/// <summary>
/// Initialize a job summary.
/// </summary>
/// <param name="cluster">Cluster where the job ran.</param>
/// <param name="clusterType">A string corresponding to the type of ClusterConfiguration.</param>
/// <param name="machine">Machine where job manager ran.</param>
/// <param name="jobId">Id of job.</param>
/// <param name="jmProcessGuid">Guid of job manager process.</param>
/// <param name="clusterJobId">Id of job on the cluster.</param>
/// <param name="friendlyname">Friendly name used.</param>
/// <param name="username">Who ran the job.</param>
/// <param name="date">Start date (not completion date).</param>
/// <param name="status">Job status.</param>
/// <param name="endTime">Estimated end running time.</param>
/// <param name="virtualcluster">Virtual cluster where job ran.</param>
public DryadLinqJobSummary(
string cluster,
ClusterConfiguration.ClusterType clusterType,
string virtualcluster,
string machine,
string jobId,
string clusterJobId,
DryadProcessIdentifier jmProcessGuid,
string friendlyname,
string username,
DateTime date,
DateTime endTime,
ClusterJobInformation.ClusterJobStatus status)
{
this.VirtualCluster = virtualcluster;
this.Cluster = cluster;
this.ClusterType = clusterType;
this.Machine = machine;
this.Name = friendlyname;
this.User = username;
this.Date = date;
this.EndTime = endTime;
this.Status = status;
this.ManagerProcessGuid = jmProcessGuid;
this.JobID = jobId;
this.ClusterJobId = clusterJobId;
}
/// <summary>
/// Extract the job information from a folder with logs on the local machine.
/// </summary>
/// <param name="jobRootFolder">Folder with logs for the specified job.</param>
/// <returns>The job information, or null if not found.</returns>
private ClusterJobInformation GetJobInfo(string jobRootFolder)
{
Uri uri = DfsFile.UriFromPath(this.config.JobsFolderUri, jobRootFolder);
long time;
long size;
this.config.DfsClient.GetFileStatus(uri, out time, out size);
DateTime date = DfsFile.TimeFromLong(time);
ClusterJobInformation.ClusterJobStatus status = ClusterJobInformation.ClusterJobStatus.Unknown;
string jobName = Path.GetFileName(jobRootFolder);
string errorMsg = "";
try
{
var jobinfo = this.yarnClient.QueryJob(jobName, uri);
var jobstatus = jobinfo.GetStatus();
errorMsg = jobinfo.ErrorMsg;
switch (jobstatus)
{
case JobStatus.NotSubmitted:
case JobStatus.Waiting:
status = ClusterJobInformation.ClusterJobStatus.Unknown;
break;
case JobStatus.Running:
status = ClusterJobInformation.ClusterJobStatus.Running;
break;
case JobStatus.Success:
status = ClusterJobInformation.ClusterJobStatus.Succeeded;
break;
case JobStatus.Cancelled:
status = ClusterJobInformation.ClusterJobStatus.Cancelled;
break;
case JobStatus.Failure:
status = ClusterJobInformation.ClusterJobStatus.Failed;
break;
default:
throw new ArgumentOutOfRangeException();
}
}
catch (Exception)
{
}
TimeSpan running = TimeSpan.Zero;
var info = new ClusterJobInformation(config.Name, "", jobName, jobName, Environment.UserName, date, running, status);
return(info);
}
/// <summary>
/// Extract the job information from a folder with logs on the local machine.
/// </summary>
/// <param name="jobRootFolder">Folder with logs for the specified job.</param>
/// <returns>The job information, or null if not found.</returns>
private ClusterJobInformation GetJobInfo(string jobRootFolder)
{
DateTime date = DateTime.MinValue;
DateTime lastHeartBeat = DateTime.MinValue;
ClusterJobInformation.ClusterJobStatus status = ClusterJobInformation.ClusterJobStatus.Unknown;
bool found = false;
Uri uri = AzureDfsFile.UriFromPath(this.config, jobRootFolder);
var jobsFolders = this.config.AzureClient.ExpandFileOrDirectory(uri).ToList();
jobRootFolder = GetBlobName(this.config.Container, jobRootFolder);
string jobName = jobRootFolder;
foreach (var file in jobsFolders)
{
if (file.AbsolutePath.EndsWith("heartbeat"))
{
string blobName = GetBlobName(this.config.Container, file.AbsolutePath);
var blob = this.config.AzureClient.Container.GetPageBlobReference(blobName);
blob.FetchAttributes();
var props = blob.Metadata;
if (props.ContainsKey("status"))
{
var st = props["status"];
switch (st)
{
case "failure":
status = ClusterJobInformation.ClusterJobStatus.Failed;
break;
case "success":
status = ClusterJobInformation.ClusterJobStatus.Succeeded;
break;
case "running":
status = ClusterJobInformation.ClusterJobStatus.Running;
break;
case "killed":
status = ClusterJobInformation.ClusterJobStatus.Cancelled;
break;
default:
Console.WriteLine("Unknown status " + st);
break;
}
}
if (props.ContainsKey("heartbeat"))
{
var hb = props["heartbeat"];
if (DateTime.TryParse(hb, out lastHeartBeat))
{
lastHeartBeat = lastHeartBeat.ToLocalTime();
if (status == ClusterJobInformation.ClusterJobStatus.Running &&
DateTime.Now - lastHeartBeat > TimeSpan.FromSeconds(40))
{
// job has in fact crashed
status = ClusterJobInformation.ClusterJobStatus.Failed;
}
}
}
if (props.ContainsKey("jobname"))
{
jobName = props["jobname"];
}
if (props.ContainsKey("starttime"))
{
var t = props["starttime"];
if (DateTime.TryParse(t, out date))
{
date = date.ToLocalTime();
}
}
found = true;
}
else if (file.AbsolutePath.Contains("DryadLinqProgram__") &&
// newer heartbeats contain the date
date != DateTime.MinValue)
{
var blob = this.config.AzureClient.Container.GetBlockBlobReference(AzureDfsFile.PathFromUri(this.config, file));
blob.FetchAttributes();
var props = blob.Properties;
if (props.LastModified.HasValue)
{
date = props.LastModified.Value.DateTime;
date = date.ToLocalTime();
}
}
}
if (!found)
{
return(null);
}
TimeSpan running = TimeSpan.Zero;
if (date != DateTime.MinValue && lastHeartBeat != DateTime.MinValue)
{
running = lastHeartBeat - date;
}
var info = new ClusterJobInformation(this.config.Name, "", jobRootFolder, jobName, Environment.UserName, date, running, status);
return(info);
}
