/** Start collecting global monitoring information for the replication domain. */
private void initializePendingMonitorData() {
// Let's process our directly connected DS
// - in the ServerHandler for a given DS1, the stored state contains :
// -- the max CSN produced by DS1
// -- the last CSN consumed by DS1 from DS2..n
// - in the ReplicationDomainDB/ReplicaDB, the built-in state contains:
// -- the max CSN produced by each server
// So for a given DS connected we can take the state and the max from
// the DS/state.
for (ServerHandler ds : domain.getConnectedDSs().values()) {
final int serverId = ds.getServerId();
final ServerState dsState = ds.getServerState().duplicate();
CSN maxCSN = dsState.getCSN(serverId);
if (maxCSN == null) {
// This directly connected LS has never produced any change
maxCSN = new CSN(0, 0, serverId);
}
pendingMonitorData.setMaxCSN(maxCSN);
pendingMonitorData.setLDAPServerState(serverId, dsState);
pendingMonitorData.setFirstMissingDate(serverId, ds.getApproxFirstMissingDate());
}
// Then initialize the max CSN for the LS that produced something
// - from our own local db state
// - whatever they are directly or indirectly connected
final ServerState dbServerState = domain.getLatestServerState();
pendingMonitorData.setRSState(domain.getLocalRSServerId(), dbServerState);
for (CSN storedCSN : dbServerState) {
pendingMonitorData.setMaxCSN(storedCSN);
}
}
/**
* Processes a Monitor message receives from a remote Replication Server and stores the data
* received.
*
* @param msg The message to be processed.
* @param serverId server handler that is receiving the message.
*/
public void receiveMonitorDataResponse(MonitorMsg msg, int serverId) {
synchronized (pendingMonitorDataLock) {
if (pendingMonitorData == null) {
// This is a response for an earlier request whose computing is
// already complete.
logger.debug(INFO_IGNORING_REMOTE_MONITOR_DATA, domain.getBaseDN(), msg.getSenderID());
return;
}
try {
// Here is the RS state : list <serverID, lastCSN>
// For each LDAP Server, we keep the max CSN across the RSes
ServerState replServerState = msg.getReplServerDbState();
pendingMonitorData.setMaxCSNs(replServerState);
// store the remote RS states.
pendingMonitorData.setRSState(msg.getSenderID(), replServerState);
// Store the remote LDAP servers states
for (int dsServerId : toIterable(msg.ldapIterator())) {
ServerState dsServerState = msg.getLDAPServerState(dsServerId);
pendingMonitorData.setMaxCSNs(dsServerState);
pendingMonitorData.setLDAPServerState(dsServerId, dsServerState);
pendingMonitorData.setFirstMissingDate(
dsServerId, msg.getLDAPApproxFirstMissingDate(dsServerId));
}
// Process the latency reported by the remote RSi on its connections
// to the other RSes
for (int rsServerId : toIterable(msg.rsIterator())) {
long newFmd = msg.getRSApproxFirstMissingDate(rsServerId);
if (rsServerId == domain.getLocalRSServerId()) {
// this is the latency of the remote RSi regarding the current RS
// let's update the first missing date of my connected LS
for (DataServerHandler ds : domain.getConnectedDSs().values()) {
int connectedServerId = ds.getServerId();
pendingMonitorData.setFirstMissingDate(connectedServerId, newFmd);
}
} else {
// this is the latency of the remote RSi regarding another RSj
// let's update the latency of the LSes connected to RSj
ReplicationServerHandler rsjHdr = domain.getConnectedRSs().get(rsServerId);
if (rsjHdr != null) {
for (int remoteServerId : rsjHdr.getConnectedDirectoryServerIds()) {
pendingMonitorData.setFirstMissingDate(remoteServerId, newFmd);
}
}
}
}
} catch (RuntimeException e) {
// FIXME: do we really expect these???
logger.error(
ERR_PROCESSING_REMOTE_MONITOR_DATA,
e.getMessage() + " " + stackTraceToSingleLineString(e));
} finally {
// Decreases the number of expected responses and potentially
// wakes up the waiting requester thread.
if (pendingMonitorDataServerIDs.remove(serverId)) {
pendingMonitorDataLatch.countDown();
}
}
}
}
/**
* Recomputes the monitor data for this replication server domain.
*
* @return The recomputed monitor data for this replication server domain.
* @throws InterruptedException If this thread is interrupted while waiting for a response.
*/
public ReplicationDomainMonitorData recomputeMonitorData() throws InterruptedException {
// Only allow monitor recalculation at a time.
synchronized (pendingMonitorLock) {
if (monitorDataLastBuildDate + monitorDataLifeTime < TimeThread.getTime()) {
try {
DN baseDN = domain.getBaseDN();
// Prevent out of band monitor responses from updating our pending
// table until we are ready.
synchronized (pendingMonitorDataLock) {
// Clear the pending monitor data.
pendingMonitorDataServerIDs.clear();
pendingMonitorData = new ReplicationDomainMonitorData();
initializePendingMonitorData();
// Send the monitor requests to the connected replication servers.
for (ServerHandler rs : domain.getConnectedRSs().values()) {
final int serverId = rs.getServerId();
MonitorRequestMsg msg = new MonitorRequestMsg(domain.getLocalRSServerId(), serverId);
try {
rs.send(msg);
// Only register this server ID to pending table if we were able
// to send the message.
pendingMonitorDataServerIDs.add(serverId);
} catch (IOException e) {
// Log a message and do a best effort from here.
logger.error(
ERR_SENDING_REMOTE_MONITOR_DATA_REQUEST, baseDN, serverId, e.getMessage());
}
}
// Create the pending response latch based on the number of expected
// monitor responses.
pendingMonitorDataLatch = new CountDownLatch(pendingMonitorDataServerIDs.size());
}
// Wait for the responses to come back.
pendingMonitorDataLatch.await(5, TimeUnit.SECONDS);
// Log messages for replication servers that have gone or come back.
synchronized (pendingMonitorDataLock) {
// Log servers that have come back.
for (int serverId : monitorDataLateServers) {
// Ensure that we only log once per server: don't fill the
// error log with repeated messages.
if (!pendingMonitorDataServerIDs.contains(serverId)) {
logger.info(NOTE_MONITOR_DATA_RECEIVED, baseDN, serverId);
}
}
// Log servers that have gone away.
for (int serverId : pendingMonitorDataServerIDs) {
// Ensure that we only log once per server: don't fill the
// error log with repeated messages.
if (!monitorDataLateServers.contains(serverId)) {
logger.warn(WARN_MISSING_REMOTE_MONITOR_DATA, baseDN, serverId);
}
}
// Remember which servers were late this time.
monitorDataLateServers.clear();
monitorDataLateServers.addAll(pendingMonitorDataServerIDs);
}
// Store the new computed data as the reference
synchronized (pendingMonitorDataLock) {
// Now we have the expected answers or an error occurred
pendingMonitorData.completeComputing();
monitorData = pendingMonitorData;
monitorDataLastBuildDate = TimeThread.getTime();
}
} finally {
synchronized (pendingMonitorDataLock) {
// Clear pending state.
pendingMonitorData = null;
pendingMonitorDataLatch = null;
pendingMonitorDataServerIDs.clear();
}
}
}
}
return monitorData;
}
