// clusterSyncLoop observes running clusters changes, and apply all services to new added cluster
// and add dns records for the changes
func (s *ServiceController) clusterSyncLoop() {
var servicesToUpdate []*cachedService
// should we remove cache for cluster from ready to not ready? should remove the condition predicate if no
clusters, err := s.clusterStore.ClusterCondition(getClusterConditionPredicate()).List()
if err != nil {
glog.Infof("Fail to get cluster list")
return
}
newClusters := clustersFromList(&clusters)
var newSet, increase sets.String
newSet = sets.NewString(newClusters...)
if newSet.Equal(s.knownClusterSet) {
// The set of cluster names in the services in the federation hasn't changed, but we can retry
// updating any services that we failed to update last time around.
servicesToUpdate = s.updateDNSRecords(servicesToUpdate, newClusters)
return
}
glog.Infof("Detected change in list of cluster names. New set: %v, Old set: %v", newSet, s.knownClusterSet)
increase = newSet.Difference(s.knownClusterSet)
// do nothing when cluster is removed.
if increase != nil {
// Try updating all services, and save the ones that fail to try again next
// round.
servicesToUpdate = s.serviceCache.allServices()
numServices := len(servicesToUpdate)
for newCluster := range increase {
glog.Infof("New cluster observed %s", newCluster)
s.updateAllServicesToCluster(servicesToUpdate, newCluster)
}
servicesToUpdate = s.updateDNSRecords(servicesToUpdate, newClusters)
glog.Infof("Successfully updated %d out of %d DNS records to direct traffic to the updated cluster",
numServices-len(servicesToUpdate), numServices)
}
s.knownClusterSet = newSet
}
// enforcePodContainerConstraints checks for required resources that are not set on this container and
// adds them to missingSet.
func enforcePodContainerConstraints(container *api.Container, requiredSet, missingSet sets.String) {
requests := container.Resources.Requests
limits := container.Resources.Limits
containerUsage := podUsageHelper(requests, limits)
containerSet := quota.ToSet(quota.ResourceNames(containerUsage))
if !containerSet.Equal(requiredSet) {
difference := requiredSet.Difference(containerSet)
missingSet.Insert(difference.List()...)
}
}
// DialFromContainers executes a curl via kubectl exec in a test container,
// which might then translate to a tcp or udp request based on the protocol
// argument in the url.
// - minTries is the minimum number of curl attempts required before declaring
// success. Set to 0 if you'd like to return as soon as all endpoints respond
// at least once.
// - maxTries is the maximum number of curl attempts. If this many attempts pass
// and we don't see all expected endpoints, the test fails.
// - expectedEps is the set of endpointnames to wait for. Typically this is also
// the hostname reported by each pod in the service through /hostName.
// maxTries == minTries will confirm that we see the expected endpoints and no
// more for maxTries. Use this if you want to eg: fail a readiness check on a
// pod and confirm it doesn't show up as an endpoint.
func (config *NetworkingTestConfig) DialFromContainer(protocol, containerIP, targetIP string, containerHttpPort, targetPort, maxTries, minTries int, expectedEps sets.String) {
cmd := fmt.Sprintf("curl -q -s 'http://%s:%d/dial?request=hostName&protocol=%s&host=%s&port=%d&tries=1'",
containerIP,
containerHttpPort,
protocol,
targetIP,
targetPort)
eps := sets.NewString()
for i := 0; i < maxTries; i++ {
stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, cmd)
if err != nil {
// A failure to kubectl exec counts as a try, not a hard fail.
// Also note that we will keep failing for maxTries in tests where
// we confirm unreachability.
Logf("Failed to execute %q: %v, stdout: %q, stderr %q", cmd, err, stdout, stderr)
} else {
var output map[string][]string
if err := json.Unmarshal([]byte(stdout), &output); err != nil {
Logf("WARNING: Failed to unmarshal curl response. Cmd %v run in %v, output: %s, err: %v",
cmd, config.HostTestContainerPod.Name, stdout, err)
continue
}
for _, hostName := range output["responses"] {
trimmed := strings.TrimSpace(hostName)
if trimmed != "" {
eps.Insert(trimmed)
}
}
}
Logf("Waiting for endpoints: %v", expectedEps.Difference(eps))
// Check against i+1 so we exit if minTries == maxTries.
if (eps.Equal(expectedEps) || eps.Len() == 0 && expectedEps.Len() == 0) && i+1 >= minTries {
return
}
// TODO: get rid of this delay #36281
time.Sleep(hitEndpointRetryDelay)
}
config.diagnoseMissingEndpoints(eps)
Failf("Failed to find expected endpoints:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", minTries, cmd, eps, expectedEps)
}
// DialFromNode executes a tcp or udp request based on protocol via kubectl exec
// in a test container running with host networking.
