/// <summary>
/// Verifies that a cluster worker node is healthy.
/// </summary>
/// <param name="node">The server node.</param>
/// <param name="clusterDefinition">The cluster definition.</param>
public static void CheckWorker(NodeSshProxy <NodeDefinition> node, ClusterDefinition clusterDefinition)
{
Covenant.Requires <ArgumentNullException>(node != null, nameof(node));
Covenant.Requires <ArgumentException>(node.Metadata.IsWorker, nameof(node));
Covenant.Requires <ArgumentNullException>(clusterDefinition != null, nameof(clusterDefinition));
if (!node.IsFaulted)
{
CheckWorkerNtp(node, clusterDefinition);
}
node.Status = "healthy";
}
/// <summary>
/// Verifies that a control-plane node's NTP health.
/// </summary>
/// <param name="node">The control-plane node.</param>
/// <param name="clusterDefinition">The cluster definition.</param>
private static void CheckControlNodeNtp(NodeSshProxy <NodeDefinition> node, ClusterDefinition clusterDefinition)
{
// We're going to use [ntpq -pw] to query the configured time sources.
// We should get something back that looks like
//
// remote refid st t when poll reach delay offset jitter
// ==============================================================================
// LOCAL(0).LOCL. 10 l 45m 64 0 0.000 0.000 0.000
// * clock.xmission. .GPS. 1 u 134 256 377 48.939 - 0.549 18.357
// + 173.44.32.10 18.26.4.105 2 u 200 256 377 96.981 - 0.623 3.284
// + pacific.latt.ne 44.24.199.34 3 u 243 256 377 41.457 - 8.929 8.497
//
// For control-plane nodes, we're simply going to verify that we have at least one external
// time source answering.
node.Status = "check: NTP";
var retryDelay = TimeSpan.FromSeconds(30);
var fault = (string)null;
for (int tryCount = 0; tryCount < 6; tryCount++)
{
var response = node.SudoCommand("/usr/bin/ntpq -pw", RunOptions.LogOutput);
if (response.ExitCode != 0)
{
Thread.Sleep(retryDelay);
continue;
}
using (var reader = response.OpenOutputTextReader())
{
string line;
// Column header and table bar lines.
line = reader.ReadLine();
if (string.IsNullOrWhiteSpace(line))
{
fault = "NTP: Invalid [ntpq -pw] response.";
Thread.Sleep(retryDelay);
continue;
}
line = reader.ReadLine();
if (string.IsNullOrWhiteSpace(line) || line[0] != '=')
{
fault = "NTP: Invalid [ntpq -pw] response.";
Thread.Sleep(retryDelay);
continue;
}
// Count the lines starting that don't include [*.LOCL.*],
// the local clock.
var sourceCount = 0;
for (line = reader.ReadLine(); line != null; line = reader.ReadLine())
{
if (line.Length > 0 && !line.Contains(".LOCL."))
{
sourceCount++;
}
}
if (sourceCount == 0)
{
fault = "NTP: No external sources are answering.";
Thread.Sleep(retryDelay);
continue;
}
// Everything looks good.
break;
}
}
if (fault != null)
{
node.Fault(fault);
}
}
/// <summary>
/// Verifies that a worker node's NTP health.
/// </summary>
/// <param name="node">The worker node.</param>
/// <param name="clusterDefinition">The cluster definition.</param>
private static void CheckWorkerNtp(NodeSshProxy <NodeDefinition> node, ClusterDefinition clusterDefinition)
{
// We're going to use [ntpq -pw] to query the configured time sources.
// We should get something back that looks like
//
// remote refid st t when poll reach delay offset jitter
// ==============================================================================
// LOCAL(0).LOCL. 10 l 45m 64 0 0.000 0.000 0.000
// * 10.0.1.5 198.60.22.240 2 u 111 128 377 0.062 3.409 0.608
// + 10.0.1.7 198.60.22.240 2 u 111 128 377 0.062 3.409 0.608
// + 10.0.1.7 198.60.22.240 2 u 111 128 377 0.062 3.409 0.608
//
// For worker nodes, we need to verify that each of the control-plane nodes are answering
// by confirming that their IP addresses are present.
