moby/libnetwork/agent.go

977 lines
28 KiB
Go
Raw Normal View History

package libnetwork
//go:generate protoc -I=. -I=../vendor/ --gogofaster_out=import_path=github.com/docker/docker/libnetwork:. agent.proto
import (
"context"
"encoding/json"
"fmt"
"net"
"sort"
"sync"
"github.com/containerd/log"
"github.com/docker/docker/libnetwork/cluster"
"github.com/docker/docker/libnetwork/discoverapi"
"github.com/docker/docker/libnetwork/driverapi"
"github.com/docker/docker/libnetwork/networkdb"
"github.com/docker/docker/libnetwork/scope"
"github.com/docker/docker/libnetwork/types"
"github.com/docker/go-events"
"github.com/gogo/protobuf/proto"
)
const (
subsysGossip = "networking:gossip"
subsysIPSec = "networking:ipsec"
keyringSize = 3
)
// ByTime implements sort.Interface for []*types.EncryptionKey based on
// the LamportTime field.
type ByTime []*types.EncryptionKey
func (b ByTime) Len() int { return len(b) }
func (b ByTime) Swap(i, j int) { b[i], b[j] = b[j], b[i] }
func (b ByTime) Less(i, j int) bool { return b[i].LamportTime < b[j].LamportTime }
type nwAgent struct {
networkDB *networkdb.NetworkDB
bindAddr net.IP
advertiseAddr string
dataPathAddr string
coreCancelFuncs []func()
driverCancelFuncs map[string][]func()
mu sync.Mutex
}
func (a *nwAgent) dataPathAddress() string {
a.mu.Lock()
defer a.mu.Unlock()
if a.dataPathAddr != "" {
return a.dataPathAddr
}
return a.advertiseAddr
}
const libnetworkEPTable = "endpoint_table"
func getBindAddr(ifaceName string) (net.IP, error) {
iface, err := net.InterfaceByName(ifaceName)
if err != nil {
return nil, fmt.Errorf("failed to find interface %s: %v", ifaceName, err)
}
addrs, err := iface.Addrs()
if err != nil {
return nil, fmt.Errorf("failed to get interface addresses: %v", err)
}
for _, a := range addrs {
addr, ok := a.(*net.IPNet)
if !ok {
continue
}
addrIP := addr.IP
if addrIP.IsLinkLocalUnicast() {
continue
}
return addrIP, nil
}
return nil, fmt.Errorf("failed to get bind address")
}
// resolveAddr resolves the given address, which can be one of, and
// parsed in the following order or priority:
//
// - a well-formed IP-address
// - a hostname
// - an interface-name
func resolveAddr(addrOrInterface string) (net.IP, error) {
// Try and see if this is a valid IP address
if ip := net.ParseIP(addrOrInterface); ip != nil {
return ip, nil
}
// If not a valid IP address, it could be a hostname.
addr, err := net.ResolveIPAddr("ip", addrOrInterface)
if err != nil {
// If hostname lookup failed, try to look for an interface with the given name.
return getBindAddr(addrOrInterface)
}
return addr.IP, nil
}
func (c *Controller) handleKeyChange(keys []*types.EncryptionKey) error {
drvEnc := discoverapi.DriverEncryptionUpdate{}
agent := c.getAgent()
if agent == nil {
log.G(context.TODO()).Debug("Skipping key change as agent is nil")
return nil
}
// Find the deleted key. If the deleted key was the primary key,
// a new primary key should be set before removing if from keyring.
c.mu.Lock()
added := []byte{}
deleted := []byte{}
j := len(c.keys)
for i := 0; i < j; {
same := false
for _, key := range keys {
if same = key.LamportTime == c.keys[i].LamportTime; same {
break
}
}
if !same {
cKey := c.keys[i]
if cKey.Subsystem == subsysGossip {
deleted = cKey.Key
}
if cKey.Subsystem == subsysIPSec {
drvEnc.Prune = cKey.Key
drvEnc.PruneTag = cKey.LamportTime
}
c.keys[i], c.keys[j-1] = c.keys[j-1], c.keys[i]
c.keys[j-1] = nil
j--
}
i++
}
c.keys = c.keys[:j]
// Find the new key and add it to the key ring
for _, key := range keys {
same := false
for _, cKey := range c.keys {
if same = cKey.LamportTime == key.LamportTime; same {
break
}
}
if !same {
c.keys = append(c.keys, key)
if key.Subsystem == subsysGossip {
added = key.Key
}
if key.Subsystem == subsysIPSec {
drvEnc.Key = key.Key
drvEnc.Tag = key.LamportTime
}
}
}
c.mu.Unlock()
if len(added) > 0 {
agent.networkDB.SetKey(added)
}
key, _, err := c.getPrimaryKeyTag(subsysGossip)
if err != nil {
return err
}
agent.networkDB.SetPrimaryKey(key)
key, tag, err := c.getPrimaryKeyTag(subsysIPSec)
if err != nil {
return err
}
drvEnc.Primary = key
drvEnc.PrimaryTag = tag
if len(deleted) > 0 {
agent.networkDB.RemoveKey(deleted)
}
c.drvRegistry.WalkDrivers(func(name string, driver driverapi.Driver, capability driverapi.Capability) bool {
dr, ok := driver.(discoverapi.Discover)
if !ok {
return false
}
if err := dr.DiscoverNew(discoverapi.EncryptionKeysUpdate, drvEnc); err != nil {
log.G(context.TODO()).Warnf("Failed to update datapath keys in driver %s: %v", name, err)
// Attempt to reconfigure keys in case of a update failure
// which can arise due to a mismatch of keys
// if worker nodes get temporarily disconnected
log.G(context.TODO()).Warnf("Reconfiguring datapath keys for %s", name)
drvCfgEnc := discoverapi.DriverEncryptionConfig{}
drvCfgEnc.Keys, drvCfgEnc.Tags = c.getKeys(subsysIPSec)
err = dr.DiscoverNew(discoverapi.EncryptionKeysConfig, drvCfgEnc)
if err != nil {
log.G(context.TODO()).Warnf("Failed to reset datapath keys in driver %s: %v", name, err)
}
}
return false
})
return nil
}
func (c *Controller) agentSetup(clusterProvider cluster.Provider) error {
agent := c.getAgent()
if agent != nil {
// agent is already present, so there is no need initialize it again.
