Files
coredns/plugin/traffic/xds/v2client.go
Miek Gieben 519ef9ca79 more
Signed-off-by: Miek Gieben <miek@miek.nl>
2020-01-13 11:21:20 +01:00

432 lines
14 KiB
Go

/*
*
* Copyright 2019 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package xds
import (
"context"
"fmt"
"sync"
"time"
"github.com/coredns/coredns/plugin/traffic/xds/buffer"
xdspb "github.com/envoyproxy/go-control-plane/envoy/api/v2"
corepb "github.com/envoyproxy/go-control-plane/envoy/api/v2/core"
adsgrpc "github.com/envoyproxy/go-control-plane/envoy/service/discovery/v2"
"google.golang.org/grpc"
)
// The value chosen here is based on the default value of the
// initial_fetch_timeout field in corepb.ConfigSource proto.
var defaultWatchExpiryTimeout = 15 * time.Second
// v2Client performs the actual xDS RPCs using the xDS v2 API. It creates a
// single ADS stream on which the different types of xDS requests and responses
// are multiplexed.
// The reason for splitting this out from the top level xdsClient object is
// because there is already an xDS v3Aplha API in development. If and when we
// want to switch to that, this separation will ease that process.
type v2Client struct {
ctx context.Context
cancelCtx context.CancelFunc
// ClientConn to the xDS gRPC server. Owned by the parent xdsClient.
cc *grpc.ClientConn
nodeProto *corepb.Node
backoff func(int) time.Duration
// sendCh in the channel onto which watchInfo objects are pushed by the
// watch API, and it is read and acted upon by the send() goroutine.
sendCh *buffer.Unbounded
mu sync.Mutex
// Message specific watch infos, protected by the above mutex. These are
// written to, after successfully reading from the update channel, and are
// read from when recovering from a broken stream to resend the xDS
// messages. When the user of this client object cancels a watch call,
// these are set to nil. All accesses to the map protected and any value
// inside the map should be protected with the above mutex.
watchMap map[string]*watchInfo
// ackMap contains the version that was acked (the version in the ack
// request that was sent on wire). The key is typeURL, the value is the
// version string, becaues the versions for different resource types
// should be independent.
ackMap map[string]string
// rdsCache maintains a mapping of {clusterName --> CDSUpdate} from
// validated cluster configurations received in CDS responses. We cache all
// valid cluster configurations, whether or not we are interested in them
// when we received them (because we could become interested in them in the
// future and the server wont send us those resources again). This is only
// to support legacy management servers that do not honor the
// resource_names field. As per the latest spec, the server should resend
// the response when the request changes, even if it had sent the same
// resource earlier (when not asked for). Protected by the above mutex.
cdsCache map[string]CDSUpdate
}
// newV2Client creates a new v2Client initialized with the passed arguments.
func newV2Client(cc *grpc.ClientConn, nodeProto *corepb.Node, backoff func(int) time.Duration) *v2Client {
v2c := &v2Client{
cc: cc,
nodeProto: nodeProto,
backoff: backoff,
sendCh: buffer.NewUnbounded(),
watchMap: make(map[string]*watchInfo),
ackMap: make(map[string]string),
cdsCache: make(map[string]CDSUpdate),
}
v2c.ctx, v2c.cancelCtx = context.WithCancel(context.Background())
go v2c.run()
return v2c
}
// close cleans up resources and goroutines allocated by this client.
func (v2c *v2Client) close() {
v2c.cancelCtx()
}
// run starts an ADS stream (and backs off exponentially, if the previous
// stream failed without receiving a single reply) and runs the sender and
// receiver routines to send and receive data from the stream respectively.
func (v2c *v2Client) run() {
retries := 0
for {
select {
case <-v2c.ctx.Done():
return
default:
}
if retries != 0 {
t := time.NewTimer(v2c.backoff(retries))
select {
case <-t.C:
case <-v2c.ctx.Done():
if !t.Stop() {
<-t.C
}
return
}
}
retries++
println("SENDING STUFF, retries", retries)
cli := adsgrpc.NewAggregatedDiscoveryServiceClient(v2c.cc)
stream, err := cli.StreamAggregatedResources(v2c.ctx) //, grpc.WaitForReady(true))
if err != nil {
log.Infof("xds: ADS stream creation failed: %v", err)
continue
}
println("created ads stream")
// send() could be blocked on reading updates from the different update
// channels when it is not actually sending out messages. So, we need a
// way to break out of send() when recv() returns. This done channel is
// used to for that purpose.
done := make(chan struct{})
go v2c.send(stream, done)
if v2c.recv(stream) {
retries = 0
}
println("sending has succeeded")
close(done)
}
}
// endRequest sends a request for provided typeURL and resource on the provided
// stream.
