consul/internal/catalog/catalogtest/test_lifecycle_v1alpha1.go

717 lines
30 KiB
Go
Raw Normal View History

[COMPLIANCE] License changes (#18443) * Adding explicit MPL license for sub-package This directory and its subdirectories (packages) contain files licensed with the MPLv2 `LICENSE` file in this directory and are intentionally licensed separately from the BSL `LICENSE` file at the root of this repository. * Adding explicit MPL license for sub-package This directory and its subdirectories (packages) contain files licensed with the MPLv2 `LICENSE` file in this directory and are intentionally licensed separately from the BSL `LICENSE` file at the root of this repository. * Updating the license from MPL to Business Source License Going forward, this project will be licensed under the Business Source License v1.1. Please see our blog post for more details at <Blog URL>, FAQ at www.hashicorp.com/licensing-faq, and details of the license at www.hashicorp.com/bsl. * add missing license headers * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 * Update copyright file headers to BUSL-1.1 --------- Co-authored-by: hashicorp-copywrite[bot] <110428419+hashicorp-copywrite[bot]@users.noreply.github.com>
2023-08-11 13:12:13 +00:00
// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package catalogtest
import (
"testing"
"github.com/hashicorp/consul/internal/catalog"
rtest "github.com/hashicorp/consul/internal/resource/resourcetest"
pbcatalog "github.com/hashicorp/consul/proto-public/pbcatalog/v1alpha1"
"github.com/hashicorp/consul/proto-public/pbresource"
"github.com/hashicorp/consul/sdk/testutil"
)
// RunCatalogV1Alpha1LifecycleIntegrationTest intends to excercise functionality of
// managing catalog resources over their normal lifecycle where they will be modified
// several times, change state etc.
func RunCatalogV1Alpha1LifecycleIntegrationTest(t *testing.T, client pbresource.ResourceServiceClient) {
t.Helper()
testutil.RunStep(t, "node-lifecycle", func(t *testing.T) {
RunCatalogV1Alpha1NodeLifecycleIntegrationTest(t, client)
})
testutil.RunStep(t, "workload-lifecycle", func(t *testing.T) {
RunCatalogV1Alpha1WorkloadLifecycleIntegrationTest(t, client)
})
testutil.RunStep(t, "endpoints-lifecycle", func(t *testing.T) {
RunCatalogV1Alpha1EndpointsLifecycleIntegrationTest(t, client)
})
}
// RunCatalogV1Alpha1NodeLifecycleIntegrationTest verifies correct functionality of
// the node-health controller. This test will exercise the following behaviors:
//
// * Creating a Node without associated HealthStatuses will mark the node as passing
// * Associating a HealthStatus with a Node will cause recomputation of the Health
// * Changing HealthStatus to a worse health will cause recomputation of the Health
// * Changing HealthStatus to a better health will cause recomputation of the Health
// * Deletion of associated HealthStatuses will recompute the Health (back to passing)
// * Deletion of the node will cause deletion of associated health statuses
func RunCatalogV1Alpha1NodeLifecycleIntegrationTest(t *testing.T, client pbresource.ResourceServiceClient) {
c := rtest.NewClient(client)
nodeName := "test-lifecycle"
nodeHealthName := "test-lifecycle-node-status"
// initial node creation
node := rtest.Resource(catalog.NodeV1Alpha1Type, nodeName).
WithData(t, &pbcatalog.Node{
Addresses: []*pbcatalog.NodeAddress{
{Host: "172.16.2.3"},
{Host: "198.18.2.3", External: true},
},
}).
Write(t, c)
// wait for the node health controller to mark the node as healthy
c.WaitForStatusCondition(t, node.Id,
catalog.NodeHealthStatusKey,
catalog.NodeHealthConditions[pbcatalog.Health_HEALTH_PASSING])
// Its easy enough to simply repeatedly set the health status and it proves
// that going both from better to worse health and worse to better all
// happen as expected. We leave the health in a warning state to allow for
// the subsequent health status deletion to cause the health to go back
// to passing.
