/**
* {@inheritDoc}
*
* <p>Carefull that stopApp() could in theory be called before startApp().
*/
public void stopApp() {
resolver.unregisterHandler(routerSName);
// unregister SRDI
resolver.unregisterSrdiHandler(routerSName);
srdiIndex.stop();
resolver = null;
srdi = null;
membership = null;
}
/**
* Make this transport as up and running.
*
* <p>When this method is called, all the services are already registered with the peergroup. So
* we do not need to delay binding any further. All the public methods, which could be called
* between init and startApp are defensive regarding the services possibly not being there.
*/
public int startApp(String[] arg) {
resolver = group.getResolverService();
membership = group.getMembershipService();
if (null == resolver) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Endpoint Router start stalled until resolver service available");
}
return Module.START_AGAIN_STALLED;
}
if (null == membership) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Endpoint Router start stalled until membership service available");
}
return Module.START_AGAIN_STALLED;
}
resolver.registerHandler(routerSName, this);
// create and register the srdi service
srdiIndex = new SrdiIndex(group, srdiIndexerFileName);
// Srdi is a thread but we are not going to start,
// since the service is reactive.
srdi = new Srdi(group, routerSName, this, srdiIndex, 0, 0);
resolver.registerSrdiHandler(routerSName, this);
// obtain credential for resolver messages
try {
// FIXME 20041008 bondolo this needs to be converted to dynamically managing credentials.
// get the only credential "nobody"
credential = (Credential) membership.getDefaultCredential();
if (null != credential) {
credentialDoc = credential.getDocument(MimeMediaType.XMLUTF8);
}
} catch (Exception e) {
if (LOG.isEnabledFor(Level.ERROR)) {
LOG.error("failed to get credential", e);
}
}
// get the RouteCM cache service
routeCM = router.getRouteCM();
return 0;
}
/**
* Process the Query, and generate response
*
* @param query the query to process
*/
public int processQuery(ResolverQueryMsg query) {
if (!useRouteResolver) { // Route resolver disabled
return ResolverService.OK;
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery starts");
}
RouteQuery routeQuery = null;
Reader ip = null;
try {
ip = new StringReader(query.getQuery());
StructuredTextDocument asDoc =
(StructuredTextDocument)
StructuredDocumentFactory.newStructuredDocument(MimeMediaType.XMLUTF8, ip);
routeQuery = new RouteQuery(asDoc);
} catch (Throwable e) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Malformed Route query ", e);
}
return ResolverService.OK;
} finally {
try {
if (null != ip) {
ip.close();
ip = null;
}
} catch (Throwable ignored) {;
}
}
PeerID pId = routeQuery.getDestPeerID();
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery: looking for route to " + pId);
}
RouteAdvertisement srcRoute = routeQuery.getSrcRoute();
List badHops = routeQuery.getBadHops();
if (LOG.isEnabledFor(Level.DEBUG)) {
if (badHops != null) {
LOG.debug("processQuery: bad Hops");
for (int i = 0; i < badHops.size(); i++) {
LOG.debug("processQuery: :" + ((PeerID) badHops.get(i)).toString());
}
}
}
// if our source route is not null, then publish it
if (srcRoute != null) {
if (!(srcRoute.getDestPeerID()).equals(localPeerId)) {
// This is not our own peer adv so we must not keep it
// longer than its expiration time.
try {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery: publishing sender route info " + srcRoute.getDestPeerID());
}
// we only need to publish this route if
// we don't know about it yet
// XXX: here is where we could be more conservative and use isNormallyReachable() instead,
// thus excluding
// incoming messengers.
if ((!router.isLocalRoute(router.pid2addr(srcRoute.getDestPeerID())))
&& (!router.isRoutedRoute(router.pid2addr(srcRoute.getDestPeerID())))) {
routeCM.publishRoute(srcRoute);
}
} catch (Exception e) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Could not publish Route Adv from query - discard", e);
}
return ResolverService.OK;
}
}
} else {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("No src Route in route query - discard ");
}
return ResolverService.OK;
}
if (pId == null) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Malformed route query request, no PeerId - discard");
}
return ResolverService.OK;
}
// We have more luck with that one because, since it is part of OUR
// message, and not part of the resolver protocol, it is in OUR
// format.
EndpointAddress qReqAddr = router.pid2addr(pId);
RouteAdvertisement route = null;
// check if this peer has a route to the destination
// requested
boolean found = false;
if (qReqAddr.equals(localPeerAddr)) {
found = true;
// return the route that is my local route
route = router.getMyLocalRoute();
} else {
// only rendezvous can respond to route requests
// if not we are generating too much traffic
// XXX: here is where we could be more conservative and use isNormallyReachable() instead,
// thus excluding
// incoming messengers.
if (router.isLocalRoute(qReqAddr)) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery: peer has direct route to destination ");
}
// we should set the route to something :-)
found = true;
// this peer has a direct route to the destination
// return the short route advertisement we know for this peer
// (For us it is zero hop, and we advertise ourself as the routing
// peer in the response. The stiching is done by whoever gets that
// response). May be there are more than one hop advertised in-there...
