You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
consul/test-integ/peering_commontopo/ac5_2_pq_failover_test.go

412 lines
15 KiB

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package peering
import (
"fmt"
"testing"
"time"
"github.com/hashicorp/consul/api"
"github.com/hashicorp/consul/sdk/testutil/retry"
"github.com/hashicorp/consul/testing/deployer/topology"
"github.com/stretchr/testify/require"
)
// 1. Setup: put health service instances in each of the 3 clusters and create the PQ in one of them
// 2. Execute the PQ: Validate that failover count == 0 and that the pq results come from the local cluster
// 3. Register a failing TTL health check with the agent managing the service instance in the local cluster
// 4. Execute the PQ: Validate that failover count == 1 and that the pq results come from the first failover target peer
// 5. Register a failing TTL health check with the agent managing the service instance in the first failover peer
// 6. Execute the PQ: Validate that failover count == 2 and that the pq results come from the second failover target
// 7. Delete failing health check from step 5
// 8. Repeat step 4
// 9. Delete failing health check from step 3
// 10. Repeat step 2
type ac5_2PQFailoverSuite struct {
clientSID topology.ID
serverSID topology.ID
nodeServer topology.NodeID
}
type nodeKey struct {
dc string
partition string
}
var ac5_2Context = make(map[nodeKey]ac5_2PQFailoverSuite)
func TestAC5PreparedQueryFailover(t *testing.T) {
ct := newCommonTopo(t, "dc2", true, true)
s := &ac5_2PQFailoverSuite{}
s.setup(t, ct)
ct.Launch(t)
s.test(t, ct)
}
func (s *ac5_2PQFailoverSuite) setup(t *testing.T, ct *commonTopo) {
s.setupDC(ct, ct.DC1, ct.DC2)
s.setupDC(ct, ct.DC2, ct.DC1)
s.setupDC3(ct, ct.DC3, ct.DC1, ct.DC2)
}
func (s *ac5_2PQFailoverSuite) setupDC(ct *commonTopo, clu, peerClu *topology.Cluster) {
// TODO: handle all partitions
partition := "default"
peer := LocalPeerName(peerClu, partition)
serverSID := topology.ID{
Name: "ac5-server-http",
Partition: partition,
}
clientSID := topology.ID{
Name: "ac5-client-http",
Partition: partition,
}
client := serviceExt{
Workload: NewFortioServiceWithDefaults(
clu.Datacenter,
clientSID,
func(s *topology.Workload) {
s.EnvoyAdminPort = 0
s.DisableServiceMesh = true
},
),
Config: &api.ServiceConfigEntry{
Kind: api.ServiceDefaults,
Name: clientSID.Name,
Partition: ConfigEntryPartition(clientSID.Partition),
Protocol: "http",
},
Exports: []api.ServiceConsumer{{Peer: peer}},
}
ct.AddServiceNode(clu, client)
server := serviceExt{
Workload: NewFortioServiceWithDefaults(
clu.Datacenter,
serverSID,
func(s *topology.Workload) {
s.EnvoyAdminPort = 0
s.DisableServiceMesh = true
},
),
Exports: []api.ServiceConsumer{{Peer: peer}},
}
serverNode := ct.AddServiceNode(clu, server)
ac5_2Context[nodeKey{clu.Datacenter, partition}] = ac5_2PQFailoverSuite{
clientSID: clientSID,
serverSID: serverSID,
nodeServer: serverNode.ID(),
}
}
func (s *ac5_2PQFailoverSuite) setupDC3(ct *commonTopo, clu, peer1, peer2 *topology.Cluster) {
var (
peers []string
partition = "default"
)
peers = append(peers, LocalPeerName(peer1, partition), LocalPeerName(peer2, partition))
serverSID := topology.ID{
Name: "ac5-server-http",
Partition: partition,
}
clientSID := topology.ID{
Name: "ac5-client-http",
Partition: partition,
}
// disable service mesh for client in DC3
client := serviceExt{
Workload: NewFortioServiceWithDefaults(
clu.Datacenter,
clientSID,
func(s *topology.Workload) {
s.EnvoyAdminPort = 0
s.DisableServiceMesh = true
},
),