//
// version is the ack version to be sent with the request
// - If this is the new request (not an ack/nack), version will be an empty
// string
// - If this is an ack, version will be the version from the response
// - If this is a nack, version will be the previous acked version (from
// ackMap). If there was no ack before, it will be an empty string
func (v2c *v2Client) sendRequest(stream adsStream, resourceNames []string, typeURL, version, nonce string) bool {
req := &xdspb.DiscoveryRequest{
Node: v2c.nodeProto,
TypeUrl: typeURL,
ResourceNames: resourceNames,
VersionInfo: version,
ResponseNonce: nonce,
// TODO: populate ErrorDetails for nack.
}
if err := stream.Send(req); err != nil {
log.Warningf("xds: request (type %s) for resource %v failed: %v", typeURL, resourceNames, err)
return false
}
return true
}
// sendExisting sends out xDS requests for registered watchers when recovering
// from a broken stream.
//
// We call stream.Send() here with the lock being held. It should be OK to do
// that here because the stream has just started and Send() usually returns
// quickly (once it pushes the message onto the transport layer) and is only
// ever blocked if we don't have enough flow control quota.
func (v2c *v2Client) sendExisting(stream adsStream) bool {
v2c.mu.Lock()
defer v2c.mu.Unlock()
// Reset the ack versions when the stream restarts.
v2c.ackMap = make(map[string]string)
for typeURL, wi := range v2c.watchMap {
if !v2c.sendRequest(stream, wi.target, typeURL, "", "") {
return false
}
}
return true
}
// processWatchInfo pulls the fields needed by the request from a watchInfo.
//
// It also calls callback with cached response, and updates the watch map in
// v2c.
//
// If the watch was already canceled, it returns false for send
func (v2c *v2Client) processWatchInfo(t *watchInfo) (target []string, typeURL, version, nonce string, send bool) {
v2c.mu.Lock()
defer v2c.mu.Unlock()
if t.state == watchCancelled {
return // This returns all zero values, and false for send.
}
t.state = watchStarted
send = true
typeURL = t.typeURL
target = t.target
v2c.checkCacheAndUpdateWatchMap(t)
// TODO: if watch is called again with the same resource names,
// there's no need to send another request.
//
// TODO: should we reset version (for ack) when a new watch is
// started? Or do this only if the resource names are different
// (so we send a new request)?
return
}
// processAckInfo pulls the fields needed by the ack request from a ackInfo.
//
// If no active watch is found for this ack, it returns false for send.
func (v2c *v2Client) processAckInfo(t *ackInfo) (target []string, typeURL, version, nonce string, send bool) {
typeURL = t.typeURL
v2c.mu.Lock()
defer v2c.mu.Unlock()
wi, ok := v2c.watchMap[typeURL]
if !ok {
// We don't send the request ack if there's no active watch (this can be
// either the server sends responses before any request, or the watch is
// canceled while the ackInfo is in queue), because there's no resource
// name. And if we send a request with empty resource name list, the
// server may treat it as a wild card and send us everything.
log.Warningf("xds: ack (type %s) not sent because there's no active watch for the type", typeURL)
return // This returns all zero values, and false for send.
}
send = true
version = t.version
nonce = t.nonce
target = wi.target
if version == "" {
// This is a nack, get the previous acked version.
version = v2c.ackMap[typeURL]
// version will still be an empty string if typeURL isn't
// found in ackMap, this can happen if there wasn't any ack
// before.
} else {
v2c.ackMap[typeURL] = version
}
return
}
// send reads watch infos from update channel and sends out actual xDS requests
// on the provided ADS stream.
func (v2c *v2Client) send(stream adsStream, done chan struct{}) {
if !v2c.sendExisting(stream) {
println("not existing stream")
return
}
println("in send")
for {
select {
case <-v2c.ctx.Done():
return
case u := <-v2c.sendCh.Get():
v2c.sendCh.Load()
var (
target []string
typeURL, version, nonce string
send bool
)
switch t := u.(type) {
case *watchInfo:
println("watchInfo")
target, typeURL, version, nonce, send = v2c.processWatchInfo(t)
case *ackInfo:
println("ackInfo")
target, typeURL, version, nonce, send = v2c.processAckInfo(t)
}
if !send {
continue
}
if !v2c.sendRequest(stream, target, typeURL, version, nonce) {
return
}
case <-done:
return
}
}
}
// recv receives xDS responses on the provided ADS stream and branches out to
// message specific handlers.