// - minTries is the minimum number of curl attempts required before declaring
// success. Set to 0 if you'd like to return as soon as all endpoints respond
// at least once.
// - maxTries is the maximum number of curl attempts. If this many attempts pass
// and we don't see all expected endpoints, the test fails.
// maxTries == minTries will confirm that we see the expected endpoints and no
// more for maxTries. Use this if you want to eg: fail a readiness check on a
// pod and confirm it doesn't show up as an endpoint.
func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
var cmd string
if protocol == "udp" {
// TODO: It would be enough to pass 1s+epsilon to timeout, but unfortunately
// busybox timeout doesn't support non-integer values.
cmd = fmt.Sprintf("echo 'hostName' | timeout -t 2 nc -w 1 -u %s %d", targetIP, targetPort)
} else {
cmd = fmt.Sprintf("timeout -t 15 curl -q -s --connect-timeout 1 http://%s:%d/hostName", targetIP, targetPort)
}
// TODO: This simply tells us that we can reach the endpoints. Check that
// the probability of hitting a specific endpoint is roughly the same as
// hitting any other.
eps := sets.NewString()
filterCmd := fmt.Sprintf("%s | grep -v '^\\s*$'", cmd)
for i := 0; i < maxTries; i++ {
stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, filterCmd)
if err != nil || len(stderr) > 0 {
// A failure to exec command counts as a try, not a hard fail.
// Also note that we will keep failing for maxTries in tests where
// we confirm unreachability.
Logf("Failed to execute %q: %v, stdout: %q, stderr: %q", filterCmd, err, stdout, stderr)
} else {
trimmed := strings.TrimSpace(stdout)
if trimmed != "" {
eps.Insert(trimmed)
}
}
Logf("Waiting for %+v endpoints, got endpoints %+v", expectedEps.Difference(eps), eps)
// Check against i+1 so we exit if minTries == maxTries.
if (eps.Equal(expectedEps) || eps.Len() == 0 && expectedEps.Len() == 0) && i+1 >= minTries {
return
}
// TODO: get rid of this delay #36281
time.Sleep(hitEndpointRetryDelay)
}
config.diagnoseMissingEndpoints(eps)
Failf("Failed to find expected endpoints:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", minTries, cmd, eps, expectedEps)
}
func TestHitNodesFromOutsideWithCount(externalIP string, httpPort int32, timeout time.Duration, expectedHosts sets.String,
countToSucceed int) error {
Logf("Waiting up to %v for satisfying expectedHosts for %v times", timeout, countToSucceed)
hittedHosts := sets.NewString()
count := 0
condition := func() (bool, error) {
var respBody bytes.Buffer
reached, err := TestReachableHTTPWithContentTimeout(externalIP, int(httpPort), "/hostname", "", &respBody,
1*time.Second)
if err != nil || !reached {
return false, nil
}
hittedHost := strings.TrimSpace(respBody.String())
if !expectedHosts.Has(hittedHost) {
Logf("Error hitting unexpected host: %v, reset counter: %v", hittedHost, count)
count = 0
return false, nil
}
if !hittedHosts.Has(hittedHost) {
hittedHosts.Insert(hittedHost)
Logf("Missing %+v, got %+v", expectedHosts.Difference(hittedHosts), hittedHosts)
}
if hittedHosts.Equal(expectedHosts) {
count++
if count >= countToSucceed {
return true, nil
}
}
return false, nil
}
if err := wait.Poll(time.Second, timeout, condition); err != nil {
return fmt.Errorf("error waiting for expectedHosts: %v, hittedHosts: %v, count: %v, expected count: %v",
expectedHosts, hittedHosts, count, countToSucceed)
}
return nil
}