Config: &api.ServiceConfigEntry{
Kind: api.ServiceDefaults,
Name: clientSID.Name,
Partition: ConfigEntryPartition(clientSID.Partition),
Protocol: "http",
},
Exports: func() []api.ServiceConsumer {
var consumers []api.ServiceConsumer
for _, peer := range peers {
consumers = append(consumers, api.ServiceConsumer{
Peer: peer,
})
}
return consumers
}(),
}
ct.AddServiceNode(clu, client)
server := serviceExt{
Workload: NewFortioServiceWithDefaults(
clu.Datacenter,
serverSID,
func(s *topology.Workload) {
s.EnvoyAdminPort = 0
s.DisableServiceMesh = true
},
),
Exports: func() []api.ServiceConsumer {
var consumers []api.ServiceConsumer
for _, peer := range peers {
consumers = append(consumers, api.ServiceConsumer{
Peer: peer,
})
}
return consumers
}(),
}
serverNode := ct.AddServiceNode(clu, server)
ac5_2Context[nodeKey{clu.Datacenter, partition}] = ac5_2PQFailoverSuite{
clientSID: clientSID,
serverSID: serverSID,
nodeServer: serverNode.ID(),
}
}
func (s *ac5_2PQFailoverSuite) createPreparedQuery(t *testing.T, ct *commonTopo, c *api.Client, serviceName, partition string) (*api.PreparedQueryDefinition, *api.PreparedQuery) {
var (
peers []string
err error
)
peers = append(peers, LocalPeerName(ct.DC2, partition), LocalPeerName(ct.DC3, partition))
def := &api.PreparedQueryDefinition{
Name: "ac5-prepared-query",
Service: api.ServiceQuery{
Service: serviceName,
Partition: ConfigEntryPartition(partition),
OnlyPassing: true,
Failover: api.QueryFailoverOptions{
Targets: func() []api.QueryFailoverTarget {
var queryFailoverTargets []api.QueryFailoverTarget
for _, peer := range peers {
queryFailoverTargets = append(queryFailoverTargets, api.QueryFailoverTarget{
Peer: peer,
})
}
return queryFailoverTargets
}(),
},
},
}
query := c.PreparedQuery()
def.ID, _, err = query.Create(def, nil)
require.NoError(t, err, "error creating prepared query in cluster")
return def, query
}
func (s *ac5_2PQFailoverSuite) test(t *testing.T, ct *commonTopo) {
partition := "default"
dc1 := ct.Sprawl.Topology().Clusters[ct.DC1.Name]
dc2 := ct.Sprawl.Topology().Clusters[ct.DC2.Name]
dc3 := ct.Sprawl.Topology().Clusters[ct.DC3.Name]
type testcase struct {
cluster *topology.Cluster
peer *topology.Cluster
targetCluster *topology.Cluster
}
tcs := []testcase{
{
cluster: dc1,
peer: dc2,
targetCluster: dc3,
},
}
for _, tc := range tcs {
client := ct.APIClientForCluster(t, tc.cluster)
t.Run(fmt.Sprintf("%#v", tc), func(t *testing.T) {
svc := ac5_2Context[nodeKey{tc.cluster.Name, partition}]
require.NotNil(t, svc.serverSID.Name, "expected service name to not be nil")
require.NotNil(t, svc.nodeServer, "expected node server to not be nil")
assertServiceHealth(t, client, svc.serverSID.Name, 1)
def, _ := s.createPreparedQuery(t, ct, client, svc.serverSID.Name, partition)
s.testPreparedQueryZeroFailover(t, client, def, tc.cluster)
s.testPreparedQuerySingleFailover(t, ct, client, def, tc.cluster, tc.peer, partition)
s.testPreparedQueryTwoFailovers(t, ct, client, def, tc.cluster, tc.peer, tc.targetCluster, partition)
// delete failing health check in peer cluster & validate single failover
s.testPQSingleFailover(t, ct, client, def, tc.cluster, tc.peer, partition)
// delete failing health check in cluster & validate zero failover
s.testPQZeroFailover(t, ct, client, def, tc.cluster, tc.peer, partition)
})
}
}
func (s *ac5_2PQFailoverSuite) testPreparedQueryZeroFailover(t *testing.T, cl *api.Client, def *api.PreparedQueryDefinition, cluster *topology.Cluster) {
t.Run(fmt.Sprintf("prepared query should not failover %s", cluster.Name), func(t *testing.T) {
