mirror of https://github.com/grpc/grpc-go.git
xds/priority: start the init timer when a child switch to Connecting from non-transient-failure state (#5334)
This commit is contained in:
parent
462d867696
commit
db79903af9
|
|
@ -100,12 +100,6 @@ type priorityBalancer struct {
|
|||
childToPriority map[string]int
|
||||
// children is a map from child name to sub-balancers.
|
||||
children map[string]*childBalancer
|
||||
// The timer to give a priority some time to connect. And if the priority
|
||||
// doesn't go into Ready/Failure, the next priority will be started.
|
||||
//
|
||||
// One timer is enough because there can be at most one priority in init
|
||||
// state.
|
||||
priorityInitTimer *timerWrapper
|
||||
}
|
||||
|
||||
func (b *priorityBalancer) UpdateClientConnState(s balancer.ClientConnState) error {
|
||||
|
|
@ -198,27 +192,17 @@ func (b *priorityBalancer) Close() {
|
|||
// Clear states of the current child in use, so if there's a race in picker
|
||||
// update, it will be dropped.
|
||||
b.childInUse = ""
|
||||
b.stopPriorityInitTimer()
|
||||
// Stop the child policies, this is necessary to stop the init timers in the
|
||||
// children.
|
||||
for _, child := range b.children {
|
||||
child.stop()
|
||||
}
|
||||
}
|
||||
|
||||
func (b *priorityBalancer) ExitIdle() {
|
||||
b.bg.ExitIdle()
|
||||
}
|
||||
|
||||
// stopPriorityInitTimer stops the priorityInitTimer if it's not nil, and set it
|
||||
// to nil.
|
||||
//
|
||||
// Caller must hold b.mu.
|
||||
func (b *priorityBalancer) stopPriorityInitTimer() {
|
||||
timerW := b.priorityInitTimer
|
||||
if timerW == nil {
|
||||
return
|
||||
}
|
||||
b.priorityInitTimer = nil
|
||||
timerW.stopped = true
|
||||
timerW.timer.Stop()
|
||||
}
|
||||
|
||||
// UpdateState implements balancergroup.BalancerStateAggregator interface. The
|
||||
// balancer group sends new connectivity state and picker here.
|
||||
func (b *priorityBalancer) UpdateState(childName string, state balancer.State) {
|
||||
|
|
|
|||
|
|
@ -19,6 +19,8 @@
|
|||
package priority
|
||||
|
||||
import (
|
||||
"time"
|
||||
|
||||
"google.golang.org/grpc/balancer"
|
||||
"google.golang.org/grpc/balancer/base"
|
||||
"google.golang.org/grpc/connectivity"
|
||||
|
|
@ -36,7 +38,16 @@ type childBalancer struct {
|
|||
rState resolver.State
|
||||
|
||||
started bool
|
||||
// This is set when the child reports TransientFailure, and unset when it
|
||||
// reports Ready or Idle. It is used to decide whether the failover timer
|
||||
// should start when the child is transitioning into Connecting. The timer
|
||||
// will be restarted if the child has not reported TF more recently than it
|
||||
// reported Ready or Idle.
|
||||
reportedTF bool
|
||||
state balancer.State
|
||||
// The timer to give a priority some time to connect. And if the priority
|
||||
// doesn't go into Ready/Failure, the next priority will be started.
|
||||
initTimer *timerWrapper
|
||||
}
|
||||
|
||||
// newChildBalancer creates a child balancer place holder, but doesn't
|
||||
|
|
@ -79,6 +90,7 @@ func (cb *childBalancer) start() {
|
|||
}
|
||||
cb.started = true
|
||||
cb.parent.bg.Add(cb.name, cb.bb)
|
||||
cb.startInitTimer()
|
||||
}
|
||||
|
||||
// sendUpdate sends the addresses and config to the child balancer.
|
||||
|
|
@ -103,10 +115,46 @@ func (cb *childBalancer) stop() {
|
|||
if !cb.started {
|
||||
return
|
||||
}
|
||||
cb.stopInitTimer()
|
||||
cb.parent.bg.Remove(cb.name)
|
||||
cb.started = false
|
||||
cb.state = balancer.State{
|
||||
ConnectivityState: connectivity.Connecting,
|
||||
Picker: base.NewErrPicker(balancer.ErrNoSubConnAvailable),
|
||||
}
|
||||
// Clear child.reportedTF, so that if this child is started later, it will
|
||||
// be given time to connect.
