* Reachability:

- Track unreachable peers persistently
        (i.e. separately from shitlist, and not cleared when they contact us)
      - Exclude detected unreachable peers from inbound tunnels
      - Exclude detected unreachable peers from selected leases
      - Exclude detected unreachable floodfill peers from lookups
      - Show unreachable status on profiles.jsp
This commit is contained in:
zzz
2008-04-17 18:59:15 +00:00
parent 2edd84e088
commit cf50b7eac1
16 changed files with 119 additions and 14 deletions

View File

@ -1,3 +1,12 @@
2008-04-17 zzz
* Reachability:
- Track unreachable peers persistently
(i.e. separately from shitlist, and not cleared when they contact us)
- Exclude detected unreachable peers from inbound tunnels
- Exclude detected unreachable peers from selected leases
- Exclude detected unreachable floodfill peers from lookups
- Show unreachable status on profiles.jsp
2008-04-16 zzz
* SSU/Reachability:
- Extend shitlist time from 4-8m to 40-60m

View File

@ -54,6 +54,7 @@ public abstract class CommSystemFacade implements Service {
public short getReachabilityStatus() { return STATUS_OK; }
public void recheckReachability() {}
public boolean isBacklogged(Hash dest) { return false; }
public boolean wasUnreachable(Hash dest) { return false; }
/**
* Tell other transports our address changed

View File

@ -17,7 +17,7 @@ import net.i2p.CoreVersion;
public class RouterVersion {
public final static String ID = "$Revision: 1.548 $ $Date: 2008-02-10 15:00:00 $";
public final static String VERSION = "0.6.1.32";
public final static long BUILD = 18;
public final static long BUILD = 19;
public static void main(String args[]) {
System.out.println("I2P Router version: " + VERSION + "-" + BUILD);
System.out.println("Router ID: " + RouterVersion.ID);

View File

@ -319,15 +319,29 @@ public class OutboundClientMessageOneShotJob extends JobImpl {
_lease = (Lease)orderedLeases.get(orderedLeases.firstKey());
} else {
****/
// Avoid a lease on a gateway we think is unreachable, if possible
for (int i = 0; i < _leaseSet.getLeaseCount(); i++) {
Lease l = _leaseSet.getLease(i);
if (!getContext().commSystem().wasUnreachable(l.getGateway())) {
_lease = l;
break;
}
if (_log.shouldLog(Log.WARN))
_log.warn(getJobId() + ": Skipping unreachable (by us) gateway " + l.getGateway());
}
if (_lease == null) {
_lease = (Lease)leases.get(0);
// }
if (_log.shouldLog(Log.WARN))
_log.warn(getJobId() + ": All leases are unreachable (by us) for " + _toString);
}
/*** removed until we fix SSU reachability
synchronized (_leaseCache) {
_leaseCache.put(_to, _lease);
}
***/
if (_log.shouldLog(Log.WARN))
_log.warn("Added to cache - lease for " + _toString);
***/
return true;
}

View File

@ -76,7 +76,7 @@ class FloodOnlySearchJob extends FloodSearchJob {
// We need to randomize our ff selection, else we stay with the same ones since
// getFloodfillPeers() is sorted by closest distance. Always using the same
// ones didn't help reliability.
// Also, query the unheard-from, unprofiled, failing and shitlisted ones last.
// Also, query the unheard-from, unprofiled, failing, unreachable and shitlisted ones last.
// We should hear from floodfills pretty frequently so set a 30m time limit.
// If unprofiled we haven't talked to them in a long time.
// We aren't contacting the peer directly, so shitlist doesn't strictly matter,
@ -90,14 +90,15 @@ class FloodOnlySearchJob extends FloodSearchJob {
Hash peer = (Hash)floodfillPeers.get(i);
PeerProfile profile = getContext().profileOrganizer().getProfile(peer);
if (profile == null || profile.getLastHeardFrom() < before ||
profile.getIsFailing() || getContext().shitlist().isShitlisted(peer)) {
profile.getIsFailing() || getContext().shitlist().isShitlisted(peer) ||
getContext().commSystem().wasUnreachable(peer)) {
failcount++;
ffp.add(peer);
} else
ffp.add(0, peer);
}
if (_log.shouldLog(Log.INFO) && failcount > 0)
_log.info(getJobId() + ": " + failcount + " of " + floodfillPeers.size() + " floodfills are not heard from, unprofiled, failing or shitlisted");
_log.info(getJobId() + ": " + failcount + " of " + floodfillPeers.size() + " floodfills are not heard from, unprofiled, failing, unreachable or shitlisted");
floodfillPeers = ffp;
}