// Validate prepared query exists in cluster
queryDef, _, err := cl.PreparedQuery().Get(def.ID, nil)
require.NoError(t, err)
require.Len(t, queryDef, 1, "expected 1 prepared query")
require.Equal(t, 2, len(queryDef[0].Service.Failover.Targets), "expected 2 prepared query failover targets to dc2 and dc3")
retry.RunWith(&retry.Timer{Timeout: 10 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
queryResult, _, err := cl.PreparedQuery().Execute(def.ID, nil)
require.NoError(r, err)
// expected outcome should show 0 failover
require.Equal(r, 0, queryResult.Failovers, "expected 0 prepared query failover")
require.Equal(r, cluster.Name, queryResult.Nodes[0].Node.Datacenter, "pq results should come from the local cluster")
})
})
}
func (s *ac5_2PQFailoverSuite) testPreparedQuerySingleFailover(t *testing.T, ct *commonTopo, cl *api.Client, def *api.PreparedQueryDefinition, cluster, peerClu *topology.Cluster, partition string) {
t.Run(fmt.Sprintf("prepared query with single failover %s", cluster.Name), func(t *testing.T) {
cfg := ct.Sprawl.Config()
svc := ac5_2Context[nodeKey{cluster.Name, partition}]
nodeCfg := DisableNode(t, cfg, cluster.Name, svc.nodeServer)
require.NoError(t, ct.Sprawl.Relaunch(nodeCfg))
// assert server health status
assertServiceHealth(t, cl, svc.serverSID.Name, 0)
// Validate prepared query exists in cluster
queryDef, _, err := cl.PreparedQuery().Get(def.ID, nil)
require.NoError(t, err)
require.Len(t, queryDef, 1, "expected 1 prepared query")
pqFailoverTargets := queryDef[0].Service.Failover.Targets
require.Len(t, pqFailoverTargets, 2, "expected 2 prepared query failover targets to dc2 and dc3")
retry.RunWith(&retry.Timer{Timeout: 10 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
queryResult, _, err := cl.PreparedQuery().Execute(def.ID, nil)
require.NoError(r, err)
require.Equal(r, 1, queryResult.Failovers, "expected 1 prepared query failover")
require.Equal(r, peerClu.Name, queryResult.Nodes[0].Node.Datacenter, fmt.Sprintf("the pq results should originate from peer clu %s", peerClu.Name))
require.Equal(r, pqFailoverTargets[0].Peer, queryResult.Nodes[0].Checks[0].PeerName,
fmt.Sprintf("pq results should come from the first failover target peer %s", pqFailoverTargets[0].Peer))
})
})
}
func (s *ac5_2PQFailoverSuite) testPreparedQueryTwoFailovers(t *testing.T, ct *commonTopo, cl *api.Client, def *api.PreparedQueryDefinition, cluster, peerClu, targetCluster *topology.Cluster, partition string) {
t.Run(fmt.Sprintf("prepared query with two failovers %s", cluster.Name), func(t *testing.T) {
cfg := ct.Sprawl.Config()
svc := ac5_2Context[nodeKey{peerClu.Name, partition}]
cfg = DisableNode(t, cfg, peerClu.Name, svc.nodeServer)
require.NoError(t, ct.Sprawl.Relaunch(cfg))
// assert server health status
assertServiceHealth(t, cl, ac5_2Context[nodeKey{cluster.Name, partition}].serverSID.Name, 0) // cluster: failing
assertServiceHealth(t, cl, svc.serverSID.Name, 0) // peer cluster: failing
queryDef, _, err := cl.PreparedQuery().Get(def.ID, nil)
require.NoError(t, err)
require.Len(t, queryDef, 1, "expected 1 prepared query")
pqFailoverTargets := queryDef[0].Service.Failover.Targets
require.Len(t, pqFailoverTargets, 2, "expected 2 prepared query failover targets to dc2 and dc3")
retry.RunWith(&retry.Timer{Timeout: 10 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
queryResult, _, err := cl.PreparedQuery().Execute(def.ID, nil)
require.NoError(r, err)
require.Equal(r, 2, queryResult.Failovers, "expected 2 prepared query failover")