return nil
}
bindAddr := clusterProvider.GetLocalAddress()
advAddr := clusterProvider.GetAdvertiseAddress()
dataAddr := clusterProvider.GetDataPathAddress()
remoteList := clusterProvider.GetRemoteAddressList()
remoteAddrList := make([]string, 0, len(remoteList))
for _, remote := range remoteList {
addr, _, _ := net.SplitHostPort(remote)
remoteAddrList = append(remoteAddrList, addr)
}
listen := clusterProvider.GetListenAddress()
listenAddr, _, _ := net.SplitHostPort(listen)
log.G(context.TODO()).WithFields(log.Fields{
"listen-addr": listenAddr,
"local-addr": bindAddr,
"advertise-addr": advAddr,
"data-path-addr": dataAddr,
"remote-addr-list": remoteAddrList,
"network-control-plane-mtu": c.Config().NetworkControlPlaneMTU,
}).Info("Initializing Libnetwork Agent")
if advAddr != "" {
if err := c.agentInit(listenAddr, bindAddr, advAddr, dataAddr); err != nil {
log.G(context.TODO()).WithError(err).Errorf("Error in agentInit")
return err
}
c.drvRegistry.WalkDrivers(func(name string, driver driverapi.Driver, capability driverapi.Capability) bool {
if capability.ConnectivityScope == scope.Global {
if d, ok := driver.(discoverapi.Discover); ok {
c.agentDriverNotify(d)
}
}
return false
})
}
if len(remoteAddrList) > 0 {
if err := c.agentJoin(remoteAddrList); err != nil {
log.G(context.TODO()).WithError(err).Error("Error in joining gossip cluster: join will be retried in background")
}
}
return nil
}
// For a given subsystem getKeys sorts the keys by lamport time and returns
// slice of keys and lamport time which can used as a unique tag for the keys
func (c *Controller) getKeys(subsystem string) (keys [][]byte, tags []uint64) {
c.mu.Lock()
defer c.mu.Unlock()
sort.Sort(ByTime(c.keys))
keys = make([][]byte, 0, len(c.keys))
tags = make([]uint64, 0, len(c.keys))
for _, key := range c.keys {
if key.Subsystem == subsystem {
keys = append(keys, key.Key)
tags = append(tags, key.LamportTime)
}
}
if len(keys) > 1 {
// TODO(thaJeztah): why are we swapping order here? This code was added in https://github.com/moby/libnetwork/commit/e83d68b7d1fd9c479120914024242238f791b4dc
keys[0], keys[1] = keys[1], keys[0]
tags[0], tags[1] = tags[1], tags[0]
}
return keys, tags
}
// getPrimaryKeyTag returns the primary key for a given subsystem from the
// list of sorted key and the associated tag
func (c *Controller) getPrimaryKeyTag(subsystem string) (key []byte, lamportTime uint64, _ error) {
c.mu.Lock()
defer c.mu.Unlock()
sort.Sort(ByTime(c.keys))
keys := make([]*types.EncryptionKey, 0, len(c.keys))
for _, k := range c.keys {
if k.Subsystem == subsystem {
keys = append(keys, k)
}
}
if len(keys) < 2 {
return nil, 0, fmt.Errorf("no primary key found for %s subsystem: %d keys found on controller, expected at least 2", subsystem, len(keys))
}
return keys[1].Key, keys[1].LamportTime, nil
}
func (c *Controller) agentInit(listenAddr, bindAddrOrInterface, advertiseAddr, dataPathAddr string) error {
bindAddr, err := resolveAddr(bindAddrOrInterface)
if err != nil {
return err
}
keys, _ := c.getKeys(subsysGossip)
netDBConf := networkdb.DefaultConfig()
netDBConf.BindAddr = listenAddr
netDBConf.AdvertiseAddr = advertiseAddr
netDBConf.Keys = keys
if c.Config().NetworkControlPlaneMTU != 0 {
// Consider the MTU remove the IP hdr (IPv4 or IPv6) and the TCP/UDP hdr.