|
||||
cb.reportedTF = false
|
||||
}
|
||||
|
||||
func (cb *childBalancer) startInitTimer() {
|
||||
if cb.initTimer != nil {
|
||||
return
|
||||
}
|
||||
// Need this local variable to capture timerW in the AfterFunc closure
|
||||
// to check the stopped boolean.
|
||||
timerW := &timerWrapper{}
|
||||
cb.initTimer = timerW
|
||||
timerW.timer = time.AfterFunc(DefaultPriorityInitTimeout, func() {
|
||||
cb.parent.mu.Lock()
|
||||
defer cb.parent.mu.Unlock()
|
||||
if timerW.stopped {
|
||||
return
|
||||
}
|
||||
cb.initTimer = nil
|
||||
// Re-sync the priority. This will switch to the next priority if
|
||||
// there's any. Note that it's important sync() is called after setting
|
||||
// initTimer to nil.
|
||||
cb.parent.syncPriority()
|
||||
})
|
||||
}
|
||||
|
||||
func (cb *childBalancer) stopInitTimer() {
|
||||
timerW := cb.initTimer
|
||||
if timerW == nil {
|
||||
return
|
||||
}
|
||||
cb.initTimer = nil
|
||||
timerW.stopped = true
|
||||
timerW.timer.Stop()
|
||||
}
|
||||
|
|
|
|||
|
|
@ -36,9 +36,10 @@ var (
|
|||
DefaultPriorityInitTimeout = 10 * time.Second
|
||||
)
|
||||
|
||||
// syncPriority handles priority after a config update. It makes sure the
|
||||
// balancer state (started or not) is in sync with the priorities (even in
|
||||
// tricky cases where a child is moved from a priority to another).
|
||||
// syncPriority handles priority after a config update or a child balancer
|
||||
// connectivity state update. It makes sure the balancer state (started or not)
|
||||
// is in sync with the priorities (even in tricky cases where a child is moved
|
||||
// from a priority to another).
|
||||
//
|
||||
// It's guaranteed that after this function returns:
|
||||
// - If some child is READY, it is childInUse, and all lower priorities are
|
||||
|
|
@ -53,8 +54,11 @@ var (
|
|||
// set parent ClientConn to TransientFailure
|
||||
// - Otherwise, Scan all children from p0, and check balancer stats:
|
||||
// - For any of the following cases:
|
||||
// - If balancer is not started (not built), this is either a new child
|
||||
// with high priority, or a new builder for an existing child.
|
||||
// - If balancer is not started (not built), this is either a new child with
|
||||
// high priority, or a new builder for an existing child.
|
||||
// - If balancer is Connecting and has non-nil initTimer (meaning it
|
||||
// transitioned from Ready or Idle to connecting, not from TF, so we
|
||||
// should give it init-time to connect).
|
||||
// - If balancer is READY
|
||||
// - If this is the lowest priority
|
||||
// - do the following:
|
||||
|
|
@ -69,9 +73,6 @@ func (b *priorityBalancer) syncPriority() {
|
|||
if len(b.priorities) == 0 {
|
||||
b.childInUse = ""
|
||||
b.priorityInUse = 0
|
||||
// Stop the init timer. This can happen if the only priority is removed
|
||||
// shortly after it's added.
|
||||
b.stopPriorityInitTimer()
|
||||
b.cc.UpdateState(balancer.State{
|
||||
ConnectivityState: connectivity.TransientFailure,
|
||||
Picker: base.NewErrPicker(ErrAllPrioritiesRemoved),
|
||||
|
|
@ -89,6 +90,7 @@ func (b *priorityBalancer) syncPriority() {
|
|||
if !child.started ||
|
||||
child.state.ConnectivityState == connectivity.Ready ||
|
||||
child.state.ConnectivityState == connectivity.Idle ||
|
||||
(child.state.ConnectivityState == connectivity.Connecting && child.initTimer != nil) ||
|
||||
p == len(b.priorities)-1 {
|
||||
if b.childInUse != "" && b.childInUse != child.name {
|
||||
// childInUse was set and is different from this child, will
|
||||
|
|
@ -123,8 +125,7 @@ func (b *priorityBalancer) stopSubBalancersLowerThanPriority(p int) {
|
|||
// - stop all child with lower priorities
|
||||
// - if childInUse is not this child
|
||||
// - set childInUse to this child
|
||||
// - stops init timer
|
||||
// - if this child is not started, start it, and start a init timer
|
||||
// - if this child is not started, start it
|
||||
//
|
||||
// Note that it does NOT send the current child state (picker) to the parent
|
||||
// ClientConn. The caller needs to send it if necessary.