View File

@ -424,6 +424,24 @@ public class ProfileOrganizer {
return;
}
public List selectPeersLocallyUnreachable() {
List n;
int count;
synchronized (_reorganizeLock) {
count = _notFailingPeers.size();
n = new ArrayList(_notFailingPeers.keySet());
}
List l = new ArrayList(count / 4);
for (Iterator iter = n.iterator(); iter.hasNext(); ) {
Hash peer = (Hash)iter.next();
if (_context.commSystem().wasUnreachable(peer))
l.add(peer);
}
if (_log.shouldLog(Log.INFO))
_log.info("Unreachable: " + l);
return l;
}
/**
* Find the hashes for all peers we are actively profiling
*

View File

@ -137,6 +137,7 @@ class ProfileOrganizerRenderer {
buf.append("<td>");
if (_context.shitlist().isShitlisted(peer)) buf.append("Shitlist");
if (prof.getIsFailing()) buf.append(" Failing");
if (_context.commSystem().wasUnreachable(peer)) buf.append(" Unreachable");
buf.append("&nbsp</td>");
//buf.append("<td><a href=\"/profile/").append(prof.getPeer().toBase64().substring(0, 32)).append("\">profile.txt</a> ");
//buf.append(" <a href=\"#").append(prof.getPeer().toBase64().substring(0, 32)).append("\">netDb</a></td>");

View File

@ -125,6 +125,10 @@ public class CommSystemFacadeImpl extends CommSystemFacade {
return _manager.isBacklogged(dest);
}
public boolean wasUnreachable(Hash dest) {
return _manager.wasUnreachable(dest);
}
public List getMostRecentErrorMessages() {
return _manager.getMostRecentErrorMessages();
}

View File

@ -48,6 +48,7 @@ public interface Transport {
public short getReachabilityStatus();
public void recheckReachability();
public boolean isBacklogged(Hash dest);
public boolean wasUnreachable(Hash dest);
public boolean isUnreachable(Hash peer);
}

View File

@ -15,6 +15,7 @@ import java.util.*;
import net.i2p.data.Hash;
import net.i2p.data.RouterAddress;
import net.i2p.data.RouterIdentity;
import net.i2p.data.RouterInfo;
import net.i2p.data.i2np.I2NPMessage;
import net.i2p.router.CommSystemFacade;
import net.i2p.router.Job;
@ -36,6 +37,7 @@ public abstract class TransportImpl implements Transport {
protected RouterContext _context;
/** map from routerIdentHash to timestamp (Long) that the peer was last unreachable */
private Map _unreachableEntries;
private Set _wasUnreachableEntries;
/**
* Initialize the new transport
@ -54,6 +56,7 @@ public abstract class TransportImpl implements Transport {
_context.statManager().createRateStat("transport.expiredOnQueueLifetime", "How long a message that expires on our outbound queue is processed", "Transport", new long[] { 60*1000l, 10*60*1000l, 60*60*1000l, 24*60*60*1000l } );
_sendPool = new ArrayList(16);
_unreachableEntries = new HashMap(16);
_wasUnreachableEntries = new HashSet(16);
_currentAddress = null;
}
@ -399,19 +402,23 @@ public abstract class TransportImpl implements Transport {
}
}
/** called when we can't reach a peer */
/** This isn't very useful since it is cleared when they contact us */
public void markUnreachable(Hash peer) {
long now = _context.clock().now();
synchronized (_unreachableEntries) {
_unreachableEntries.put(peer, new Long(now));
}
markWasUnreachable(peer, true);
}
/** called when we establish a peer connection (outbound or inbound) */
public void markReachable(Hash peer) {
public void markReachable(Hash peer, boolean isInbound) {
// if *some* transport can reach them, then we shouldn't shitlist 'em
_context.shitlist().unshitlistRouter(peer);
synchronized (_unreachableEntries) {
_unreachableEntries.remove(peer);
}
if (!isInbound)
markWasUnreachable(peer, false);
}
private class CleanupUnreachable extends JobImpl {
public CleanupUnreachable(RouterContext ctx) {
@ -432,6 +439,34 @@ public abstract class TransportImpl implements Transport {
}
}
/**
* Was the peer UNreachable (outbound only) the last time we tried it?
* This is NOT reset if the peer contacts us and it is never expired.
*/
public boolean wasUnreachable(Hash peer) {
synchronized (_wasUnreachableEntries) {
if (_wasUnreachableEntries.contains(peer))
return true;
}
RouterInfo ri = _context.netDb().lookupRouterInfoLocally(peer);
if (ri == null)
return false;
return null == ri.getTargetAddress(this.getStyle());
}
/**
* Maintain the WasUnreachable list
*/
public void markWasUnreachable(Hash peer, boolean yes) {
synchronized (_wasUnreachableEntries) {
if (yes)
_wasUnreachableEntries.add(peer);
else
_wasUnreachableEntries.remove(peer);
}
if (_log.shouldLog(Log.WARN))
_log.warn(this.getStyle() + " setting wasUnreachable to " + yes + " for " + peer);
}
public static boolean isPubliclyRoutable(byte addr[]) {
if (addr.length == 4) {
if ((addr[0]&0xFF) == 127) return false;