// To be on the safe side let's cut 100 bytes
netDBConf.PacketBufferSize = (c.Config().NetworkControlPlaneMTU - 100)
log.G(context.TODO()).Debugf("Control plane MTU: %d will initialize NetworkDB with: %d",
c.Config().NetworkControlPlaneMTU, netDBConf.PacketBufferSize)
}
nDB, err := networkdb.New(netDBConf)
if err != nil {
return err
}
// Register the diagnostic handlers
nDB.RegisterDiagnosticHandlers(c.DiagnosticServer)
var cancelList []func()
ch, cancel := nDB.Watch(libnetworkEPTable, "")
cancelList = append(cancelList, cancel)
nodeCh, cancel := nDB.Watch(networkdb.NodeTable, "")
cancelList = append(cancelList, cancel)
c.mu.Lock()
c.agent = &nwAgent{
networkDB: nDB,
bindAddr: bindAddr,
advertiseAddr: advertiseAddr,
dataPathAddr: dataPathAddr,
coreCancelFuncs: cancelList,
driverCancelFuncs: make(map[string][]func()),
}
c.mu.Unlock()
go c.handleTableEvents(ch, c.handleEpTableEvent)
go c.handleTableEvents(nodeCh, c.handleNodeTableEvent)
keys, tags := c.getKeys(subsysIPSec)
c.drvRegistry.WalkDrivers(func(name string, driver driverapi.Driver, capability driverapi.Capability) bool {
if dr, ok := driver.(discoverapi.Discover); ok {
if err := dr.DiscoverNew(discoverapi.EncryptionKeysConfig, discoverapi.DriverEncryptionConfig{
Keys: keys,
Tags: tags,
}); err != nil {
log.G(context.TODO()).Warnf("Failed to set datapath keys in driver %s: %v", name, err)
}
}
return false
})
c.WalkNetworks(joinCluster)
return nil
}
func (c *Controller) agentJoin(remoteAddrList []string) error {
agent := c.getAgent()
if agent == nil {
return nil
}
return agent.networkDB.Join(remoteAddrList)
}
func (c *Controller) agentDriverNotify(d discoverapi.Discover) {
agent := c.getAgent()
if agent == nil {
return
}
if err := d.DiscoverNew(discoverapi.NodeDiscovery, discoverapi.NodeDiscoveryData{
Address: agent.dataPathAddress(),
BindAddress: agent.bindAddr.String(),
Self: true,
}); err != nil {
log.G(context.TODO()).Warnf("Failed the node discovery in driver: %v", err)
}
keys, tags := c.getKeys(subsysIPSec)
if err := d.DiscoverNew(discoverapi.EncryptionKeysConfig, discoverapi.DriverEncryptionConfig{
Keys: keys,
Tags: tags,
}); err != nil {
log.G(context.TODO()).Warnf("Failed to set datapath keys in driver: %v", err)
}
}
func (c *Controller) agentClose() {
// Acquire current agent instance and reset its pointer
// then run closing functions
c.mu.Lock()
agent := c.agent
c.agent = nil
c.mu.Unlock()
// when the agent is closed the cluster provider should be cleaned up
c.SetClusterProvider(nil)
if agent == nil {
return
}
var cancelList []func()
agent.mu.Lock()
for _, cancelFuncs := range agent.driverCancelFuncs {
cancelList = append(cancelList, cancelFuncs...)
}
// Add also the cancel functions for the network db
cancelList = append(cancelList, agent.coreCancelFuncs...)
agent.mu.Unlock()
for _, cancel := range cancelList {
cancel()
}
agent.networkDB.Close()
}
// Task has the backend container details
type Task struct {
Name string
EndpointID string
EndpointIP string
Info map[string]string
}
// ServiceInfo has service specific details along with the list of backend tasks
type ServiceInfo struct {
VIP string
LocalLBIndex int
Tasks []Task
Ports []string
}
type epRecord struct {
ep EndpointRecord
info map[string]string
lbIndex int
}
// Services returns a map of services keyed by the service name with the details
// of all the tasks that belong to the service. Applicable only in swarm mode.
func (n *Network) Services() map[string]ServiceInfo {
agent, ok := n.clusterAgent()
if !ok {
return nil
}
nwID := n.ID()
d, err := n.driver(true)
if err != nil {
log.G(context.TODO()).Errorf("Could not resolve driver for network %s/%s while fetching services: %v", n.networkType, nwID, err)
return nil
}
// Walk through libnetworkEPTable and fetch the driver agnostic endpoint info
eps := make(map[string]epRecord)
c := n.getController()
for eid, value := range agent.networkDB.GetTableByNetwork(libnetworkEPTable, nwID) {
var epRec EndpointRecord
if err := proto.Unmarshal(value.Value, &epRec); err != nil {
log.G(context.TODO()).Errorf("Unmarshal of libnetworkEPTable failed for endpoint %s in network %s, %v", eid, nwID, err)
continue
}
eps[eid] = epRecord{
ep: epRec,
lbIndex: c.getLBIndex(epRec.ServiceID, nwID, epRec.IngressPorts),
}
}
// Walk through the driver's tables, have the driver decode the entries
// and return the tuple {ep ID, value}. value is a string that coveys
// relevant info about the endpoint.