node.Status = "check: NTP";
var retryDelay = TimeSpan.FromSeconds(30);
var fault = (string)null;
var firstTry = true;
tryAgain:
for (var tries = 0; tries < 6; tries++)
{
var output = node.SudoCommand("/usr/bin/ntpq -pw", RunOptions.LogOutput).OutputText;
foreach (var controlNode in clusterDefinition.SortedControlNodes)
{
// We're going to check the for presence of the control-plane node's IP address
// or its name, the latter because [ntpq] appears to attempt a reverse
// IP address lookup which will resolve into one of the DNS names defined
// in the local [/etc/hosts] file.
if (!output.Contains(controlNode.Address.ToString()) && !output.Contains(controlNode.Name.ToLower()))
{
fault = $"NTP: Manager [{controlNode.Name}/{controlNode.Address}] is not answering.";
Thread.Sleep(retryDelay);
continue;
}
// Everything looks OK.
break;
}
}
if (fault != null)
{
if (firstTry)
{
// $hack(jefflill):
//
// I've seen the NTP check fail on worker nodes, complaining
// that the connection attempt was rejected. I manually restarted
// the node and then it worked. I'm not sure if the rejected connection
// was being made to the local NTP service or from the local service
// to NTP running on the control-plane.
//
// I'm going to assume that it was to the local NTP service and I'm
// going to try mitigating this by restarting the local NTP service
// and then re-running the tests. I'm only going to do this once.
node.SudoCommand("systemctl restart ntp", node.DefaultRunOptions & ~RunOptions.FaultOnError);
firstTry = false;
goto tryAgain;
}
node.Fault(fault);
}
}
/// <summary>
/// Inspects the node to determine physical machine capabilities like
/// processor count, RAM, and primary disk capacity and then sets the
/// corresponding node labels in the cluster definition.
/// </summary>
/// <param name="node">The target node.</param>
private void DetectLabels(NodeSshProxy <NodeDefinition> node)
{
CommandResponse result;
// Download [/proc/meminfo] and extract the [MemTotal] value (in kB).
result = node.SudoCommand("cat /proc/meminfo", RunOptions.FaultOnError);
if (result.ExitCode == 0)
{
var memInfo = result.OutputText;
var memTotalRegex = new Regex(@"^MemTotal:\s*(?<size>\d+)\s*kB", RegexOptions.Multiline);
var memMatch = memTotalRegex.Match(memInfo);
if (memMatch.Success && long.TryParse(memMatch.Groups["size"].Value, out var memSizeKiB))
{
// Note that the RAM reported by Linux is somewhat less than the
// physical RAM installed.
node.Metadata.Labels.ComputeRam = (int)(memSizeKiB / 1024); // Convert KiB --> MiB
}
}
// Download [/proc/cpuinfo] and count the number of processors.
result = node.SudoCommand("cat /proc/cpuinfo", RunOptions.FaultOnError);
if (result.ExitCode == 0)
{
var cpuInfo = result.OutputText;
var processorRegex = new Regex(@"^processor\s*:\s*\d+", RegexOptions.Multiline);
var processorMatches = processorRegex.Matches(cpuInfo);
node.Metadata.Labels.ComputeCores = processorMatches.Count;
}
// Determine the primary disk size.
// $hack(jefflill):
//
// I'm not entirely sure how to determine which block device is hosting
// the primary file system for all systems. For now, I'm just going to
// assume that this can be one of:
//
// /dev/sda1
// /dev/sda
// /dev/xvda1
// /dev/xvda
//
// I'll try each of these in order and setting the label for the
// first reasonable result we get back.
var blockDevices = new string[]
{
"/dev/sda1",
"/dev/sda",
"/dev/xvda1",
"/dev/xvda"
};
foreach (var blockDevice in blockDevices)
{
result = node.SudoCommand($"lsblk -b --output SIZE -n -d {blockDevice}", RunOptions.LogOutput | RunOptions.FaultOnError);
if (result.ExitCode == 0)
{
if (long.TryParse(result.OutputText.Trim(), out var deviceSize) && deviceSize > 0)
{
node.Metadata.Labels.StorageSize = ByteUnits.ToGiB(deviceSize);
break;
}
}
}
}
