moby--moby/libnetwork/service_common.go

414 lines
13 KiB
Go
Raw Normal View History

//go:build linux || windows
// +build linux windows
package libnetwork
import (
"net"
"github.com/docker/docker/libnetwork/internal/setmatrix"
"github.com/sirupsen/logrus"
)
const maxSetStringLen = 350
func (c *controller) addEndpointNameResolution(svcName, svcID, nID, eID, containerName string, vip net.IP, serviceAliases, taskAliases []string, ip net.IP, addService bool, method string) error {
n, err := c.NetworkByID(nID)
if err != nil {
return err
}
logrus.Debugf("addEndpointNameResolution %s %s add_service:%t sAliases:%v tAliases:%v", eID, svcName, addService, serviceAliases, taskAliases)
// Add container resolution mappings
if err := c.addContainerNameResolution(nID, eID, containerName, taskAliases, ip, method); err != nil {
return err
}
serviceID := svcID
if serviceID == "" {
// This is the case of a normal container not part of a service
serviceID = eID
}
// Add endpoint IP to special "tasks.svc_name" so that the applications have access to DNS RR.
n.(*network).addSvcRecords(eID, "tasks."+svcName, serviceID, ip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).addSvcRecords(eID, "tasks."+alias, serviceID, ip, nil, false, method)
}
// Add service name to vip in DNS, if vip is valid. Otherwise resort to DNS RR
if len(vip) == 0 {
n.(*network).addSvcRecords(eID, svcName, serviceID, ip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).addSvcRecords(eID, alias, serviceID, ip, nil, false, method)
}
}
if addService && len(vip) != 0 {
n.(*network).addSvcRecords(eID, svcName, serviceID, vip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).addSvcRecords(eID, alias, serviceID, vip, nil, false, method)
}
}
return nil
}
func (c *controller) addContainerNameResolution(nID, eID, containerName string, taskAliases []string, ip net.IP, method string) error {
n, err := c.NetworkByID(nID)
if err != nil {
return err
}
logrus.Debugf("addContainerNameResolution %s %s", eID, containerName)
// Add resolution for container name
n.(*network).addSvcRecords(eID, containerName, eID, ip, nil, true, method)
// Add resolution for taskaliases
for _, alias := range taskAliases {
n.(*network).addSvcRecords(eID, alias, eID, ip, nil, false, method)
}
return nil
}
func (c *controller) deleteEndpointNameResolution(svcName, svcID, nID, eID, containerName string, vip net.IP, serviceAliases, taskAliases []string, ip net.IP, rmService, multipleEntries bool, method string) error {
n, err := c.NetworkByID(nID)
if err != nil {
return err
}
logrus.Debugf("deleteEndpointNameResolution %s %s rm_service:%t suppress:%t sAliases:%v tAliases:%v", eID, svcName, rmService, multipleEntries, serviceAliases, taskAliases)
// Delete container resolution mappings
if err := c.delContainerNameResolution(nID, eID, containerName, taskAliases, ip, method); err != nil {
logrus.WithError(err).Warn("Error delting container from resolver")
}
serviceID := svcID
if serviceID == "" {
// This is the case of a normal container not part of a service
serviceID = eID
}
// Delete the special "tasks.svc_name" backend record.
if !multipleEntries {
n.(*network).deleteSvcRecords(eID, "tasks."+svcName, serviceID, ip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).deleteSvcRecords(eID, "tasks."+alias, serviceID, ip, nil, false, method)
}
}
// If we are doing DNS RR delete the endpoint IP from DNS record right away.