|
||||
|
|
@ -156,33 +157,8 @@ func (b *priorityBalancer) switchToChild(child *childBalancer, priority int) {
|
|||
b.childInUse = child.name
|
||||
b.priorityInUse = priority
|
||||
|
||||
// Init timer is always for childInUse. Since we are switching to a
|
||||
// different child, we will stop the init timer no matter what. If this
|
||||
// child is not started, we will start the init timer later.
|
||||
b.stopPriorityInitTimer()
|
||||
|
||||
if !child.started {
|
||||
child.start()
|
||||
// Need this local variable to capture timerW in the AfterFunc closure
|
||||
// to check the stopped boolean.
|
||||
timerW := &timerWrapper{}
|
||||
b.priorityInitTimer = timerW
|
||||
timerW.timer = time.AfterFunc(DefaultPriorityInitTimeout, func() {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
if timerW.stopped {
|
||||
return
|
||||
}
|
||||
b.priorityInitTimer = nil
|
||||
// Switch to the next priority if there's any.
|
||||
if pNext := priority + 1; pNext < len(b.priorities) {
|
||||
nameNext := b.priorities[pNext]
|
||||
if childNext, ok := b.children[nameNext]; ok {
|
||||
b.switchToChild(childNext, pNext)
|
||||
childNext.sendUpdate()
|
||||
}
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -222,141 +198,57 @@ func (b *priorityBalancer) handleChildStateUpdate(childName string, s balancer.S
|
|||
b.logger.Warningf("priority: child balancer not found for child %v, priority %v", childName, priority)
|
||||
return
|
||||
}
|
||||
oldState := child.state.ConnectivityState
|
||||
oldChildState := child.state
|
||||
child.state = s
|
||||
|
||||
// We start/stop the init timer of this child based on the new connectivity
|
||||
// state. syncPriority() later will need the init timer (to check if it's
|
||||
// nil or not) to decide which child to switch to.
|
||||
switch s.ConnectivityState {
|
||||
case connectivity.Ready, connectivity.Idle:
|
||||
// Note that idle is also handled as if it's Ready. It will close the
|
||||
// lower priorities (which will be kept in a cache, not deleted), and
|
||||
// new picks will use the Idle picker.
|
||||
b.handlePriorityWithNewStateReady(child, priority)
|
||||
child.reportedTF = false
|
||||
child.stopInitTimer()
|
||||
case connectivity.TransientFailure:
|
||||
b.handlePriorityWithNewStateTransientFailure(child, priority)
|
||||
child.reportedTF = true
|
||||
child.stopInitTimer()
|
||||
case connectivity.Connecting:
|
||||
b.handlePriorityWithNewStateConnecting(child, priority, oldState)
|
||||
if !child.reportedTF {
|
||||
child.startInitTimer()
|
||||
}
|
||||
default:
|
||||
// New state is Shutdown, should never happen. Don't forward.
|
||||
}
|
||||
}
|
||||
|
||||
// handlePriorityWithNewStateReady handles state Ready from a higher or equal
|
||||
// priority.
|
||||
//
|
||||
// An update with state Ready:
|
||||
// - If it's from higher priority:
|
||||
// - Switch to this priority
|
||||
// - Forward the update
|
||||
// - If it's from priorityInUse:
|
||||
// - Forward only
|
||||
//
|
||||
// Caller must make sure priorityInUse is not higher than priority.
|
||||
//
|
||||
// Caller must hold mu.
|
||||
func (b *priorityBalancer) handlePriorityWithNewStateReady(child *childBalancer, priority int) {
|
||||
// If one priority higher or equal to priorityInUse goes Ready, stop the
|
||||
// init timer. If update is from higher than priorityInUse, priorityInUse
|
||||
// will be closed, and the init timer will become useless.
|
||||
b.stopPriorityInitTimer()
|
||||
|
||||
// priorityInUse is lower than this priority, switch to this.
|
||||
if b.priorityInUse > priority {
|
||||
b.logger.Infof("Switching priority from %v to %v, because latter became Ready", b.priorityInUse, priority)
|
||||
b.switchToChild(child, priority)
|
||||
}
|
||||
// Forward the update since it's READY.
|
||||
b.cc.UpdateState(child.state)
|
||||
}
|
||||
|
||||
// handlePriorityWithNewStateTransientFailure handles state TransientFailure
|
||||
// from a higher or equal priority.
|
||||
//
|
||||
// An update with state TransientFailure:
|
||||
// - If it's from a higher priority:
|
||||
// - Do not forward, and do nothing
|
||||
// - If it's from priorityInUse:
|
||||
// - If there's no lower:
|
||||
// - Forward and do nothing else
|
||||
// - If there's a lower priority:
|
||||
// - Switch to the lower
|
||||
// - Forward the lower child's state
|
||||
// - Do NOT forward this update
|
||||
//
|
||||
// Caller must make sure priorityInUse is not higher than priority.
