.lang.Object[] {exc.toString(),"OTSResource rollback"});
exceptionisThrown = false;
}
}
}
} else {
// NOTE: Currently unimplemented. The coordinator needs to
// check if duplicate resources are being registered for the
// same RM for the same xid. Also the coordinator should
// not go away, until all its resources have been sent
// completion notification. The keypointing should not
// be done *as is* in the removeCoordinator() method.
// waitForResync semaphore needs to be flagged when the
// recovery thread goes away.
// Register the OTSResource with the Coordinator.
// It will be called for commit or rollback during resync.
if(_logger.isLoggable(Level.FINE)) {
_logger.logp(Level.FINE,"DelegatedRecoveryManager",
"proceedWithXARecovery()",
"Recognized OTSResource: " + otsResource +
" with tid: " +
LogFormatter.convertToString(globalTID.realTID.tid) +
";Hence registering this resource with coordinator...");
}
coord.directRegisterResource(otsResource);
}
}
| static boolean | removeCoordinator(GlobalTID globalTID, java.lang.Long localTID, boolean aborted, java.lang.String logPath)
boolean result = false;
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
// Remove the global identifier to Coordinator mapping if possible.
CoordinatorImpl coord = null;
result = (state.coordsByGlobalTID.remove(globalTID) != null);
// Remove the InternalTid to Coordinator mapping if possible.
if (result) {
coord = (CoordinatorImpl) state.coordsByLocalTID.remove(localTID);
result = (coord != null);
}
// If that succeeded, forget the CoordinatorLog object, if the
// transaction is not a subtransaction. The following may return
// FALSE if there are no log records available
// (i.e. non-recoverable OTS).
if (coord != null) {
try {
if (coord.is_top_level_transaction()) {
CoordinatorLog.removeLog(localTID, logPath);
}
} catch(SystemException exc) {
result = false;
}
}
// Clear the timeout for the transaction, if any.
// Perform the removal under the timer mutex.
DelegatedTimeoutManager tmoutMgr = getTimeoutManager(logPath);
tmoutMgr.setTimeout(localTID, DelegatedTimeoutManager.CANCEL_TIMEOUT, 0);
// Modify any thread associations there may be for the transaction, to
// indicate that the transaction has ended.
// COMMENT(Ram J) 09/19/2001 This below line is commented out since in
// the J2EE controlled environment, all threads are associated and
// dissociated in an orderly fashion, as well as there is no possibility
// of concurrent threads active in a given transaction.
//CurrentTransaction.endAll(globalTID, aborted);
// If the count of resyncing Coordinators is greater than zero,
// this means we are still in resync. Decrease the count.
if (state.resyncCoords > 0) {
state.resyncCoords--;
// If the number of resyncing Coordinators is now zero,
// we may allow new work.
if (state.resyncCoords == 0) {
try {
resyncComplete(true, true, logPath);
} catch (Throwable exc) {}
}
}
return result;
| static void | resync(java.lang.String logPath)
// If there are any transactions, proceed with resync. The map of
// coordinators by global identifier is created during the
// TopCoordinator reconstruct method when the coordinators are added
// via addCoordinator. We copy the contents to another map as
// Coordinators will remove themselves from the map during resync.
// Now that the Coordinators have been reconstructed, record
// the number of transactions requiring resync,
// and make an event trace point. We must clone the Hashtable
// here so that the Enumeration does not get
// changed when any subsequent transaction is created (this can happen
// when the last Coordinator is removed).
RecoveryStateHolder recoveryState = (RecoveryStateHolder)recoveryStatetable.get(logPath);
recoveryState.resyncCoords = recoveryState.coordsByGlobalTID.size();
Enumeration resyncList =
((Hashtable) recoveryState.coordsByGlobalTID.clone()).elements();
boolean isRoot[] = new boolean[1];
// Go through and resync each transaction. The transaction lock
// for each transaction is obtained to avoid deadlocks during recovery.
while (resyncList.hasMoreElements()) {
TopCoordinator coord = (TopCoordinator)resyncList.nextElement();
try {
// Before performing recovery, lock the coordinator.
synchronized (coord) {
Status state = coord.recover(isRoot);
if (state == Status.StatusUnknown) {
// If the coordinator can be locked, then perform
// recovery on it. If the outcome is not currently
// known, we do nothing with the transaction,
// as we expect to eventually get an outcome
// from the parent. In this case an in-doubt timeout
// is established for the
// transaction so that it will continue to retry.
// For subordinates, the Coordinator will compl-ete the
// transaction itself as it will have no
// Synchronization objects.
DelegatedTimeoutManager tmoutMgr = getTimeoutManager(logPath);
tmoutMgr.setTimeout(
new Long(coord.getLocalTID()),
DelegatedTimeoutManager.IN_DOUBT_TIMEOUT,
60);
} else if (state == Status.StatusCommitted) {
// For committed or rolled back, proceed with
// completion of the transaction, regardless of
// whether it is the root or a subordinate.