for _, table := range n.driverTables {
if table.objType != driverapi.EndpointObject {
continue
}
for key, value := range agent.networkDB.GetTableByNetwork(table.name, nwID) {
epID, info := d.DecodeTableEntry(table.name, key, value.Value)
if ep, ok := eps[epID]; !ok {
log.G(context.TODO()).Errorf("Inconsistent driver and libnetwork state for endpoint %s", epID)
} else {
ep.info = info
eps[epID] = ep
}
}
}
// group the endpoints into a map keyed by the service name
sinfo := make(map[string]ServiceInfo)
for ep, epr := range eps {
s, ok := sinfo[epr.ep.ServiceName]
if !ok {
s = ServiceInfo{
VIP: epr.ep.VirtualIP,
LocalLBIndex: epr.lbIndex,
}
}
if s.Ports == nil {
ports := make([]string, 0, len(epr.ep.IngressPorts))
for _, port := range epr.ep.IngressPorts {
ports = append(ports, fmt.Sprintf("Target: %d, Publish: %d", port.TargetPort, port.PublishedPort))
}
s.Ports = ports
}
s.Tasks = append(s.Tasks, Task{
Name: epr.ep.Name,
EndpointID: ep,
EndpointIP: epr.ep.EndpointIP,
Info: epr.info,
})
sinfo[epr.ep.ServiceName] = s
}
return sinfo
}
// clusterAgent returns the cluster agent if the network is a swarm-scoped,
// multi-host network.
func (n *Network) clusterAgent() (agent *nwAgent, ok bool) {
if n.scope != scope.Swarm || !n.driverIsMultihost() {
return nil, false
}
a := n.getController().getAgent()
return a, a != nil
}
func (n *Network) joinCluster() error {
agent, ok := n.clusterAgent()
if !ok {
return nil
}
return agent.networkDB.JoinNetwork(n.ID())
}
func (n *Network) leaveCluster() error {
agent, ok := n.clusterAgent()
if !ok {
return nil
}
return agent.networkDB.LeaveNetwork(n.ID())
}
func (ep *Endpoint) addDriverInfoToCluster() error {
if ep.joinInfo == nil || len(ep.joinInfo.driverTableEntries) == 0 {
return nil
}
n := ep.getNetwork()
agent, ok := n.clusterAgent()
if !ok {
return nil
}
nwID := n.ID()
for _, te := range ep.joinInfo.driverTableEntries {
if err := agent.networkDB.CreateEntry(te.tableName, nwID, te.key, te.value); err != nil {
return err
}
}
return nil
}
func (ep *Endpoint) deleteDriverInfoFromCluster() error {
if ep.joinInfo == nil || len(ep.joinInfo.driverTableEntries) == 0 {
return nil
}
n := ep.getNetwork()
agent, ok := n.clusterAgent()
if !ok {
return nil
}
nwID := n.ID()
for _, te := range ep.joinInfo.driverTableEntries {
if err := agent.networkDB.DeleteEntry(te.tableName, nwID, te.key); err != nil {
return err
}
}
return nil
}
func (ep *Endpoint) addServiceInfoToCluster(sb *Sandbox) error {
if len(ep.dnsNames) == 0 || ep.Iface() == nil || ep.Iface().Address() == nil {
return nil
}
n := ep.getNetwork()
agent, ok := n.clusterAgent()
if !ok {
libnetwork: Endpoint: return early if no agent was found This removes redundant nil-checks in Endpoint.deleteServiceInfoFromCluster and Endpoint.addServiceInfoToCluster. These functions return early if the network is not ["cluster eligible"][1], and the function used for that (`Network.isClusterEligible`) requires the [agent to not be `nil`][2]. This check moved around a few times ([3][3], [4][4]), but was originally added in [libnetwork 1570][5] which, among others, tried to avoid a nil-pointer exception reported in [moby 28712][6], which accessed the `Controller.agent` [without locking][7]. That issue was addressed by adding locks, adding a `Controller.getAgent` accessor, and updating deleteServiceInfoFromCluster to use a local var. It also sprinkled this `nil` check to be on the safe side, but as `Network.isClusterEligible` already checks for the agent to not be `nil`, this should not be redundant. [1]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L529-L534 [2]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L688-L696 [3]: https://github.com/moby/libnetwork/commit/f2307265c77fd6db5d92310857c3ffad1184ad24 [4]: https://github.com/moby/libnetwork/commit/6426d1e66f33c0b0c8bb135b7ee547447f54d043 [5]: https://github.com/moby/libnetwork/commit/8dcf9960aa81767376e580f20d9ecea3e68f62ad [6]: https://github.com/moby/moby/issues/28712 [7]: https://github.com/moby/moby/blob/75fd88ba8938f120ce1f2a21b19cf13c12071b16/vendor/github.com/docker/libnetwork/agent.go#L452 Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2023-08-27 09:15:42 +00:00
return nil
}
sb.service.Lock()
defer sb.service.Unlock()
log.G(context.TODO()).Debugf("addServiceInfoToCluster START for %s %s", ep.svcName, ep.ID())
// Check that the endpoint is still present on the sandbox before adding it to the service discovery.