healthChanges := []pbcatalog.Health{
pbcatalog.Health_HEALTH_PASSING,
pbcatalog.Health_HEALTH_WARNING,
pbcatalog.Health_HEALTH_CRITICAL,
pbcatalog.Health_HEALTH_MAINTENANCE,
pbcatalog.Health_HEALTH_CRITICAL,
pbcatalog.Health_HEALTH_WARNING,
pbcatalog.Health_HEALTH_PASSING,
pbcatalog.Health_HEALTH_WARNING,
}
// This will be set within the loop and used afterwards to delete the health status
var nodeHealth *pbresource.Resource
// Iterate through the various desired health statuses, updating
// a HealthStatus resource owned by the node and waiting for
// reconciliation at each point
for _, health := range healthChanges {
// update the health check
nodeHealth = setHealthStatus(t, c, node.Id, nodeHealthName, health)
// wait for reconciliation to kick in and put the node into the right
// health status.
c.WaitForStatusCondition(t, node.Id,
catalog.NodeHealthStatusKey,
catalog.NodeHealthConditions[health])
}
// now delete the health status and ensure things go back to passing
c.MustDelete(t, nodeHealth.Id)
// wait for the node health controller to mark the node as healthy
c.WaitForStatusCondition(t, node.Id,
catalog.NodeHealthStatusKey,
catalog.NodeHealthConditions[pbcatalog.Health_HEALTH_PASSING])
// Add the health status back once more, the actual status doesn't matter.
// It just must be owned by the node so that we can show cascading
// deletions of owned health statuses working.
healthStatus := setHealthStatus(t, c, node.Id, nodeHealthName, pbcatalog.Health_HEALTH_CRITICAL)
// Delete the node and wait for the health status to be deleted.
c.MustDelete(t, node.Id)
c.WaitForDeletion(t, healthStatus.Id)
}
// RunCatalogV1Alpha1WorkloadLifecycleIntegrationTest verifies correct functionality of
// the workload-health controller. This test will exercise the following behaviors:
//
// - Associating a workload with a node causes recomputation of the health and takes
// into account the nodes health
// - Modifying the workloads associated node causes health recomputation and takes into
// account the new nodes health
// - Removal of the node association causes recomputation of health and for no node health
// to be taken into account.
// - Creating a workload without associated health statuses or node association will
// be marked passing
// - Creating a workload without associated health statuses but with a node will
// inherit its health from the node.
// - Changing HealthStatus to a worse health will cause recompuation of the Health
// - Changing HealthStatus to a better health will cause recompuation of the Health
// - Overall health is computed as the worst health amongst the nodes health and all
// of the workloads associated HealthStatuses
// - Deletion of the workload will cause deletion of all associated health statuses.
func RunCatalogV1Alpha1WorkloadLifecycleIntegrationTest(t *testing.T, client pbresource.ResourceServiceClient) {
c := rtest.NewClient(client)
testutil.RunStep(t, "nodeless-workload", func(t *testing.T) {
runV1Alpha1NodelessWorkloadLifecycleIntegrationTest(t, c)
})
testutil.RunStep(t, "node-associated-workload", func(t *testing.T) {
runV1Alpha1NodeAssociatedWorkloadLifecycleIntegrationTest(t, c)
})
}
// runV1Alpha1NodelessWorkloadLifecycleIntegrationTest verifies correct functionality of
// the workload-health controller for workloads without node associations. In particular
// the following behaviors are being tested
//
// - Creating a workload without associated health statuses or node association will
// be marked passing
// - Changing HealthStatus to a worse health will cause recompuation of the Health
// - Changing HealthStatus to a better health will cause recompuation of the Health
// - Deletion of associated HealthStatus for a nodeless workload will be set back to passing
// - Deletion of the workload will cause deletion of all associated health statuses.
func runV1Alpha1NodelessWorkloadLifecycleIntegrationTest(t *testing.T, c *rtest.Client) {
workloadName := "test-lifecycle-workload"
workloadHealthName := "test-lifecycle-workload-status"
// create a workload without a node association or health statuses yet
workload := rtest.Resource(catalog.WorkloadV1Alpha1Type, workloadName).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "198.18.9.8"},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
Identity: "test-lifecycle",
}).