require.Equal(r, targetCluster.Name, queryResult.Nodes[0].Node.Datacenter, fmt.Sprintf("the pq results should originate from cluster %s", targetCluster.Name))
require.Equal(r, pqFailoverTargets[1].Peer, queryResult.Nodes[0].Checks[0].PeerName,
fmt.Sprintf("pq results should come from the second failover target peer %s", pqFailoverTargets[1].Peer))
})
})
}
func (s *ac5_2PQFailoverSuite) testPQSingleFailover(t *testing.T, ct *commonTopo, cl *api.Client, def *api.PreparedQueryDefinition, cluster, peerClu *topology.Cluster, partition string) {
t.Run(fmt.Sprintf("delete failing health check in %s and validate single failover %s", peerClu.Name, cluster.Name), func(t *testing.T) {
cfg := ct.Sprawl.Config()
svc := ac5_2Context[nodeKey{peerClu.Name, partition}]
cfg = EnableNode(t, cfg, peerClu.Name, svc.nodeServer)
require.NoError(t, ct.Sprawl.Relaunch(cfg))
queryDef, _, err := cl.PreparedQuery().Get(def.ID, nil)
require.NoError(t, err)
pqFailoverTargets := queryDef[0].Service.Failover.Targets
require.Len(t, pqFailoverTargets, 2, "expected 2 prepared query failover targets to dc2 and dc3")
retry.RunWith(&retry.Timer{Timeout: 10 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
queryResult, _, err := cl.PreparedQuery().Execute(def.ID, nil)
require.NoError(r, err)
require.Equal(r, 1, queryResult.Failovers, "expected 1 prepared query failover")
require.Equal(r, peerClu.Name, queryResult.Nodes[0].Node.Datacenter, fmt.Sprintf("the pq results should originate from cluster %s", peerClu.Name))
require.Equal(r, pqFailoverTargets[0].Peer, queryResult.Nodes[0].Checks[0].PeerName,
fmt.Sprintf("pq results should come from the second failover target peer %s", pqFailoverTargets[0].Peer))
})
})
}
func (s *ac5_2PQFailoverSuite) testPQZeroFailover(t *testing.T, ct *commonTopo, cl *api.Client, def *api.PreparedQueryDefinition, cluster, _ *topology.Cluster, partition string) {
t.Run(fmt.Sprintf("delete failing health check in %s and validate zero failover %s", cluster.Name, cluster.Name), func(t *testing.T) {
cfg := ct.Sprawl.Config()
svc := ac5_2Context[nodeKey{cluster.Name, partition}]
cfg = EnableNode(t, cfg, cluster.Name, svc.nodeServer)
require.NoError(t, ct.Sprawl.Relaunch(cfg))
// assert server health status
assertServiceHealth(t, cl, ac5_2Context[nodeKey{cluster.Name, partition}].serverSID.Name, 1) // cluster: passing
assertServiceHealth(t, cl, svc.serverSID.Name, 1) // peer cluster: passing
queryDef, _, err := cl.PreparedQuery().Get(def.ID, nil)
require.NoError(t, err)
pqFailoverTargets := queryDef[0].Service.Failover.Targets
require.Len(t, pqFailoverTargets, 2, "expected 2 prepared query failover targets to dc2 and dc3")
retry.RunWith(&retry.Timer{Timeout: 10 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
queryResult, _, err := cl.PreparedQuery().Execute(def.ID, nil)
require.NoError(r, err)
// expected outcome should show 0 failover
require.Equal(r, 0, queryResult.Failovers, "expected 0 prepared query failover")
require.Equal(r, cluster.Name, queryResult.Nodes[0].Node.Datacenter, "pq results should come from the local cluster")
})
})
}
// assertServiceHealth checks that a service health status before running tests
func assertServiceHealth(t *testing.T, cl *api.Client, serverSVC string, count int) {
t.Helper()
t.Log("validate service health in catalog")
retry.RunWith(&retry.Timer{Timeout: time.Second * 20, Wait: time.Millisecond * 500}, t, func(r *retry.R) {
svcs, _, err := cl.Health().Service(
serverSVC,
"",
true,
nil,
)
require.NoError(r, err)
require.Equal(r, count, len(svcs))
})
}