// This is to handle a race between the EnableService and the sbLeave
// It is possible that the EnableService starts, fetches the list of the endpoints and
// by the time the addServiceInfoToCluster is called the endpoint got removed from the sandbox
// The risk is that the deleteServiceInfoToCluster happens before the addServiceInfoToCluster.
// This check under the Service lock of the sandbox ensure the correct behavior.
// If the addServiceInfoToCluster arrives first may find or not the endpoint and will proceed or exit
// but in any case the deleteServiceInfoToCluster will follow doing the cleanup if needed.
// In case the deleteServiceInfoToCluster arrives first, this one is happening after the endpoint is
// removed from the list, in this situation the delete will bail out not finding any data to cleanup
// and the add will bail out not finding the endpoint on the sandbox.
if err := sb.GetEndpoint(ep.ID()); err == nil {
log.G(context.TODO()).Warnf("addServiceInfoToCluster suppressing service resolution ep is not anymore in the sandbox %s", ep.ID())
return nil
}
dnsNames := ep.getDNSNames()
primaryDNSName, dnsAliases := dnsNames[0], dnsNames[1:]
var ingressPorts []*PortConfig
if ep.svcID != "" {
// This is a task part of a service
// Gossip ingress ports only in ingress network.
if n.ingress {
ingressPorts = ep.ingressPorts
}
if err := n.getController().addServiceBinding(ep.svcName, ep.svcID, n.ID(), ep.ID(), primaryDNSName, ep.virtualIP, ingressPorts, ep.svcAliases, dnsAliases, ep.Iface().Address().IP, "addServiceInfoToCluster"); err != nil {
return err
}
} else {
// This is a container simply attached to an attachable network
if err := n.getController().addContainerNameResolution(n.ID(), ep.ID(), primaryDNSName, dnsAliases, ep.Iface().Address().IP, "addServiceInfoToCluster"); err != nil {
return err
}
}
buf, err := proto.Marshal(&EndpointRecord{
Name: primaryDNSName,
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
ServiceName: ep.svcName,
ServiceID: ep.svcID,
VirtualIP: ep.virtualIP.String(),
IngressPorts: ingressPorts,
Aliases: ep.svcAliases,
TaskAliases: dnsAliases,
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
EndpointIP: ep.Iface().Address().IP.String(),
ServiceDisabled: false,
})
if err != nil {
return err
}
libnetwork: Endpoint: return early if no agent was found This removes redundant nil-checks in Endpoint.deleteServiceInfoFromCluster and Endpoint.addServiceInfoToCluster. These functions return early if the network is not ["cluster eligible"][1], and the function used for that (`Network.isClusterEligible`) requires the [agent to not be `nil`][2]. This check moved around a few times ([3][3], [4][4]), but was originally added in [libnetwork 1570][5] which, among others, tried to avoid a nil-pointer exception reported in [moby 28712][6], which accessed the `Controller.agent` [without locking][7]. That issue was addressed by adding locks, adding a `Controller.getAgent` accessor, and updating deleteServiceInfoFromCluster to use a local var. It also sprinkled this `nil` check to be on the safe side, but as `Network.isClusterEligible` already checks for the agent to not be `nil`, this should not be redundant. [1]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L529-L534 [2]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L688-L696 [3]: https://github.com/moby/libnetwork/commit/f2307265c77fd6db5d92310857c3ffad1184ad24 [4]: https://github.com/moby/libnetwork/commit/6426d1e66f33c0b0c8bb135b7ee547447f54d043 [5]: https://github.com/moby/libnetwork/commit/8dcf9960aa81767376e580f20d9ecea3e68f62ad [6]: https://github.com/moby/moby/issues/28712 [7]: https://github.com/moby/moby/blob/75fd88ba8938f120ce1f2a21b19cf13c12071b16/vendor/github.com/docker/libnetwork/agent.go#L452 Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2023-08-27 09:15:42 +00:00
if err := agent.networkDB.CreateEntry(libnetworkEPTable, n.ID(), ep.ID(), buf); err != nil {
log.G(context.TODO()).Warnf("addServiceInfoToCluster NetworkDB CreateEntry failed for %s %s err:%s", ep.id, n.id, err)
return err
}
log.G(context.TODO()).Debugf("addServiceInfoToCluster END for %s %s", ep.svcName, ep.ID())
return nil
}
func (ep *Endpoint) deleteServiceInfoFromCluster(sb *Sandbox, fullRemove bool, method string) error {
if len(ep.dnsNames) == 0 {
return nil
}
n := ep.getNetwork()
agent, ok := n.clusterAgent()
if !ok {