Write(t, c)
// wait for the workload health controller to mark the workload as healthy
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadHealthConditions[pbcatalog.Health_HEALTH_PASSING])
// We may not need to iterate through all of these states but its easy
// enough and quick enough to do so. The general rationale is that we
// should move through changing the workloads associated health status
// in this progression. We can prove that moving from better to worse
// health or worse to better both function correctly.
healthChanges := []pbcatalog.Health{
pbcatalog.Health_HEALTH_PASSING,
pbcatalog.Health_HEALTH_WARNING,
pbcatalog.Health_HEALTH_CRITICAL,
pbcatalog.Health_HEALTH_MAINTENANCE,
pbcatalog.Health_HEALTH_CRITICAL,
pbcatalog.Health_HEALTH_WARNING,
pbcatalog.Health_HEALTH_PASSING,
pbcatalog.Health_HEALTH_WARNING,
}
var workloadHealth *pbresource.Resource
// Iterate through the various desired health statuses, updating
// a HealthStatus resource owned by the workload and waiting for
// reconciliation at each point
for _, health := range healthChanges {
// update the health status
workloadHealth = setHealthStatus(t, c, workload.Id, workloadHealthName, health)
// wait for reconciliation to kick in and put the workload into
// the right health status.
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadHealthConditions[health])
}
// Now delete the health status, things should go back to passing status
c.MustDelete(t, workloadHealth.Id)
// ensure the workloads health went back to passing
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadHealthConditions[pbcatalog.Health_HEALTH_PASSING])
// Reset the workload health. The actual health is irrelevant, we just want it
// to exist to provde that Health Statuses get deleted along with the workload
// when its deleted.
workloadHealth = setHealthStatus(t, c, workload.Id, workloadHealthName, pbcatalog.Health_HEALTH_WARNING)
// Delete the workload and wait for the HealthStatus to also be deleted
c.MustDelete(t, workload.Id)
c.WaitForDeletion(t, workloadHealth.Id)
}
// runV1Alpha1NodeAssociatedWorkloadLifecycleIntegrationTest verifies correct functionality of
// the workload-health controller. This test will exercise the following behaviors:
//
// - Associating a workload with a node causes recomputation of the health and takes
// into account the nodes health
// - Modifying the workloads associated node causes health recomputation and takes into
// account the new nodes health
// - Removal of the node association causes recomputation of health and for no node health
// to be taken into account.
// - Creating a workload without associated health statuses but with a node will
// inherit its health from the node.
// - Overall health is computed as the worst health amongst the nodes health and all
// of the workloads associated HealthStatuses
func runV1Alpha1NodeAssociatedWorkloadLifecycleIntegrationTest(t *testing.T, c *rtest.Client) {
workloadName := "test-lifecycle"
workloadHealthName := "test-lifecycle"
nodeName1 := "test-lifecycle-1"
nodeName2 := "test-lifecycle-2"
nodeHealthName1 := "test-lifecycle-node-1"
nodeHealthName2 := "test-lifecycle-node-2"
// Insert a some nodes to link the workloads to at various points throughout the test
node1 := rtest.Resource(catalog.NodeV1Alpha1Type, nodeName1).
WithData(t, &pbcatalog.Node{
Addresses: []*pbcatalog.NodeAddress{{Host: "172.17.9.10"}},
}).
Write(t, c)
node2 := rtest.Resource(catalog.NodeV1Alpha1Type, nodeName2).
WithData(t, &pbcatalog.Node{
Addresses: []*pbcatalog.NodeAddress{{Host: "172.17.9.11"}},
}).
Write(t, c)
// Set some non-passing health statuses for those nodes. Using non-passing will make
// it easy to see that changing a passing workloads node association appropriately
// impacts the overall workload health.
setHealthStatus(t, c, node1.Id, nodeHealthName1, pbcatalog.Health_HEALTH_CRITICAL)
setHealthStatus(t, c, node2.Id, nodeHealthName2, pbcatalog.Health_HEALTH_WARNING)
// Add the workload but don't immediately associate with any node.
workload := rtest.Resource(catalog.WorkloadV1Alpha1Type, workloadName).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "198.18.9.8"},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
Identity: "test-lifecycle",
}).
Write(t, c)
// wait for the workload health controller to mark the workload as healthy
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadHealthConditions[pbcatalog.Health_HEALTH_PASSING])
// now modify the workload to associate it with node 1 (currently with CRITICAL health)
workload = rtest.ResourceID(workload.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{{Host: "198.18.9.8"}},
Ports: map[string]*pbcatalog.WorkloadPort{"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
Identity: "test-lifecycle",
// this is the only difference from the previous write
NodeName: node1.Id.Name,
}).