// alternate routing peers...should we leave them ?
// For now, we keep the full dest, but wack the hops.
route =
(RouteAdvertisement)
AdvertisementFactory.newAdvertisement(RouteAdvertisement.getAdvertisementType());
AccessPointAdvertisement ap =
(AccessPointAdvertisement)
AdvertisementFactory.newAdvertisement(
AccessPointAdvertisement.getAdvertisementType());
ap.setPeerID(pId);
route.setDest(ap);
} else {
route = router.getRoute(qReqAddr, false);
if (route != null) {
found = true;
// check if we were given some bad hops info
// and see if the found route contains
// any of these bad hops. In that case, we need
// to mark this route as bad
for (int i = 0; i < badHops.size(); i++) {
// destination is known to be bad
if (router.addr2pid(qReqAddr).equals(((PeerID) badHops.get(i)))) {
processBadRoute((PeerID) badHops.get(i), route);
found = false;
break;
}
if (route.containsHop((PeerID) badHops.get(i))) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug(
"processQuery: peer has bad route due to "
+ ((PeerID) badHops.get(i)).toString());
}
processBadRoute((PeerID) badHops.get(i), route);
found = false;
break;
}
}
}
}
}
if (!found) {
// discard the request if we are not a rendezvous
// else forward to the next peers
if (!group.isRendezvous()) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("discard query forwarding as not a rendezvous");
}
return ResolverService.OK;
}
// did not find a route, check our srdi cache
// make sure we protect against out of sync
// SRDI index
// srdi forwarding is only involved once the Index entry has
// been found and we forwarded the resolver query. Afterward a
// normal walk proceeds from the initial SRDI index pointing
// rdv. This is done to protect against potential loopback
// entries in the SRDI cache index due to out of sync peerview
// and index.
if (query.getHopCount() < 2) {
// check local SRDI cache to see if we have the entry
// we look for 10 entries, will pickup one randomly
Vector results = srdiIndex.query("route", "DstPID", pId.toString(), 10);
if (results.size() > 0) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery srdiIndex lookup match :" + results.size());
}
// remove any non-rdv peers to avoid sending
// to a non-rdv peers and garbage collect the SRDI
// index in the process
Vector clean = cleanupAnyEdges(query.getSrc(), results);
if (clean.size() > 0) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("found an srdi entry forwarding query to SRDI peer");
}
// The purpose of incrementing the hopcount
// when an SRDI index match is found (we got a
// pointer to a rdv that should have the route) is to
// restrict any further forwarding. The increment
// count is only done when a matching SRDI index is
// found. Not when the replica is selected as we
// still need to forward the query. This restriction
// is purposelly done to avoid too many longjumps
// within a walk.
query.incrementHopCount();
// Note: this forwards the query to 1 peer randomly
// selected from the result
srdi.forwardQuery(clean, query, 1);
// tell the resolver no further action is needed.
return ResolverService.OK;
}
}
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("did not find a route or SRDI index");
}
// force a walk
return ResolverService.Repropagate;
}
// we found a route send the response
try {
if (route == null) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("we should have had a route at this point");
}
return ResolverService.OK;
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery: we have a route build route response" + route.display());
}
RouteAdvertisement myRoute = router.getMyLocalRoute();
// make sure we initialized our local
// route info as we will need it to respond. We may
// not have our route if we are still
// waiting for a relay connection.
if (myRoute == null) {
return ResolverService.OK;
}
RouteResponse routeResponse = new RouteResponse();
routeResponse.setDestRoute(route);
routeResponse.setSrcRoute(myRoute);
if (routeResponse == null) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("error creating route response");
}
return ResolverService.OK;
}
// construct a response from the query
ResolverResponseMsg res = query.makeResponse();
res.setCredential(credentialDoc);
res.setResponse(routeResponse.toString());
resolver.sendResponse(query.getSrc(), res);
return ResolverService.OK;
} catch (Exception ee) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery: error while processing query ", ee);
}
return ResolverService.OK;
}
}
/**
* issue a new route discovery resolver request
*
* @param peer the destination as a logical enpoint address
*/
protected void findRoute(EndpointAddress peer) {
RouteAdvertisement myRoute = router.getMyLocalRoute();
// No need to pursue further if we haven't
// initialize our own route as responding
// peers are not going to be able to respond to us.