// If the transaction represents a root, it would
// normally wait for the CoordinatorTerm object to
// call before completing the transaction. As there is
// no CoordinatorTerm in recovery, we must do it here.
if(_logger.isLoggable(Level.FINE)) {
_logger.logp(Level.FINE,"DelegatedRecoveryManager","resync()",
"Before invoking commit on the reconstructed coordinator"+
"GTID is: "+
((TopCoordinator)coord).superInfo.globalTID.toString());
}
try {
coord.commit();
} catch (Throwable exc) {
_logger.log(Level.WARNING,"jts.exception_during_resync",
new java.lang.Object[] {exc.toString(),"commit"});
}
if (isRoot[0]) {
try {
coord.afterCompletion(state);
} catch (Throwable exc) {
_logger.log(Level.WARNING,"jts.exception_during_resync",
new java.lang.Object[] {exc.toString(),
"after_completion"});
}
}
} else {
// By default, roll the transaction back.
try {
if(_logger.isLoggable(Level.FINE)) {
_logger.logp(Level.FINE,"DelegatedRecoveryManager","resync()",
"Before invoking rollback on the"+
"reconstructed coordinator :"+
"GTID is : "+
((TopCoordinator)coord).superInfo.globalTID.toString());
}
coord.rollback(true);
} catch (Throwable exc) {
_logger.log(Level.WARNING,"jts.resync_failed",
new java.lang.Object [] {exc.toString(),"rollback"});
}
if (isRoot[0]) {
try {
coord.afterCompletion(Status.StatusRolledBack);
} catch (Throwable exc) {
_logger.log(Level.WARNING,"jts.resync_failed",
new java.lang.Object[]
{ exc.toString(), "after_completion"});
}
}
}
}
} catch (Throwable exc) {}
}
// Note that resyncComplete will be called by the
// last TopCoordinator to complete resync (in removeCoordinator)
// so we do not need to do it here.
| private static void | resyncComplete(boolean resynced, boolean keypointRequired, java.lang.String logPath)
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
// Inform JTSXA that resync is complete, and trace the fact
// that resync has completed.
// COMMENT(Ram J) not needed anymore
//JTSXA.resyncComplete();
// Perform a keypoint of the log if required.
if (keypointRequired) {
CoordinatorLog.keypoint(logPath);
}
// Post the resync in progress event semaphore.
state.resyncInProgress.post();
state.resyncInProgress = null;
| static void | shutdown(boolean immediate)Informs the DelegatedRecoveryManager that the transaction service is being shut
down.
For immediate shutdown,
For quiesce,
Enumeration keys = recoveryStatetable.keys();
if (keys.hasMoreElements()) {
String logPath = (String)keys.nextElement();
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
if (immediate) {
// If immediate, stop the resync thread if any.
} else {
// Otherwise ensure that resync has completed.
if (state.resyncInProgress != null) {
try {
state.resyncInProgress.waitEvent();
} catch (InterruptedException exc) {}
}
}
// COMMENT(Ram J) not needed anymore.
//JTSXA.shutdown(immediate);
// If not immediate shutdown, keypoint and close the log.
// Only do this if the process is recoverable!
if (!immediate) {
CoordinatorLog.keypoint(logPath);
CoordinatorLog.finalizeAll(logPath);
}
//$Continue with shutdown/quiesce.
}
| private static java.lang.String | stringifyXid(javax.transaction.xa.Xid xid)
int glen = xid.getGlobalTransactionId().length;
int blen = xid.getBranchQualifier().length;
byte[] xidRep = new byte[glen + 1 + blen];
System.arraycopy(xid.getGlobalTransactionId(), 0, xidRep, 0, glen);
xidRep[glen] = (byte) ',";
System.arraycopy(xid.getBranchQualifier(), 0, xidRep, glen + 1, blen);
return new String(xidRep);
| static boolean | validLocalTID(java.lang.Long localTID, java.lang.String logPath)Determines whether the local transaction identifier represents a valid
transaction.
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
boolean result = state.coordsByLocalTID.containsKey(localTID);
return result;
| public static void | waitForRecovery(java.lang.String logPath)
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
if (state.recoveryInProgress != null) {
try {
state.recoveryInProgress.waitEvent();
} catch (InterruptedException exc) {
_logger.log(Level.SEVERE,"jts.wait_for_resync_complete_interrupted");
String msg = LogFormatter.getLocalizedMessage(_logger,
"jts.wait_for_resync_complete_interrupted");
throw new org.omg.CORBA.INTERNAL(msg);
}
}
| public static void | waitForResync(java.lang.String logPath)Waits for resync to complete.
RecoveryStateHolder state = (RecoveryStateHolder)recoveryStatetable.get(logPath);
if (state.resyncInProgress != null) {
try {
state.resyncInProgress.waitEvent();
} catch (InterruptedException exc) {
_logger.log(Level.SEVERE,"jts.wait_for_resync_complete_interrupted");
String msg = LogFormatter.getLocalizedMessage(_logger,
"jts.wait_for_resync_complete_interrupted");
throw new org.omg.CORBA.INTERNAL(msg);
}
}
|
|