Write(t, c)
// wait for the workload health controller to mark the workload as critical (due to node 1 having critical health)
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_PASSING][pbcatalog.Health_HEALTH_CRITICAL])
// Now reassociate the workload with node 2. This should cause recalculation of its health into the warning state
workload = rtest.ResourceID(workload.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{{Host: "198.18.9.8"}},
Ports: map[string]*pbcatalog.WorkloadPort{"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
Identity: "test-lifecycle",
// this is the only difference from the previous write
NodeName: node2.Id.Name,
}).
Write(t, c)
// Wait for the workload health controller to mark the workload as warning (due to node 2 having warning health)
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_PASSING][pbcatalog.Health_HEALTH_WARNING])
// Delete the node, this should cause the health to be recalculated as critical because the node association
// is broken.
c.MustDelete(t, node2.Id)
// Wait for the workload health controller to mark the workload as critical due to the missing node
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_PASSING][pbcatalog.Health_HEALTH_CRITICAL])
// Now fixup the node association to point at node 1
workload = rtest.ResourceID(workload.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{{Host: "198.18.9.8"}},
Ports: map[string]*pbcatalog.WorkloadPort{"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
Identity: "test-lifecycle",
// this is the only difference from the previous write
NodeName: node1.Id.Name,
}).
Write(t, c)
// Also set node 1 health down to WARNING
setHealthStatus(t, c, node1.Id, nodeHealthName1, pbcatalog.Health_HEALTH_WARNING)
// Wait for the workload health controller to mark the workload as warning (due to node 1 having warning health now)
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_PASSING][pbcatalog.Health_HEALTH_WARNING])
// Now add a critical workload health check to ensure that both node and workload health are accounted for.
setHealthStatus(t, c, workload.Id, workloadHealthName, pbcatalog.Health_HEALTH_CRITICAL)
// Wait for the workload health to be recomputed and put into the critical status.
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_CRITICAL][pbcatalog.Health_HEALTH_WARNING])
// Reset the workloads health to passing. We expect the overall health to go back to warning
setHealthStatus(t, c, workload.Id, workloadHealthName, pbcatalog.Health_HEALTH_PASSING)
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadAndNodeHealthConditions[pbcatalog.Health_HEALTH_PASSING][pbcatalog.Health_HEALTH_WARNING])
// Remove the node association and wait for the health to go back to passing
workload = rtest.ResourceID(workload.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{{Host: "198.18.9.8"}},
Ports: map[string]*pbcatalog.WorkloadPort{"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
Identity: "test-lifecycle",
}).
Write(t, c)
c.WaitForStatusCondition(t, workload.Id,
catalog.WorkloadHealthStatusKey,
catalog.WorkloadHealthConditions[pbcatalog.Health_HEALTH_PASSING])
}
// RunCatalogV1Alpha1EndpointsLifecycleIntegrationTest verifies the correct functionality of
// the endpoints controller. This test will exercise the following behaviors:
//
// * Services without a selector get marked with status indicating their endpoints are unmanaged
// * Services with a selector get marked with status indicating their endpoints are managed
// * Deleting a service will delete the associated endpoints (regardless of them being managed or not)
// * Moving from managed to unmanaged endpoints will delete the managed endpoints
// * Moving from unmanaged to managed endpoints will overwrite any previous endpoints.
// * A service with a selector that matches no workloads will still have the endpoints object written.
// * Adding ports to a service will recalculate the endpoints
// * Removing ports from a service will recalculate the endpoints
// * Changing the workload will recalculate the endpoints (ports, addresses, or health)
func RunCatalogV1Alpha1EndpointsLifecycleIntegrationTest(t *testing.T, client pbresource.ResourceServiceClient) {
c := rtest.NewClient(client)
serviceName := "test-lifecycle"
// Create the service without a selector. We should not see endpoints generated but we should see the
// status updated to note endpoints are not being managed.
service := rtest.Resource(catalog.ServiceV1Alpha1Type, serviceName).
WithData(t, &pbcatalog.Service{
Ports: []*pbcatalog.ServicePort{{TargetPort: "http", Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
}).
Write(t, c)
// Wait to ensure the status is updated accordingly
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionUnmanaged)
// Verify that no endpoints were created.
endpointsID := rtest.Resource(catalog.ServiceEndpointsV1Alpha1Type, serviceName).ID()
c.RequireResourceNotFound(t, endpointsID)
// Add some empty endpoints (type validations enforce that they are owned by the service)
rtest.ResourceID(endpointsID).