if !multipleEntries && len(vip) == 0 {
n.(*network).deleteSvcRecords(eID, svcName, serviceID, ip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).deleteSvcRecords(eID, alias, serviceID, ip, nil, false, method)
}
}
// Remove the DNS record for VIP only if we are removing the service
if rmService && len(vip) != 0 && !multipleEntries {
n.(*network).deleteSvcRecords(eID, svcName, serviceID, vip, nil, false, method)
for _, alias := range serviceAliases {
n.(*network).deleteSvcRecords(eID, alias, serviceID, vip, nil, false, method)
}
}
return nil
}
func (c *controller) delContainerNameResolution(nID, eID, containerName string, taskAliases []string, ip net.IP, method string) error {
n, err := c.NetworkByID(nID)
if err != nil {
return err
}
logrus.Debugf("delContainerNameResolution %s %s", eID, containerName)
// Delete resolution for container name
n.(*network).deleteSvcRecords(eID, containerName, eID, ip, nil, true, method)
// Delete resolution for taskaliases
for _, alias := range taskAliases {
n.(*network).deleteSvcRecords(eID, alias, eID, ip, nil, true, method)
}
return nil
}
func newService(name string, id string, ingressPorts []*PortConfig, serviceAliases []string) *service {
return &service{
name: name,
id: id,
ingressPorts: ingressPorts,
loadBalancers: make(map[string]*loadBalancer),
aliases: serviceAliases,
ipToEndpoint: setmatrix.NewSetMatrix(),
}
}
func (c *controller) getLBIndex(sid, nid string, ingressPorts []*PortConfig) int {
skey := serviceKey{
id: sid,
ports: portConfigs(ingressPorts).String(),
}
c.Lock()
s, ok := c.serviceBindings[skey]
c.Unlock()
if !ok {
return 0
}
s.Lock()
lb := s.loadBalancers[nid]
s.Unlock()
return int(lb.fwMark)
}
// cleanupServiceDiscovery when the network is being deleted, erase all the associated service discovery records
func (c *controller) cleanupServiceDiscovery(cleanupNID string) {
c.Lock()
defer c.Unlock()
if cleanupNID == "" {
logrus.Debugf("cleanupServiceDiscovery for all networks")
c.svcRecords = make(map[string]svcInfo)
return
}
logrus.Debugf("cleanupServiceDiscovery for network:%s", cleanupNID)
delete(c.svcRecords, cleanupNID)
}
func (c *controller) cleanupServiceBindings(cleanupNID string) {
var cleanupFuncs []func()
logrus.Debugf("cleanupServiceBindings for %s", cleanupNID)
c.Lock()
services := make([]*service, 0, len(c.serviceBindings))
for _, s := range c.serviceBindings {
services = append(services, s)
}
c.Unlock()
for _, s := range services {
s.Lock()
// Skip the serviceBindings that got deleted
if s.deleted {
s.Unlock()
continue
}
for nid, lb := range s.loadBalancers {
if cleanupNID != "" && nid != cleanupNID {
continue
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
for eid, be := range lb.backEnds {
cleanupFuncs = append(cleanupFuncs, makeServiceCleanupFunc(c, s, nid, eid, lb.vip, be.ip))
}
}
s.Unlock()
}
for _, f := range cleanupFuncs {
f()
}
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
func makeServiceCleanupFunc(c *controller, s *service, nID, eID string, vip net.IP, ip net.IP) func() {
// ContainerName and taskAliases are not available here, this is still fine because the Service discovery
// cleanup already happened before. The only thing that rmServiceBinding is still doing here a part from the Load
// Balancer bookeeping, is to keep consistent the mapping of endpoint to IP.
return func() {
if err := c.rmServiceBinding(s.name, s.id, nID, eID, "", vip, s.ingressPorts, s.aliases, []string{}, ip, "cleanupServiceBindings", false, true); err != nil {
logrus.Errorf("Failed to remove service bindings for service %s network %s endpoint %s while cleanup: %v", s.id, nID, eID, err)
}
}
}
func (c *controller) addServiceBinding(svcName, svcID, nID, eID, containerName string, vip net.IP, ingressPorts []*PortConfig, serviceAliases, taskAliases []string, ip net.IP, method string) error {
var addService bool
// Failure to lock the network ID on add can result in racing
// racing against network deletion resulting in inconsistent
// state in the c.serviceBindings map and it's sub-maps. Also,
// always lock network ID before services to avoid deadlock.
c.networkLocker.Lock(nID)
defer c.networkLocker.Unlock(nID) //nolint:errcheck
n, err := c.NetworkByID(nID)
if err != nil {
return err
}
skey := serviceKey{
id: svcID,
ports: portConfigs(ingressPorts).String(),
}
var s *service
for {
c.Lock()
var ok bool
s, ok = c.serviceBindings[skey]
if !ok {
// Create a new service if we are seeing this service
// for the first time.
s = newService(svcName, svcID, ingressPorts, serviceAliases)
c.serviceBindings[skey] = s
}
c.Unlock()
s.Lock()
if !s.deleted {
// ok the object is good to be used
break
}
s.Unlock()
}
logrus.Debugf("addServiceBinding from %s START for %s %s p:%p nid:%s skey:%v", method, svcName, eID, s, nID, skey)
defer s.Unlock()
lb, ok := s.loadBalancers[nID]
if !ok {
// Create a new load balancer if we are seeing this
// network attachment on the service for the first
// time.
