public void LogFinishedSplit(string workerId, long totalSplits, long remainingSplits)
{
lock (LockLogFinished)
{
long splitId = remainingSplits;
SplitInfo splitInfo = splitInfos[splitId];
if (splitInfo.DidFinished())
{
//Logger.LogInfo("[" + workerId + " SLOW FINISHED " + splitId + "]");
return;
}
splitInfos[splitId].EndedSplit();
averageSplitTime = splitInfos[splitId].SplitTime() + 500;
numSplits--;
Logger.LogInfo("[" + workerId + " ENDED " + splitId + "] in " + splitInfos[splitId].SplitTime() + " ms. " + numSplits + " splits remaining");
}
}
public SplitInfo FindSlowSplit()
{
foreach (KeyValuePair <string, long> keyValue in workersSplits)
{
string key = keyValue.Key;
SplitInfo splitInfo = splitInfos[keyValue.Value];
if (averageSplitTime < long.MaxValue)
{
long waitTime = splitInfo.splitId == 0 ? 3 * averageSplitTime : 2 * averageSplitTime;
if (!splitInfo.DidFinished() && splitInfo.SplitTime() > waitTime)
{
Logger.LogWarn("[SPLIT " + splitInfo.splitId + " SLOW] taking " + splitInfo.SplitTime() + " vs " + waitTime);
return(splitInfo);
}
}
}
return(null);
}
private void StartPrimaryJobTrackerProcess(string clientUrl, long fileSize, long numSplits, string mapperName, byte[] mapperCode)
{
currentJobTrackerUrl = this.myURL;
serverRole = ServerRole.JOB_TRACKER;
status = ExecutionState.WORKING;
isPrimary = true;
jtInformation = new JobTrackerInformation(this, numSplits);
this.clientURL = clientUrl;
this.fileSize = fileSize;
client = (IClient)Activator.GetObject(typeof(IClient), clientURL);
this.mapperName = mapperName;
this.mapperCode = mapperCode;
Thread trackWorkersThread = new Thread(() =>
{
while (!jtInformation.DidFinishJob() && isPrimary && serverRole == ServerRole.JOB_TRACKER)
{
/* wait until if I am unfrozen and revert to Worker if needed */
if (WaitForUnfreezeAndCheckChanges())
{
break;
}
/* --------------------------- */
Logger.LogInfo("[CHECKING SLOW WORKERS]");
SplitInfo slowSplit = jtInformation.FindSlowSplit();
if (slowSplit != null)
{
//Logger.LogWarn("There is a slow split - " + slowSplit.splitId);
IWorker freeWorker = jtInformation.GetFirstFreeWorker();
if (freeWorker != null)
{
Logger.LogWarn("[SLOWWWWWWW SPLIT] RESENDING " + slowSplit.remainingSplits);
ResendSplitToNextWorker(freeWorker, slowSplit.fileSize, slowSplit.totalSplits, slowSplit.remainingSplits);
}
}
Thread.Sleep(4000);
}
serverRole = ServerRole.NONE;
status = ExecutionState.WAITING;
});
trackWorkersThread.Start();
Thread ConfigureSecondaryServerThread = new Thread(() =>
{
//wait for an available backUrl, then pings it then sets up as primary Server
while (isPrimary && serverRole == ServerRole.JOB_TRACKER && !jtInformation.DidFinishJob())
{
if (backURL == myURL)
{
continue;
}
//Logger.LogInfo("Waiting for an available url to be the secondary JT");
secondaryJT = (IWorker)Activator.GetObject(typeof(IWorker), backURL);
try
{
secondaryJT.PingJT();
backupSecondaryServerIfFailsUrl = secondaryJT.SetUpAsSecondaryServer(this.clientURL, this.myURL, fileSize, numSplits, mapperName, mapperCode);
backupSecondaryServerIfFails = (IWorker)Activator.GetObject(typeof(IWorker), backupSecondaryServerIfFailsUrl);
Logger.LogInfo("Success setting setup backupserver");
break;
}
catch (Exception)
{
Logger.LogInfo("There is still no backUrl available to become backup server");
}
Thread.Sleep(1000);
}
});
ConfigureSecondaryServerThread.Start();
if (!primaryDidStartedJob && !firstRequest)
{
primaryDidStartedJob = true;
ReceiveWork(clientUrl, fileSize, numSplits, mapperName, mapperCode);
}
firstRequest = false;
}