func (v2c *v2Client) recv(stream adsStream) bool {
success := false
for {
resp, err := stream.Recv()
// TODO: call watch callbacks with error when stream is broken.
if err != nil {
log.Warningf("xds: ADS stream recv failed: %v", err)
return success
}
var respHandleErr error
switch resp.GetTypeUrl() {
case cdsURL:
respHandleErr = v2c.handleCDSResponse(resp)
case edsURL:
respHandleErr = v2c.handleEDSResponse(resp)
default:
log.Warningf("xds: unknown response URL type: %v", resp.GetTypeUrl())
continue
}
typeURL := resp.GetTypeUrl()
if respHandleErr != nil {
log.Warningf("xds: response (type %s) handler failed: %v", typeURL, respHandleErr)
v2c.sendCh.Put(&ackInfo{
typeURL: typeURL,
version: "",
nonce: resp.GetNonce(),
})
continue
}
v2c.sendCh.Put(&ackInfo{
typeURL: typeURL,
version: resp.GetVersionInfo(),
nonce: resp.GetNonce(),
})
success = true
}
}
// watchCDS registers an CDS watcher for the provided clusterName. Updates
// corresponding to received CDS responses will be pushed to the provided
// callback. The caller can cancel the watch by invoking the returned cancel
// function.
// The provided callback should not block or perform any expensive operations
// or call other methods of the v2Client object.
func (v2c *v2Client) watchCDS(clusterName string, cdsCb cdsCallback) (cancel func()) {
return v2c.watch(&watchInfo{
typeURL: cdsURL,
target: []string{clusterName},
callback: cdsCb,
})
}
// watchEDS registers an EDS watcher for the provided clusterName. Updates
// corresponding to received EDS responses will be pushed to the provided
// callback. The caller can cancel the watch by invoking the returned cancel
// function.
// The provided callback should not block or perform any expensive operations
// or call other methods of the v2Client object.
func (v2c *v2Client) watchEDS(clusterName string, edsCb edsCallback) (cancel func()) {
return v2c.watch(&watchInfo{
typeURL: edsURL,
target: []string{clusterName},
callback: edsCb,
})
// TODO: Once a registered EDS watch is cancelled, we should send an EDS
// request with no resources. This will let the server know that we are no
// longer interested in this resource.
}
func (v2c *v2Client) watch(wi *watchInfo) (cancel func()) {
println("watch")
v2c.sendCh.Put(wi)
println("returning from watch")
return func() {
v2c.mu.Lock()
defer v2c.mu.Unlock()
if wi.state == watchEnqueued {
wi.state = watchCancelled
return
}
v2c.watchMap[wi.typeURL].cancel()
delete(v2c.watchMap, wi.typeURL)
// TODO: should we reset ack version string when cancelling the watch?
}
}
// checkCacheAndUpdateWatchMap is called when a new watch call is handled in
// send(). If an existing watcher is found, its expiry timer is stopped. If the
// watchInfo to be added to the watchMap is found in the cache, the watcher
// callback is immediately invoked.
//
// Caller should hold v2c.mu
func (v2c *v2Client) checkCacheAndUpdateWatchMap(wi *watchInfo) {
if existing := v2c.watchMap[wi.typeURL]; existing != nil {
existing.cancel()
}
v2c.watchMap[wi.typeURL] = wi
switch wi.typeURL {
// We need to grab the lock inside of the expiryTimer's afterFunc because
// we need to access the watchInfo, which is stored in the watchMap.
case cdsURL:
clusterName := wi.target[0]
if update, ok := v2c.cdsCache[clusterName]; ok {
var err error
if v2c.watchMap[cdsURL] == nil {
err = fmt.Errorf("xds: no CDS watcher found when handling CDS watch for cluster {%v} from cache", clusterName)
}
wi.callback.(cdsCallback)(update, err)
return
}
wi.expiryTimer = time.AfterFunc(defaultWatchExpiryTimeout, func() {
v2c.mu.Lock()
wi.callback.(cdsCallback)(CDSUpdate{}, fmt.Errorf("xds: CDS target %s not found", wi.target))
v2c.mu.Unlock()
})
case edsURL:
wi.expiryTimer = time.AfterFunc(defaultWatchExpiryTimeout, func() {
v2c.mu.Lock()
wi.callback.(edsCallback)(nil, fmt.Errorf("xds: EDS target %s not found", wi.target))
v2c.mu.Unlock()
})
}
}