WithData(t, &pbcatalog.ServiceEndpoints{}).
WithOwner(service.Id).
Write(t, c)
// Now delete the service and ensure that they are cleaned up.
c.MustDelete(t, service.Id)
c.WaitForDeletion(t, endpointsID)
// Add some workloads to eventually select by the service
// api-1 has all ports (http, grpc and mesh). It also has a mixture of Addresses
// that select individual ports and one that selects all ports implicitly
api1 := rtest.Resource(catalog.WorkloadV1Alpha1Type, "api-1").
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1"},
{Host: "::1", Ports: []string{"grpc"}},
{Host: "127.0.0.2", Ports: []string{"http"}},
{Host: "172.17.1.1", Ports: []string{"mesh"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"mesh": {Port: 10000, Protocol: pbcatalog.Protocol_PROTOCOL_MESH},
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
Identity: "api",
}).
Write(t, c)
// api-2 has only grpc and mesh ports. It also has a mixture of Addresses that
// select individual ports and one that selects all ports implicitly
api2 := rtest.Resource(catalog.WorkloadV1Alpha1Type, "api-2").
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1"},
{Host: "::1", Ports: []string{"grpc"}},
{Host: "172.17.1.2", Ports: []string{"mesh"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"mesh": {Port: 10000, Protocol: pbcatalog.Protocol_PROTOCOL_MESH},
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
Identity: "api",
}).
Write(t, c)
// api-3 has the mesh and HTTP ports. It also has a mixture of Addresses that
// select individual ports and one that selects all ports.
api3 := rtest.Resource(catalog.WorkloadV1Alpha1Type, "api-3").
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1"},
{Host: "172.17.1.3", Ports: []string{"mesh"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"mesh": {Port: 10000, Protocol: pbcatalog.Protocol_PROTOCOL_MESH},
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
Identity: "api",
}).
Write(t, c)
// Now create a service with unmanaged endpoints again
service = rtest.Resource(catalog.ServiceV1Alpha1Type, serviceName).
WithData(t, &pbcatalog.Service{
Ports: []*pbcatalog.ServicePort{{TargetPort: "http", Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
}).
Write(t, c)
// Inject the endpoints resource. We want to prove that transition from unmanaged to
// managed endpoints results in overwriting of the old endpoints
rtest.ResourceID(endpointsID).
WithData(t, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "198.18.1.1", External: true},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"http": {Port: 443, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
},
}).
WithOwner(service.Id).
Write(t, c)
// Wait to ensure the status is updated accordingly
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionUnmanaged)
// Now move the service to having managed endpoints
service = rtest.ResourceID(service.Id).
WithData(t, &pbcatalog.Service{
Workloads: &pbcatalog.WorkloadSelector{Names: []string{"bar"}},
Ports: []*pbcatalog.ServicePort{{TargetPort: "http", Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
}).
Write(t, c)
// Verify that this status is updated to show this service as having managed endpoints
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionManaged)
// Verify that the service endpoints are created. In this case they will be empty
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{})
// Rewrite the service to select the API workloads - just select the singular port for now
service = rtest.ResourceID(service.Id).
WithData(t, &pbcatalog.Service{
Workloads: &pbcatalog.WorkloadSelector{Prefixes: []string{"api-"}},
Ports: []*pbcatalog.ServicePort{{TargetPort: "http", Protocol: pbcatalog.Protocol_PROTOCOL_HTTP}},
}).
Write(t, c)
// Wait for the status to be updated. The condition itself will remain unchanged but we are waiting for
// the generations to match to know that the endpoints would have been regenerated
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionManaged)
// ensure that api-1 and api-3 are selected but api-2 is excluded due to not having the desired port
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
TargetRef: api1.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"http"}},
{Host: "127.0.0.2", Ports: []string{"http"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
{
TargetRef: api3.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"http"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
},
})
// Rewrite the service to select the API workloads - changing from selecting the HTTP port to the gRPC port
service = rtest.ResourceID(service.Id).
WithData(t, &pbcatalog.Service{
Workloads: &pbcatalog.WorkloadSelector{Prefixes: []string{"api-"}},
Ports: []*pbcatalog.ServicePort{{TargetPort: "grpc", Protocol: pbcatalog.Protocol_PROTOCOL_GRPC}},
}).