|
||||
//
|
||||
// Caller must hold mu.
|
||||
func (b *priorityBalancer) handlePriorityWithNewStateTransientFailure(child *childBalancer, priority int) {
|
||||
// priorityInUse is lower than this priority, do nothing.
|
||||
if b.priorityInUse > priority {
|
||||
oldPriorityInUse := b.priorityInUse
|
||||
child.parent.syncPriority()
|
||||
// If child is switched by syncPriority(), it also sends the update from the
|
||||
// new child to overwrite the old picker used by the parent.
|
||||
//
|
||||
// But no update is sent if the child is not switches. That means if this
|
||||
// update is from childInUse, and this child is still childInUse after
|
||||
// syncing, the update being handled here is not sent to the parent. In that
|
||||
// case, we need to do an explicit check here to forward the update.
|
||||
if b.priorityInUse == oldPriorityInUse && b.priorityInUse == priority {
|
||||
// Special handling for Connecting. If child was not switched, and this
|
||||
// is a Connecting->Connecting transition, do not send the redundant
|
||||
// update, since all Connecting pickers are the same (they tell the RPCs
|
||||
// to repick).
|
||||
//
|
||||
// This can happen because the initial state of a child (before any
|
||||
// update is received) is Connecting. When the child is started, it's
|
||||
// picker is sent to the parent by syncPriority (to overwrite the old
|
||||
// picker if there's any). When it reports Connecting after being
|
||||
// started, it will send a Connecting update (handled here), causing a
|
||||
// Connecting->Connecting transition.
|
||||
if oldChildState.ConnectivityState == connectivity.Connecting && s.ConnectivityState == connectivity.Connecting {
|
||||
return
|
||||
}
|
||||
// priorityInUse sends a failure. Stop its init timer.
|
||||
b.stopPriorityInitTimer()
|
||||
priorityNext := priority + 1
|
||||
if priorityNext >= len(b.priorities) {
|
||||
// Forward this update.
|
||||
// Only forward this update if sync() didn't switch child, and this
|
||||
// child is in use.
|
||||
//
|
||||
// sync() forwards the update if the child was switched, so there's no
|
||||
// need to forward again.
|
||||
b.cc.UpdateState(child.state)
|
||||
return
|
||||
}
|
||||
b.logger.Infof("Switching priority from %v to %v, because former became TransientFailure", priority, priorityNext)
|
||||
nameNext := b.priorities[priorityNext]
|
||||
childNext := b.children[nameNext]
|
||||
b.switchToChild(childNext, priorityNext)
|
||||
b.cc.UpdateState(childNext.state)
|
||||
childNext.sendUpdate()
|
||||
}
|
||||
|
||||
// handlePriorityWithNewStateConnecting handles state Connecting from a higher
|
||||
// than or equal priority.
|
||||
//
|
||||
// An update with state Connecting:
|
||||
// - If it's from a higher priority
|
||||
// - Do nothing
|
||||
// - If it's from priorityInUse, the behavior depends on previous state.
|
||||
//
|
||||
// When new state is Connecting, the behavior depends on previous state. If the
|
||||
// previous state was Ready, this is a transition out from Ready to Connecting.
|
||||
// Assuming there are multiple backends in the same priority, this mean we are
|
||||
// in a bad situation and we should failover to the next priority (Side note:
|
||||
// the current connectivity state aggregating algorithm (e.g. round-robin) is
|
||||
// not handling this right, because if many backends all go from Ready to
|
||||
// Connecting, the overall situation is more like TransientFailure, not
|
||||
// Connecting).
|
||||
//
|
||||
// If the previous state was Idle, we don't do anything special with failure,
|
||||
// and simply forward the update. The init timer should be in process, will
|
||||
// handle failover if it timeouts. If the previous state was TransientFailure,
|
||||
// we do not forward, because the lower priority is in use.
|
||||
//
|
||||
// Caller must make sure priorityInUse is not higher than priority.
|
||||
//
|
||||
// Caller must hold mu.
|
||||
func (b *priorityBalancer) handlePriorityWithNewStateConnecting(child *childBalancer, priority int, oldState connectivity.State) {
|
||||
// priorityInUse is lower than this priority, do nothing.