View File

@ -202,6 +202,20 @@ public class TransportManager implements TransportEventListener {
return false;
}
/**
* Was the peer UNreachable (outbound only) on any transport,
* based on the last time we tried it for each transport?
* This is NOT reset if the peer contacts us.
*/
public boolean wasUnreachable(Hash dest) {
for (int i = 0; i < _transports.size(); i++) {
Transport t = (Transport)_transports.get(i);
if (!t.wasUnreachable(dest))
return false;
}
return true;
}
Map getAddresses() {
Map rv = new HashMap(_transports.size());
for (int i = 0; i < _transports.size(); i++) {

View File

@ -351,7 +351,7 @@ public class EstablishState {
long diff = 1000*Math.abs(_tsA-_tsB);
if (diff >= Router.CLOCK_FUDGE_FACTOR) {
_context.statManager().addRateData("ntcp.invalidOutboundSkew", diff, 0);
_transport.markReachable(_con.getRemotePeer().calculateHash());
_transport.markReachable(_con.getRemotePeer().calculateHash(), false);
_context.shitlist().shitlistRouter(_con.getRemotePeer().calculateHash(), "Outbound clock skew of " + diff + " ms");
fail("Clocks too skewed (" + diff + " ms)", null, true);
return;
@ -536,7 +536,7 @@ public class EstablishState {
long diff = 1000*Math.abs(tsA-_tsB);
if (diff >= Router.CLOCK_FUDGE_FACTOR) {
_context.statManager().addRateData("ntcp.invalidInboundSkew", diff, 0);
_transport.markReachable(alice.calculateHash());
_transport.markReachable(alice.calculateHash(), true);
_context.shitlist().shitlistRouter(alice.calculateHash(), "Clock skew of " + diff + " ms");
fail("Clocks too skewed (" + diff + " ms)", null, true);
return;

View File

@ -419,7 +419,7 @@ public class NTCPConnection implements FIFOBandwidthLimiter.CompleteListener {
_established = true;
_establishedOn = System.currentTimeMillis();
_establishState = null;
_transport.markReachable(getRemotePeer().calculateHash());
_transport.markReachable(getRemotePeer().calculateHash(), false);
//_context.shitlist().unshitlistRouter(getRemotePeer().calculateHash(), NTCPTransport.STYLE);
boolean msgs = false;
synchronized (_outbound) {

View File

@ -132,7 +132,7 @@ public class NTCPTransport extends TransportImpl {
void inboundEstablished(NTCPConnection con) {
_context.statManager().addRateData("ntcp.inboundEstablished", 1, 0);
markReachable(con.getRemotePeer().calculateHash());
markReachable(con.getRemotePeer().calculateHash(), true);
//_context.shitlist().unshitlistRouter(con.getRemotePeer().calculateHash());
NTCPConnection old = null;
synchronized (_conLock) {

View File

@ -555,7 +555,7 @@ public class UDPTransport extends TransportImpl implements TimedWeightedPriority
_log.warn("Peer already connected: old=" + oldPeer + " new=" + peer, new Exception("dup"));
_activeThrottle.unchoke(peer.getRemotePeer());
markReachable(peer.getRemotePeer());
markReachable(peer.getRemotePeer(), peer.isInbound());
//_context.shitlist().unshitlistRouter(peer.getRemotePeer(), STYLE);
if (SHOULD_FLOOD_PEERS)

View File

@ -155,8 +155,12 @@ public abstract class TunnelPeerSelector {
//
// Unreachable peers at the inbound gateway is a major cause of problems.
// Due to a bug in SSU peer testing in 0.6.1.32 and earlier, peers don't know
// if they are unreachable, so this won't help much. As of 0.6.1.33 we should have
// lots of unreachables, so enable this for now.
// if they are unreachable, so the netdb indication won't help much.
// As of 0.6.1.33 we should have lots of unreachables, so enable this for now.
// Also (and more effectively) exclude peers we detect are unreachable,
// this should be much more effective, especially on a router that has been
// up a few hours.
//
// We could just try and exclude them as the inbound gateway but that's harder
// (and even worse for anonymity?).
//
@ -168,6 +172,9 @@ public abstract class TunnelPeerSelector {
List caps = ctx.peerManager().getPeersByCapability(Router.CAPABILITY_UNREACHABLE);
if (caps != null)
peers.addAll(caps);
caps = ctx.profileOrganizer().selectPeersLocallyUnreachable();
if (caps != null)
peers.addAll(caps);
}
if (filterSlow(ctx, isInbound, isExploratory)) {
Log log = ctx.logManager().getLog(TunnelPeerSelector.class);