fwMarkCtrMu.Lock()
lb = &loadBalancer{
vip: vip,
fwMark: fwMarkCtr,
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
backEnds: make(map[string]*lbBackend),
service: s,
}
fwMarkCtr++
fwMarkCtrMu.Unlock()
s.loadBalancers[nID] = lb
addService = true
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
lb.backEnds[eID] = &lbBackend{ip, false}
ok, entries := s.assignIPToEndpoint(ip.String(), eID)
if !ok || entries > 1 {
setStr, b := s.printIPToEndpoint(ip.String())
if len(setStr) > maxSetStringLen {
setStr = setStr[:maxSetStringLen]
}
logrus.Warnf("addServiceBinding %s possible transient state ok:%t entries:%d set:%t %s", eID, ok, entries, b, setStr)
}
Add endpoint load-balancing mode This is the heart of the scalability change for services in libnetwork. The present routing mesh adds load-balancing rules for a network to every container connected to the network. This newer approach creates a load-balancing endpoint per network per node. For every service on a network, libnetwork assigns the VIP of the service to the endpoint's interface as an alias. This endpoint must have a unique IP address in order to route return traffic to it. Traffic destined for a service's VIP arrives at the load-balancing endpoint on the VIP and from there, Linux load balances it among backend destinations while SNATing said traffic to the endpoint's unique IP address. The net result of this scheme is that each node in a swarm need only have one set of load balancing state per service instead of one per container on the node. This scheme is very similar to how services currently operate on Windows nodes in libnetwork. It (as with Windows nodes) costs the use of extra IP addresses in a network (one per node) and an extra network hop in the stack, although, always in the stack local to the container. In order to prevent existing deployments from suddenly failing if they failed to allocate sufficient address space to include per-node load-balancing endpoint IP addresses, this patch preserves the existing functionality and activates the new functionality on a per-network basis depending on whether the network has a load-balancing endpoint. Eventually, moby should always set this option when creating new networks and should only omit it for networks created as part of a swarm that are not marked to use endpoint load balancing. This patch also normalizes the code to treat "load" and "balancer" as two separate words from the perspectives of variable/function naming. This means that the 'b' in "balancer" must be capitalized. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-04-10 16:34:41 +00:00
// Add loadbalancer service and backend to the network
n.(*network).addLBBackend(ip, lb)
// Add the appropriate name resolutions
if err := c.addEndpointNameResolution(svcName, svcID, nID, eID, containerName, vip, serviceAliases, taskAliases, ip, addService, "addServiceBinding"); err != nil {
return err
}
logrus.Debugf("addServiceBinding from %s END for %s %s", method, svcName, eID)
return nil
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
func (c *controller) rmServiceBinding(svcName, svcID, nID, eID, containerName string, vip net.IP, ingressPorts []*PortConfig, serviceAliases []string, taskAliases []string, ip net.IP, method string, deleteSvcRecords bool, fullRemove bool) error {
var rmService bool
skey := serviceKey{
id: svcID,
ports: portConfigs(ingressPorts).String(),
}
c.Lock()
s, ok := c.serviceBindings[skey]
c.Unlock()
if !ok {
logrus.Warnf("rmServiceBinding %s %s %s aborted c.serviceBindings[skey] !ok", method, svcName, eID)
return nil
}
s.Lock()
defer s.Unlock()
logrus.Debugf("rmServiceBinding from %s START for %s %s p:%p nid:%s sKey:%v deleteSvc:%t", method, svcName, eID, s, nID, skey, deleteSvcRecords)
lb, ok := s.loadBalancers[nID]
if !ok {
logrus.Warnf("rmServiceBinding %s %s %s aborted s.loadBalancers[nid] !ok", method, svcName, eID)
return nil
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
be, ok := lb.backEnds[eID]
if !ok {
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
logrus.Warnf("rmServiceBinding %s %s %s aborted lb.backEnds[eid] && lb.disabled[eid] !ok", method, svcName, eID)
return nil
}
Gracefully remove LB endpoints from services This patch attempts to allow endpoints to complete servicing connections while being removed from a service. The change adds a flag to the endpoint.deleteServiceInfoFromCluster() method to indicate whether this removal should fully remove connectivity through the load balancer to the endpoint or should just disable directing further connections to the endpoint. If the flag is 'false', then the load balancer assigns a weight of 0 to the endpoint but does not remove it as a linux load balancing destination. It does remove the endpoint as a docker load balancing endpoint but tracks it in a special map of "disabled-but-not- destroyed" load balancing endpoints. This allows traffic to continue flowing, at least under Linux. If the flag is 'true', then the code removes the endpoint entirely as a load balancing destination. The sandbox.DisableService() method invokes deleteServiceInfoFromCluster() with the flag sent to 'false', while the endpoint.sbLeave() method invokes it with the flag set to 'true' to complete the removal on endpoint finalization. Renaming the endpoint invokes deleteServiceInfoFromCluster() with the flag set to 'true' because renaming attempts to completely remove and then re-add each endpoint service entry. The controller.rmServiceBinding() method, which carries out the operation, similarly gets a new flag for whether to fully remove the endpoint. If the flag is false, it does the job of moving the endpoint from the load balancing set to the 'disabled' set. It then removes or de-weights the entry in the OS load balancing table via network.rmLBBackend(). It removes the service entirely via said method ONLY IF there are no more live or disabled load balancing endpoints. Similarly network.addLBBackend() requires slight tweaking to properly manage the disabled set. Finally, this change requires propagating the status of disabled service endpoints via the networkDB. Accordingly, the patch includes both code to generate and handle service update messages. It also augments the service structure with a ServiceDisabled boolean to convey whether an endpoint should ultimately be removed or just disabled. This, naturally, required a rebuild of the protocol buffer code as well. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-02-14 22:04:23 +00:00
if fullRemove {
// delete regardless
delete(lb.backEnds, eID)
} else {
be.disabled = true
}
if len(lb.backEnds) == 0 {
// All the backends for this service have been
// removed. Time to remove the load balancer and also
// remove the service entry in IPVS.