libnetwork: Endpoint: return early if no agent was found This removes redundant nil-checks in Endpoint.deleteServiceInfoFromCluster and Endpoint.addServiceInfoToCluster. These functions return early if the network is not ["cluster eligible"][1], and the function used for that (`Network.isClusterEligible`) requires the [agent to not be `nil`][2]. This check moved around a few times ([3][3], [4][4]), but was originally added in [libnetwork 1570][5] which, among others, tried to avoid a nil-pointer exception reported in [moby 28712][6], which accessed the `Controller.agent` [without locking][7]. That issue was addressed by adding locks, adding a `Controller.getAgent` accessor, and updating deleteServiceInfoFromCluster to use a local var. It also sprinkled this `nil` check to be on the safe side, but as `Network.isClusterEligible` already checks for the agent to not be `nil`, this should not be redundant. [1]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L529-L534 [2]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L688-L696 [3]: https://github.com/moby/libnetwork/commit/f2307265c77fd6db5d92310857c3ffad1184ad24 [4]: https://github.com/moby/libnetwork/commit/6426d1e66f33c0b0c8bb135b7ee547447f54d043 [5]: https://github.com/moby/libnetwork/commit/8dcf9960aa81767376e580f20d9ecea3e68f62ad [6]: https://github.com/moby/moby/issues/28712 [7]: https://github.com/moby/moby/blob/75fd88ba8938f120ce1f2a21b19cf13c12071b16/vendor/github.com/docker/libnetwork/agent.go#L452 Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2023-08-27 09:15:42 +00:00
return nil
}
sb.service.Lock()
defer sb.service.Unlock()
log.G(context.TODO()).Debugf("deleteServiceInfoFromCluster from %s START for %s %s", method, ep.svcName, ep.ID())
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
// Avoid a race w/ with a container that aborts preemptively. This would
// get caught in disableServceInNetworkDB, but we check here to make the
// nature of the condition more clear.
// See comment in addServiceInfoToCluster()
if err := sb.GetEndpoint(ep.ID()); err == nil {
log.G(context.TODO()).Warnf("deleteServiceInfoFromCluster suppressing service resolution ep is not anymore in the sandbox %s", ep.ID())
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
return nil
}
dnsNames := ep.getDNSNames()
primaryDNSName, dnsAliases := dnsNames[0], dnsNames[1:]
libnetwork: Endpoint: return early if no agent was found This removes redundant nil-checks in Endpoint.deleteServiceInfoFromCluster and Endpoint.addServiceInfoToCluster. These functions return early if the network is not ["cluster eligible"][1], and the function used for that (`Network.isClusterEligible`) requires the [agent to not be `nil`][2]. This check moved around a few times ([3][3], [4][4]), but was originally added in [libnetwork 1570][5] which, among others, tried to avoid a nil-pointer exception reported in [moby 28712][6], which accessed the `Controller.agent` [without locking][7]. That issue was addressed by adding locks, adding a `Controller.getAgent` accessor, and updating deleteServiceInfoFromCluster to use a local var. It also sprinkled this `nil` check to be on the safe side, but as `Network.isClusterEligible` already checks for the agent to not be `nil`, this should not be redundant. [1]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L529-L534 [2]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L688-L696 [3]: https://github.com/moby/libnetwork/commit/f2307265c77fd6db5d92310857c3ffad1184ad24 [4]: https://github.com/moby/libnetwork/commit/6426d1e66f33c0b0c8bb135b7ee547447f54d043 [5]: https://github.com/moby/libnetwork/commit/8dcf9960aa81767376e580f20d9ecea3e68f62ad [6]: https://github.com/moby/moby/issues/28712 [7]: https://github.com/moby/moby/blob/75fd88ba8938f120ce1f2a21b19cf13c12071b16/vendor/github.com/docker/libnetwork/agent.go#L452 Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2023-08-27 09:15:42 +00:00
// First update the networkDB then locally
if fullRemove {
if err := agent.networkDB.DeleteEntry(libnetworkEPTable, n.ID(), ep.ID()); err != nil {
log.G(context.TODO()).Warnf("deleteServiceInfoFromCluster NetworkDB DeleteEntry failed for %s %s err:%s", ep.id, n.id, err)
}
libnetwork: Endpoint: return early if no agent was found This removes redundant nil-checks in Endpoint.deleteServiceInfoFromCluster and Endpoint.addServiceInfoToCluster. These functions return early if the network is not ["cluster eligible"][1], and the function used for that (`Network.isClusterEligible`) requires the [agent to not be `nil`][2]. This check moved around a few times ([3][3], [4][4]), but was originally added in [libnetwork 1570][5] which, among others, tried to avoid a nil-pointer exception reported in [moby 28712][6], which accessed the `Controller.agent` [without locking][7]. That issue was addressed by adding locks, adding a `Controller.getAgent` accessor, and updating deleteServiceInfoFromCluster to use a local var. It also sprinkled this `nil` check to be on the safe side, but as `Network.isClusterEligible` already checks for the agent to not be `nil`, this should not be redundant. [1]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L529-L534 [2]: https://github.com/moby/moby/blob/5b53ddfcdd1c0721f875d3e237078ab7de891a57/libnetwork/agent.go#L688-L696 [3]: https://github.com/moby/libnetwork/commit/f2307265c77fd6db5d92310857c3ffad1184ad24 [4]: https://github.com/moby/libnetwork/commit/6426d1e66f33c0b0c8bb135b7ee547447f54d043 [5]: https://github.com/moby/libnetwork/commit/8dcf9960aa81767376e580f20d9ecea3e68f62ad [6]: https://github.com/moby/moby/issues/28712 [7]: https://github.com/moby/moby/blob/75fd88ba8938f120ce1f2a21b19cf13c12071b16/vendor/github.com/docker/libnetwork/agent.go#L452 Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2023-08-27 09:15:42 +00:00
