/// <summary>process datanode heartbeat or stats initialization.</summary>
public virtual void UpdateHeartbeatState(StorageReport[] reports, long cacheCapacity
, long cacheUsed, int xceiverCount, int volFailures, VolumeFailureSummary volumeFailureSummary
)
{
long totalCapacity = 0;
long totalRemaining = 0;
long totalBlockPoolUsed = 0;
long totalDfsUsed = 0;
ICollection <DatanodeStorageInfo> failedStorageInfos = null;
// Decide if we should check for any missing StorageReport and mark it as
// failed. There are different scenarios.
// 1. When DN is running, a storage failed. Given the current DN
// implementation doesn't add recovered storage back to its storage list
// until DN restart, we can assume volFailures won't decrease
// during the current DN registration session.
// When volumeFailures == this.volumeFailures, it implies there is no
// state change. No need to check for failed storage. This is an
// optimization. Recent versions of the DataNode report a
// VolumeFailureSummary containing the date/time of the last volume
// failure. If that's available, then we check that instead for greater
// accuracy.
// 2. After DN restarts, volFailures might not increase and it is possible
// we still have new failed storage. For example, admins reduce
// available storages in configuration. Another corner case
// is the failed volumes might change after restart; a) there
// is one good storage A, one restored good storage B, so there is
// one element in storageReports and that is A. b) A failed. c) Before
// DN sends HB to NN to indicate A has failed, DN restarts. d) After DN
// restarts, storageReports has one element which is B.
bool checkFailedStorages;
if (volumeFailureSummary != null && this.volumeFailureSummary != null)
{
checkFailedStorages = volumeFailureSummary.GetLastVolumeFailureDate() > this.volumeFailureSummary
.GetLastVolumeFailureDate();
}
else
{
checkFailedStorages = (volFailures > this.volumeFailures) || !heartbeatedSinceRegistration;
}
if (checkFailedStorages)
{
Log.Info("Number of failed storage changes from " + this.volumeFailures + " to "
+ volFailures);
failedStorageInfos = new HashSet <DatanodeStorageInfo>(storageMap.Values);
}
SetCacheCapacity(cacheCapacity);
SetCacheUsed(cacheUsed);
SetXceiverCount(xceiverCount);
SetLastUpdate(Time.Now());
SetLastUpdateMonotonic(Time.MonotonicNow());
this.volumeFailures = volFailures;
this.volumeFailureSummary = volumeFailureSummary;
foreach (StorageReport report in reports)
{
DatanodeStorageInfo storage = UpdateStorage(report.GetStorage());
if (checkFailedStorages)
{
failedStorageInfos.Remove(storage);
}
storage.ReceivedHeartbeat(report);
totalCapacity += report.GetCapacity();
totalRemaining += report.GetRemaining();
totalBlockPoolUsed += report.GetBlockPoolUsed();
totalDfsUsed += report.GetDfsUsed();
}
RollBlocksScheduled(GetLastUpdateMonotonic());
// Update total metrics for the node.
SetCapacity(totalCapacity);
SetRemaining(totalRemaining);
SetBlockPoolUsed(totalBlockPoolUsed);
SetDfsUsed(totalDfsUsed);
if (checkFailedStorages)
{
UpdateFailedStorage(failedStorageInfos);
}
if (storageMap.Count != reports.Length)
{
PruneStorageMap(reports);
}
}