rmService = true
delete(s.loadBalancers, nID)
logrus.Debugf("rmServiceBinding %s delete %s, p:%p in loadbalancers len:%d", eID, nID, lb, len(s.loadBalancers))
}
ok, entries := s.removeIPToEndpoint(ip.String(), eID)
if !ok || entries > 0 {
setStr, b := s.printIPToEndpoint(ip.String())
if len(setStr) > maxSetStringLen {
setStr = setStr[:maxSetStringLen]
}
logrus.Warnf("rmServiceBinding %s possible transient state ok:%t entries:%d set:%t %s", eID, ok, entries, b, setStr)
}
// Remove loadbalancer service(if needed) and backend in all
// sandboxes in the network only if the vip is valid.
if entries == 0 {
// The network may well have been deleted from the store (and
// dataplane) before the last of the service bindings. On Linux that's
// ok because removing the network sandbox from the dataplane
// implicitly cleans up all related dataplane state.
// On the Windows dataplane, VFP policylists must be removed
// independently of the network, and they must be removed before the HNS
// network. Otherwise, policylist removal fails with "network not
// found." On Windows cleanupServiceBindings must be called prior to
// removing the network from the store or dataplane.
Add endpoint load-balancing mode This is the heart of the scalability change for services in libnetwork. The present routing mesh adds load-balancing rules for a network to every container connected to the network. This newer approach creates a load-balancing endpoint per network per node. For every service on a network, libnetwork assigns the VIP of the service to the endpoint's interface as an alias. This endpoint must have a unique IP address in order to route return traffic to it. Traffic destined for a service's VIP arrives at the load-balancing endpoint on the VIP and from there, Linux load balances it among backend destinations while SNATing said traffic to the endpoint's unique IP address. The net result of this scheme is that each node in a swarm need only have one set of load balancing state per service instead of one per container on the node. This scheme is very similar to how services currently operate on Windows nodes in libnetwork. It (as with Windows nodes) costs the use of extra IP addresses in a network (one per node) and an extra network hop in the stack, although, always in the stack local to the container. In order to prevent existing deployments from suddenly failing if they failed to allocate sufficient address space to include per-node load-balancing endpoint IP addresses, this patch preserves the existing functionality and activates the new functionality on a per-network basis depending on whether the network has a load-balancing endpoint. Eventually, moby should always set this option when creating new networks and should only omit it for networks created as part of a swarm that are not marked to use endpoint load balancing. This patch also normalizes the code to treat "load" and "balancer" as two separate words from the perspectives of variable/function naming. This means that the 'b' in "balancer" must be capitalized. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-04-10 16:34:41 +00:00
n, err := c.NetworkByID(nID)
if err == nil {
n.(*network).rmLBBackend(ip, lb, rmService, fullRemove)
}
}
// Delete the name resolutions
if deleteSvcRecords {
if err := c.deleteEndpointNameResolution(svcName, svcID, nID, eID, containerName, vip, serviceAliases, taskAliases, ip, rmService, entries > 0, "rmServiceBinding"); err != nil {
return err
}
}
if len(s.loadBalancers) == 0 {
// All loadbalancers for the service removed. Time to
// remove the service itself.
c.Lock()
// Mark the object as deleted so that the add won't use it wrongly
s.deleted = true
// NOTE The delete from the serviceBindings map has to be the last operation else we are allowing a race between this service
// that is getting deleted and a new service that will be created if the entry is not anymore there
delete(c.serviceBindings, skey)
c.Unlock()
}
logrus.Debugf("rmServiceBinding from %s END for %s %s", method, svcName, eID)
return nil
}