Add rich HTTP health endpoints for cluster monitoring

Enhance /api/health with component-level health, ServiceLevel, and
redundancy state for load balancer probes. Add /health HTML page for
operators to monitor node health in clustered System Platform deployments.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Joseph Doherty
2026-03-28 16:44:31 -04:00
parent f0a076ec26
commit 9d3599fbb6
5 changed files with 381 additions and 3 deletions

View File

@@ -132,6 +132,139 @@ namespace ZB.MOM.WW.LmxOpcUa.Tests.Status
sut.IsHealthy().ShouldBe(false);
}
[Fact]
public void GetHealthData_WhenConnected_ReturnsHealthyStatus()
{
var sut = CreateService();
var data = sut.GetHealthData();
data.Status.ShouldBe("Healthy");
data.Components.MxAccess.ShouldBe("Connected");
data.Components.Database.ShouldBe("Connected");
}
[Fact]
public void GetHealthData_WhenDisconnected_ReturnsUnhealthyStatus()
{
var mxClient = new FakeMxAccessClient { State = ConnectionState.Disconnected };
var galaxyStats = new GalaxyRepositoryStats { DbConnected = false };
var sut = new StatusReportService(new HealthCheckService(), 10);
sut.SetComponents(mxClient, null, galaxyStats, null);
var data = sut.GetHealthData();
data.Status.ShouldBe("Unhealthy");
data.ServiceLevel.ShouldBe((byte)0);
data.Components.MxAccess.ShouldBe("Disconnected");
data.Components.Database.ShouldBe("Disconnected");
}
[Fact]
public void GetHealthData_NoRedundancy_ServiceLevel255WhenHealthy()
{
var sut = CreateService();
var data = sut.GetHealthData();
data.RedundancyEnabled.ShouldBe(false);
data.ServiceLevel.ShouldBe((byte)255);
data.RedundancyRole.ShouldBeNull();
data.RedundancyMode.ShouldBeNull();
}
[Fact]
public void GetHealthData_WithRedundancy_IncludesRoleAndServiceLevel()
{
var sut = CreateServiceWithRedundancy("Primary");
var data = sut.GetHealthData();
data.RedundancyEnabled.ShouldBe(true);
data.RedundancyRole.ShouldBe("Primary");
data.RedundancyMode.ShouldBe("Warm");
data.ServiceLevel.ShouldBe((byte)200);
}
[Fact]
public void GetHealthData_SecondaryRole_LowerServiceLevel()
{
var sut = CreateServiceWithRedundancy("Secondary");
var data = sut.GetHealthData();
data.ServiceLevel.ShouldBe((byte)150);
}
[Fact]
public void GetHealthData_ContainsUptime()
{
var sut = CreateService();
var data = sut.GetHealthData();
data.Uptime.ShouldNotBeNullOrWhiteSpace();
}
[Fact]
public void GetHealthData_ContainsTimestamp()
{
var sut = CreateService();
var data = sut.GetHealthData();
data.Timestamp.ShouldBeGreaterThan(DateTime.UtcNow.AddMinutes(-1));
}
[Fact]
public void GenerateHealthJson_ContainsExpectedFields()
{
var sut = CreateService();
var json = sut.GenerateHealthJson();
json.ShouldContain("Status");
json.ShouldContain("ServiceLevel");
json.ShouldContain("Components");
json.ShouldContain("MxAccess");
json.ShouldContain("Database");
json.ShouldContain("OpcUaServer");
json.ShouldContain("Uptime");
}
[Fact]
public void GenerateHealthHtml_ContainsStatusBadge()
{
var sut = CreateService();
var html = sut.GenerateHealthHtml();
html.ShouldContain("HEALTHY");
html.ShouldContain("SERVICE LEVEL");
html.ShouldContain("255");
}
[Fact]
public void GenerateHealthHtml_ContainsComponentCards()
{
var sut = CreateService();
var html = sut.GenerateHealthHtml();
html.ShouldContain("MXAccess");
html.ShouldContain("Galaxy Database");
html.ShouldContain("OPC UA Server");
}
[Fact]
public void GenerateHealthHtml_WithRedundancy_ShowsRoleAndMode()
{
var sut = CreateServiceWithRedundancy("Primary");
var html = sut.GenerateHealthHtml();
html.ShouldContain("Primary");
html.ShouldContain("Warm");
}
[Fact]
public void GenerateHealthHtml_ContainsAutoRefresh()
{
var sut = CreateService();
var html = sut.GenerateHealthHtml();
html.ShouldContain("meta http-equiv='refresh' content='10'");
}
/// <summary>
/// Creates a status report service preloaded with representative runtime, Galaxy, and metrics data.
/// </summary>
@@ -157,5 +290,21 @@ namespace ZB.MOM.WW.LmxOpcUa.Tests.Status
sut.SetComponents(mxClient, metrics, galaxyStats, null);
return sut;
}
private static StatusReportService CreateServiceWithRedundancy(string role)
{
var mxClient = new FakeMxAccessClient();
var galaxyStats = new GalaxyRepositoryStats { GalaxyName = "TestGalaxy", DbConnected = true };
var redundancyConfig = new Host.Configuration.RedundancyConfiguration
{
Enabled = true,
Mode = "Warm",
Role = role,
ServiceLevelBase = 200
};
var sut = new StatusReportService(new HealthCheckService(), 10);
sut.SetComponents(mxClient, null, galaxyStats, null, null, redundancyConfig, "urn:test:instance1");
return sut;
}
}
}

View File

@@ -107,6 +107,35 @@ namespace ZB.MOM.WW.LmxOpcUa.Tests.Status
response.Headers.CacheControl?.NoStore.ShouldBe(true);
}
/// <summary>
/// Confirms that the /health route returns an HTML health page.
/// </summary>
[Fact]
public async Task HealthPage_ReturnsHtml200()
{
var response = await _client.GetAsync("/health");
response.StatusCode.ShouldBe(HttpStatusCode.OK);
response.Content.Headers.ContentType?.MediaType.ShouldBe("text/html");
var body = await response.Content.ReadAsStringAsync();
body.ShouldContain("SERVICE LEVEL");
body.ShouldContain("MXAccess");
}
/// <summary>
/// Confirms that /api/health returns rich JSON with component health details.
/// </summary>
[Fact]
public async Task ApiHealth_ReturnsRichJson()
{
var response = await _client.GetAsync("/api/health");
response.StatusCode.ShouldBe(HttpStatusCode.OK);
response.Content.Headers.ContentType?.MediaType.ShouldBe("application/json");
var body = await response.Content.ReadAsStringAsync();
body.ShouldContain("ServiceLevel");
body.ShouldContain("Components");
body.ShouldContain("Uptime");
}
/// <summary>
/// Confirms that the server can be started and stopped cleanly.
/// </summary>