} else {
disableServiceInNetworkDB(agent, n, ep)
}
if ep.Iface() != nil && ep.Iface().Address() != nil {
if ep.svcID != "" {
// This is a task part of a service
var ingressPorts []*PortConfig
if n.ingress {
ingressPorts = ep.ingressPorts
}
if err := n.getController().rmServiceBinding(ep.svcName, ep.svcID, n.ID(), ep.ID(), primaryDNSName, ep.virtualIP, ingressPorts, ep.svcAliases, dnsAliases, ep.Iface().Address().IP, "deleteServiceInfoFromCluster", true, fullRemove); err != nil {
return err
}
} else {
// This is a container simply attached to an attachable network
if err := n.getController().delContainerNameResolution(n.ID(), ep.ID(), primaryDNSName, dnsAliases, ep.Iface().Address().IP, "deleteServiceInfoFromCluster"); err != nil {
return err
}
}
}
log.G(context.TODO()).Debugf("deleteServiceInfoFromCluster from %s END for %s %s", method, ep.svcName, ep.ID())
return nil
}
func disableServiceInNetworkDB(a *nwAgent, n *Network, ep *Endpoint) {
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
var epRec EndpointRecord
log.G(context.TODO()).Debugf("disableServiceInNetworkDB for %s %s", ep.svcName, ep.ID())
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
// Update existing record to indicate that the service is disabled
inBuf, err := a.networkDB.GetEntry(libnetworkEPTable, n.ID(), ep.ID())
if err != nil {
log.G(context.TODO()).Warnf("disableServiceInNetworkDB GetEntry failed for %s %s err:%s", ep.id, n.id, err)
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
return
}
// Should never fail
if err := proto.Unmarshal(inBuf, &epRec); err != nil {
log.G(context.TODO()).Errorf("disableServiceInNetworkDB unmarshal failed for %s %s err:%s", ep.id, n.id, err)
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
return
}
epRec.ServiceDisabled = true
// Should never fail
outBuf, err := proto.Marshal(&epRec)
if err != nil {
log.G(context.TODO()).Errorf("disableServiceInNetworkDB marshalling failed for %s %s err:%s", ep.id, n.id, err)
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
return
}
// Send update to the whole cluster
if err := a.networkDB.UpdateEntry(libnetworkEPTable, n.ID(), ep.ID(), outBuf); err != nil {
log.G(context.TODO()).Warnf("disableServiceInNetworkDB UpdateEntry failed for %s %s err:%s", ep.id, n.id, err)
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 (n *Network) addDriverWatches() {
if len(n.driverTables) == 0 {
return
}
agent, ok := n.clusterAgent()
if !ok {
return
}
c := n.getController()
for _, table := range n.driverTables {
ch, cancel := agent.networkDB.Watch(table.name, n.ID())
agent.mu.Lock()
agent.driverCancelFuncs[n.ID()] = append(agent.driverCancelFuncs[n.ID()], cancel)
agent.mu.Unlock()
go c.handleTableEvents(ch, n.handleDriverTableEvent)
d, err := n.driver(false)
if err != nil {
log.G(context.TODO()).Errorf("Could not resolve driver %s while walking driver tabl: %v", n.networkType, err)
return
}
err = agent.networkDB.WalkTable(table.name, func(nid, key string, value []byte, deleted bool) bool {
// skip the entries that are mark for deletion, this is safe because this function is
// called at initialization time so there is no state to delete
if nid == n.ID() && !deleted {
d.EventNotify(driverapi.Create, nid, table.name, key, value)
}
return false
})
if err != nil {
log.G(context.TODO()).WithError(err).Warn("Error while walking networkdb")
}
}
}
func (n *Network) cancelDriverWatches() {
agent, ok := n.clusterAgent()
if !ok {
return
}
agent.mu.Lock()
cancelFuncs := agent.driverCancelFuncs[n.ID()]
delete(agent.driverCancelFuncs, n.ID())
agent.mu.Unlock()
for _, cancel := range cancelFuncs {
cancel()
}
}
func (c *Controller) handleTableEvents(ch *events.Channel, fn func(events.Event)) {
for {
select {
case ev := <-ch.C:
fn(ev)
case <-ch.Done():
return
}
}
}
func (n *Network) handleDriverTableEvent(ev events.Event) {
d, err := n.driver(false)
if err != nil {
log.G(context.TODO()).Errorf("Could not resolve driver %s while handling driver table event: %v", n.networkType, err)
return
}
var (
etype driverapi.EventType