if (myRoute == null) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Cannot issue a find route if we don't know our own route");
}
return;
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Find route for peer = " + peer);
}
try {
// create a new RouteQuery message
RouteQuery doc = null;
// check if we have some bad route information
// for that peer, in that case pass the bad hop count
BadRoute badRoute;
badRoute = (BadRoute) router.getBadRoute(peer);
if (badRoute != null) {
// ok we have a bad route
// pass the bad hops info as part of the query
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("findRoute sends query: known bad Hops" + badRoute.display());
}
doc = new RouteQuery(router.addr2pid(peer), myRoute, badRoute.getHops());
} else {
doc = new RouteQuery(router.addr2pid(peer), myRoute, null);
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("Sending query for peer : " + peer);
}
ResolverQuery query =
new ResolverQuery(
routerSName, credentialDoc, localPeerId.toString(), doc.toString(), qid++);
// only run SRDI if we are a rendezvous
if (group.isRendezvous()) {
// check where to send the query via SRDI
Vector results = null;
if (srdiIndex != null) {
// try to find a least 10 entries, will pick up one
// randomly. This will protect against retry. It is
// likely that a number of RDV will know about a route
results = srdiIndex.query("route", "DstPID", router.addr2pid(peer).toString(), 10);
if (results != null && results.size() > 0) {
// use SRDI to send the query
// remove any non rdv peers from the candidate list
// and garbage collect the index in the process
Vector clean = cleanupAnyEdges(query.getSrc(), results);
if (clean.size() > 0) {
// The purpose of incrementing the hopcount
// when an SRDI index match is found (we got a
// pointer to a rdv that should have the route) is to
// restrict any further forwarding. The increment
// count is only done when a matching SRDI index is
// found. Not when the replica is selected as we
// still need to forward the query. This restriction
// is purposelly done to avoid too many longjumps
// within a walk.
query.incrementHopCount();
srdi.forwardQuery(clean, query, 1);
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("found an srdi entry forwarding query to SRDI peer");
}
return;
}
} else {
// it is not in our cache, look for the replica peer
// we need to send the query
PeerID destPeer = srdi.getReplicaPeer(router.addr2pid(peer).toString());
if (destPeer != null && !destPeer.equals(localPeerId)) {
// don't push anywhere if we do not have a replica
// or we are trying to push to ourself
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("processQuery srdiIndex DHT forward :" + destPeer);
}
srdi.forwardQuery(destPeer.toString(), query);
return;
}
}
}
}
// if we reach that point then we just use the resolver walk
resolver = group.getResolverService();
if (resolver != null) {
resolver.sendQuery(null, query);
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("find route query sent");
}
} else {
if (LOG.isEnabledFor(Level.WARN)) {
LOG.warn("cannot get the resolver service");
}
}
} catch (Exception ee) {
if (LOG.isEnabledFor(Level.WARN)) {
LOG.warn("Exception in findRoute", ee);
}
}
}
/**
* Return a route error in case a route was found to be invalid as the current hop cannot find a
* way to forward the message to the destination or any other hops in the forward part of the
* route. In that case a negative route response is forwarded to the orginal source of the
* message. Now of course we do not have any way to guarantee that the NACK message will be
* received by the sender, but the best we can do is to try :-)
*
* <p>we send a query ID to NACKROUTE_QUERYID to indicate this is a bad Route
*
* @param src original source of the message
* @param dest original destination of the message
*/
protected void generateNACKRoute(PeerID src, PeerID dest, Vector origHops) {
// As long as the group is partially initialized, do not bother
// trying to send NACKS. We can't: it just causes NPEs.
if (resolver == null) {
return;
}
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("generate NACK Route response " + src);
}
// check first, if we are not already in process of looking for a
// route to the destination peer of the NACK. We should not try to
// send a NACK to that destination at that point as this will block
// our incoming processing thread while it is looking for a route to
// that destination. If there a no pending route requests to that
// destination then we can go ahead an attempt to send the NACK. At
// the maximum we should have only one thread block while looking for
// a specific destination. When we find a route to the destination,
// the next NACK processing will be sent.
if (router.isPendingRouteQuery(router.pid2addr(src))) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("drop NACK due to pending route discovery " + src);
}
return;
}
// Generate a route response
RouteAdvertisement route =
(RouteAdvertisement)
AdvertisementFactory.newAdvertisement(RouteAdvertisement.getAdvertisementType());
AccessPointAdvertisement ap =
(AccessPointAdvertisement)
AdvertisementFactory.newAdvertisement(AccessPointAdvertisement.getAdvertisementType());
ap.setPeerID(dest);
route.setDest(ap);
route.setHops(origHops);
// set the the route of the peer that
// detected the bad route
RouteAdvertisement routeSrc =
(RouteAdvertisement)
AdvertisementFactory.newAdvertisement(RouteAdvertisement.getAdvertisementType());
AccessPointAdvertisement apSrc =
(AccessPointAdvertisement)
AdvertisementFactory.newAdvertisement(AccessPointAdvertisement.getAdvertisementType());
apSrc.setPeerID((PeerID) localPeerId);
routeSrc.setDest(apSrc);
RouteResponse routeResponse = new RouteResponse();
routeResponse.setDestRoute(route);
routeResponse.setSrcRoute(routeSrc);
if (routeResponse == null) {
if (LOG.isEnabledFor(Level.DEBUG)) {
LOG.debug("error creating route NACKresponse");
}
return;
}
ResolverResponse res =
new ResolverResponse(
routerSName,
credentialDoc,
NACKROUTE_QUERYID, // mean NACKRoute
routeResponse.toString());
// send the NACK response back to the originator
resolver.sendResponse(src.toString(), res);
}