Write(t, c)
// Wait for the status to be updated. The condition itself will remain unchanged but we are waiting for
// the generations to match to know that the endpoints would have been regenerated
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionManaged)
// Check that the endpoints were generated as expected
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
TargetRef: api1.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"grpc"}},
{Host: "::1", Ports: []string{"grpc"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
{
TargetRef: api2.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"grpc"}},
{Host: "::1", Ports: []string{"grpc"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
},
})
// Update the service to change the ports used. This should result in the workload being removed
// from the endpoints
rtest.ResourceID(api2.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1"},
{Host: "::1", Ports: []string{"http"}},
{Host: "172.17.1.2", Ports: []string{"mesh"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"mesh": {Port: 10000, Protocol: pbcatalog.Protocol_PROTOCOL_MESH},
"http": {Port: 8080, Protocol: pbcatalog.Protocol_PROTOCOL_HTTP},
},
Identity: "api",
}).
Write(t, c)
// Verify that api-2 was removed from the service endpoints as it no longer has a grpc port
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
TargetRef: api1.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"grpc"}},
{Host: "::1", Ports: []string{"grpc"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
},
})
// Remove the ::1 address from workload api1 which should result in recomputing endpoints
rtest.ResourceID(api1.Id).
WithData(t, &pbcatalog.Workload{
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1"},
{Host: "172.17.1.1", Ports: []string{"mesh"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"mesh": {Port: 10000, Protocol: pbcatalog.Protocol_PROTOCOL_MESH},
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
Identity: "api",
}).
Write(t, c)
// Verify that api-1 had its addresses modified appropriately
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
TargetRef: api1.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"grpc"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
HealthStatus: pbcatalog.Health_HEALTH_PASSING,
Identity: "api",
},
},
})
// Add a failing health status to the api1 workload to force recomputation of endpoints
setHealthStatus(t, c, api1.Id, "api-failed", pbcatalog.Health_HEALTH_CRITICAL)
// Verify that api-1 within the endpoints has the expected health
verifyServiceEndpoints(t, c, endpointsID, &pbcatalog.ServiceEndpoints{
Endpoints: []*pbcatalog.Endpoint{
{
TargetRef: api1.Id,
Addresses: []*pbcatalog.WorkloadAddress{
{Host: "127.0.0.1", Ports: []string{"grpc"}},
},
Ports: map[string]*pbcatalog.WorkloadPort{
"grpc": {Port: 9090, Protocol: pbcatalog.Protocol_PROTOCOL_GRPC},
},
HealthStatus: pbcatalog.Health_HEALTH_CRITICAL,
Identity: "api",
},
},
})
// Move the service to being unmanaged. We should see the ServiceEndpoints being removed.
service = rtest.ResourceID(service.Id).
WithData(t, &pbcatalog.Service{
Ports: []*pbcatalog.ServicePort{{TargetPort: "grpc", Protocol: pbcatalog.Protocol_PROTOCOL_GRPC}},
}).
Write(t, c)
// Wait for the endpoints controller to inform us that the endpoints are not being managed
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionUnmanaged)
// Ensure that the managed endpoints were deleted
c.WaitForDeletion(t, endpointsID)
// Put the service back into managed mode.
service = rtest.ResourceID(service.Id).
WithData(t, &pbcatalog.Service{
Workloads: &pbcatalog.WorkloadSelector{Prefixes: []string{"api-"}},
Ports: []*pbcatalog.ServicePort{{TargetPort: "grpc", Protocol: pbcatalog.Protocol_PROTOCOL_GRPC}},
}).
Write(t, c)
// Wait for the service endpoints to be regenerated
c.WaitForStatusCondition(t, service.Id, catalog.EndpointsStatusKey, catalog.EndpointsStatusConditionManaged)
c.RequireResourceExists(t, endpointsID)
// Now delete the service and ensure that the endpoints eventually are deleted as well
c.MustDelete(t, service.Id)
c.WaitForDeletion(t, endpointsID)
}
func setHealthStatus(t *testing.T, client *rtest.Client, owner *pbresource.ID, name string, health pbcatalog.Health) *pbresource.Resource {
return rtest.Resource(catalog.HealthStatusV1Alpha1Type, name).
WithData(t, &pbcatalog.HealthStatus{
Type: "synthetic",
Status: health,
}).
WithOwner(owner).
Write(t, client)
}