tname string
key string
value []byte
)
switch event := ev.(type) {
case networkdb.CreateEvent:
tname = event.Table
key = event.Key
value = event.Value
etype = driverapi.Create
case networkdb.DeleteEvent:
tname = event.Table
key = event.Key
value = event.Value
etype = driverapi.Delete
case networkdb.UpdateEvent:
tname = event.Table
key = event.Key
value = event.Value
etype = driverapi.Delete
}
d.EventNotify(etype, n.ID(), tname, key, value)
}
func (c *Controller) handleNodeTableEvent(ev events.Event) {
var (
value []byte
isAdd bool
nodeAddr networkdb.NodeAddr
)
switch event := ev.(type) {
case networkdb.CreateEvent:
value = event.Value
isAdd = true
case networkdb.DeleteEvent:
value = event.Value
case networkdb.UpdateEvent:
log.G(context.TODO()).Errorf("Unexpected update node table event = %#v", event)
}
err := json.Unmarshal(value, &nodeAddr)
if err != nil {
log.G(context.TODO()).Errorf("Error unmarshalling node table event %v", err)
return
}
c.processNodeDiscovery([]net.IP{nodeAddr.Addr}, isAdd)
}
func (c *Controller) handleEpTableEvent(ev events.Event) {
var (
nid string
eid string
value []byte
epRec EndpointRecord
)
switch event := ev.(type) {
case networkdb.CreateEvent:
nid = event.NetworkID
eid = event.Key
value = event.Value
case networkdb.DeleteEvent:
nid = event.NetworkID
eid = event.Key
value = event.Value
case networkdb.UpdateEvent:
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
nid = event.NetworkID
eid = event.Key
value = event.Value
default:
log.G(context.TODO()).Errorf("Unexpected update service table event = %#v", event)
return
}
err := proto.Unmarshal(value, &epRec)
if err != nil {
log.G(context.TODO()).Errorf("Failed to unmarshal service table value: %v", err)
return
}
containerName := epRec.Name
svcName := epRec.ServiceName
svcID := epRec.ServiceID
vip := net.ParseIP(epRec.VirtualIP)
ip := net.ParseIP(epRec.EndpointIP)
ingressPorts := epRec.IngressPorts
serviceAliases := epRec.Aliases
taskAliases := epRec.TaskAliases
if containerName == "" || ip == nil {
log.G(context.TODO()).Errorf("Invalid endpoint name/ip received while handling service table event %s", value)
return
}
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
switch ev.(type) {
case networkdb.CreateEvent:
log.G(context.TODO()).Debugf("handleEpTableEvent ADD %s R:%v", eid, epRec)
if svcID != "" {
// This is a remote task part of a service
if err := c.addServiceBinding(svcName, svcID, nid, eid, containerName, vip, ingressPorts, serviceAliases, taskAliases, ip, "handleEpTableEvent"); err != nil {
log.G(context.TODO()).Errorf("failed adding service binding for %s epRec:%v err:%v", eid, epRec, err)
return
}
} else {
// This is a remote container simply attached to an attachable network
if err := c.addContainerNameResolution(nid, eid, containerName, taskAliases, ip, "handleEpTableEvent"); err != nil {
log.G(context.TODO()).Errorf("failed adding container name resolution for %s epRec:%v err:%v", eid, epRec, err)
}
}
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
case networkdb.DeleteEvent:
log.G(context.TODO()).Debugf("handleEpTableEvent DEL %s R:%v", eid, epRec)
if svcID != "" {
// This is a remote task part of a service
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 err := c.rmServiceBinding(svcName, svcID, nid, eid, containerName, vip, ingressPorts, serviceAliases, taskAliases, ip, "handleEpTableEvent", true, true); err != nil {
log.G(context.TODO()).Errorf("failed removing service binding for %s epRec:%v err:%v", eid, epRec, err)
return
}
} else {
// This is a remote container simply attached to an attachable network
if err := c.delContainerNameResolution(nid, eid, containerName, taskAliases, ip, "handleEpTableEvent"); err != nil {
log.G(context.TODO()).Errorf("failed removing container name resolution for %s epRec:%v err:%v", eid, epRec, err)
}
}
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
case networkdb.UpdateEvent:
log.G(context.TODO()).Debugf("handleEpTableEvent UPD %s R:%v", eid, epRec)
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
// We currently should only get these to inform us that an endpoint
// is disabled. Report if otherwise.
if svcID == "" || !epRec.ServiceDisabled {
log.G(context.TODO()).Errorf("Unexpected update table event for %s epRec:%v", eid, epRec)
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
return
}
// This is a remote task that is part of a service that is now disabled
if err := c.rmServiceBinding(svcName, svcID, nid, eid, containerName, vip, ingressPorts, serviceAliases, taskAliases, ip, "handleEpTableEvent", true, false); err != nil {
log.G(context.TODO()).Errorf("failed disabling service binding for %s epRec:%v err:%v", eid, epRec, err)
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
return
}
}
}