|
||||
if b.priorityInUse > priority {
|
||||
return
|
||||
}
|
||||
|
||||
switch oldState {
|
||||
case connectivity.Ready:
|
||||
// Handling transition from Ready to Connecting, is same as handling
|
||||
// TransientFailure. There's no need to stop the init timer, because it
|
||||
// should have been stopped when state turned Ready.
|
||||
priorityNext := priority + 1
|
||||
if priorityNext >= len(b.priorities) {
|
||||
// Forward this update.
|
||||
b.cc.UpdateState(child.state)
|
||||
return
|
||||
}
|
||||
b.logger.Infof("Switching priority from %v to %v, because former became TransientFailure", priority, priorityNext)
|
||||
nameNext := b.priorities[priorityNext]
|
||||
childNext := b.children[nameNext]
|
||||
b.switchToChild(childNext, priorityNext)
|
||||
b.cc.UpdateState(childNext.state)
|
||||
childNext.sendUpdate()
|
||||
case connectivity.Idle:
|
||||
b.cc.UpdateState(child.state)
|
||||
default:
|
||||
// Old state is Connecting, TransientFailure or Shutdown. Don't forward.
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -414,9 +414,8 @@ func (s) TestPriority_HighPriorityToConnectingFromReady(t *testing.T) {
|
|||
t.Fatalf("want %v, got %v", want, err)
|
||||
}
|
||||
|
||||
// Turn 0 to Connecting, will start and use 1. Because 0 changing from Ready
|
||||
// to Connecting is a failure.
|
||||
pb.UpdateSubConnState(sc0, balancer.SubConnState{ConnectivityState: connectivity.Connecting})
|
||||
// Turn 0 to TransientFailure, will start and use 1.
|
||||
pb.UpdateSubConnState(sc0, balancer.SubConnState{ConnectivityState: connectivity.TransientFailure})
|
||||
|
||||
// Before 1 gets READY, picker should return NoSubConnAvailable, so RPCs
|
||||
// will retry.
|
||||
|
|
@ -741,6 +740,9 @@ func (s) TestPriority_InitTimeout(t *testing.T) {
|
|||
}
|
||||
sc1 := <-cc.NewSubConnCh
|
||||
|
||||
// After the init timer of p0, when switching to p1, a connecting picker
|
||||
// will be sent to the parent. Clear it here.
|
||||
<-cc.NewPickerCh
|
||||
pb.UpdateSubConnState(sc1, balancer.SubConnState{ConnectivityState: connectivity.Connecting})
|
||||
pb.UpdateSubConnState(sc1, balancer.SubConnState{ConnectivityState: connectivity.Ready})
|
||||
|
||||
|
|
@ -859,6 +861,7 @@ func (s) TestPriority_RemovesAllPriorities(t *testing.T) {
|
|||
// Don't send any update to p0, so to not override the old state of p0.
|
||||
// Later, connect to p1 and then remove p1. This will fallback to p0, and
|
||||
// will send p0's old picker if they are not correctly removed.
|
||||
<-cc.NewPickerCh // Clear the picker from old p0.
|
||||
|
||||
// p1 will be used after priority init timeout.
|
||||
addrs11 := <-cc.NewSubConnAddrsCh
|
||||
|
|
@ -1851,6 +1854,22 @@ func (s) TestPriority_HighPriorityInitIdle(t *testing.T) {
|
|||
t.Fatalf("pick returned %v, %v, want _, %v", pr, err, errsTestInitIdle[0])
|
||||
}
|
||||
|
||||
// The ClientConn state update triggers a priority switch, from p0 -> p0
|
||||
// (since p0 is still in use). Along with this the update, p0 also gets a
|
||||
// ClientConn state update, with the addresses, which didn't change in this
|
||||
// test (this update to the child is necessary in case the addresses are
|
||||
// different).
|
||||
//
|
||||
// The test child policy, initIdleBalancer, blindly calls NewSubConn with
|
||||
// all the addresses it receives, so this will trigger a NewSubConn with the
|
||||
// old p0 addresses. (Note that in a real balancer, like roundrobin, no new
|
||||
// SubConn will be created because the addresses didn't change).
|
||||
//
|
||||
// Clear those from the channel so the rest of the test can get the expected
|
||||
// behavior.
|
||||
<-cc.NewSubConnAddrsCh
|
||||
<-cc.NewSubConnCh
|
||||
|
||||
// Turn p0 down, to start p1.
|
||||
pb.UpdateSubConnState(sc0, balancer.SubConnState{ConnectivityState: connectivity.TransientFailure})
|
||||
// Before 1 gets READY, picker should return NoSubConnAvailable, so RPCs
|
||||
|
|
|
|||
Loading…
Reference in New Issue