feat: grpc add retry middleware (#1561)

Signed-off-by: Gaius <gaius.qi@gmail.com>

Signed-off-by: Gaius <gaius.qi@gmail.com>
This commit is contained in:
Gaius 2022-08-12 18:11:18 +08:00
parent 242fb2bc37
commit 22b8e6368e
No known key found for this signature in database
GPG Key ID: 8B4E5D1290FA2FFB
4 changed files with 130 additions and 134 deletions

View File

@ -24,9 +24,9 @@ import (
grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware" grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware"
grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap" grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap"
grpc_retry "github.com/grpc-ecosystem/go-grpc-middleware/retry"
grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus" grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus"
"google.golang.org/grpc" "google.golang.org/grpc"
"google.golang.org/grpc/backoff"
"google.golang.org/grpc/credentials/insecure" "google.golang.org/grpc/credentials/insecure"
cdnsystemv1 "d7y.io/api/pkg/apis/cdnsystem/v1" cdnsystemv1 "d7y.io/api/pkg/apis/cdnsystem/v1"
@ -39,25 +39,30 @@ import (
) )
const ( const (
backoffBaseDelay = 1 * time.Second // maxRetries is maximum number of retries.
backoffMultiplier = 1.2 maxRetries = 3
backoffJitter = 0.2
backoffMaxDelay = 120 * time.Second // backoffWaitBetween is waiting for a fixed period of
minConnectTime = 3 * time.Second //fast fail, leave time to try other scheduler // time between calls in backoff linear.
backoffWaitBetween = 500 * time.Millisecond
// perRetryTimeout is GRPC timeout per call (including initial call) on this call.
perRetryTimeout = 3 * time.Second
) )
// defaultDialOptions is default dial options of manager client.
var defaultDialOptions = []grpc.DialOption{ var defaultDialOptions = []grpc.DialOption{
grpc.WithDefaultServiceConfig(balancer.BalancerServiceConfig), grpc.WithDefaultServiceConfig(balancer.BalancerServiceConfig),
grpc.WithTransportCredentials(insecure.NewCredentials()), grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithConnectParams(grpc.ConnectParams{ grpc.WithUnaryInterceptor(grpc_middleware.ChainUnaryClient(
Backoff: backoff.Config{ grpc_prometheus.UnaryClientInterceptor,
BaseDelay: backoffBaseDelay, grpc_zap.UnaryClientInterceptor(logger.GrpcLogger.Desugar()),
Multiplier: backoffMultiplier, grpc_retry.UnaryClientInterceptor(
Jitter: backoffJitter, grpc_retry.WithPerRetryTimeout(perRetryTimeout),
MaxDelay: backoffMaxDelay, grpc_retry.WithMax(maxRetries),
}, grpc_retry.WithBackoff(grpc_retry.BackoffLinear(backoffWaitBetween)),
MinConnectTimeout: minConnectTime, ),
}), )),
grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient( grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient(
grpc_prometheus.StreamClientInterceptor, grpc_prometheus.StreamClientInterceptor,
grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()), grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()),

View File

@ -25,9 +25,9 @@ import (
grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware" grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware"
grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap" grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap"
grpc_retry "github.com/grpc-ecosystem/go-grpc-middleware/retry"
grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus" grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus"
"google.golang.org/grpc" "google.golang.org/grpc"
"google.golang.org/grpc/backoff"
"google.golang.org/grpc/codes" "google.golang.org/grpc/codes"
"google.golang.org/grpc/credentials/insecure" "google.golang.org/grpc/credentials/insecure"
"google.golang.org/grpc/status" "google.golang.org/grpc/status"
@ -40,13 +40,65 @@ import (
) )
const ( const (
contextTimeout = 2 * time.Minute // maxRetries is maximum number of retries.
backoffBaseDelay = 1 * time.Second maxRetries = 3
backoffMultiplier = 1.6
backoffJitter = 0.2 // backoffWaitBetween is waiting for a fixed period of
backoffMaxDelay = 10 * time.Second // time between calls in backoff linear.
backoffWaitBetween = 500 * time.Millisecond
// perRetryTimeout is GRPC timeout per call (including initial call) on this call.
perRetryTimeout = 5 * time.Second
) )
// defaultDialOptions is default dial options of manager client.
var defaultDialOptions = []grpc.DialOption{
grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithUnaryInterceptor(grpc_middleware.ChainUnaryClient(
grpc_prometheus.UnaryClientInterceptor,
grpc_zap.UnaryClientInterceptor(logger.GrpcLogger.Desugar()),
grpc_retry.UnaryClientInterceptor(
grpc_retry.WithPerRetryTimeout(perRetryTimeout),
grpc_retry.WithMax(maxRetries),
grpc_retry.WithBackoff(grpc_retry.BackoffLinear(backoffWaitBetween)),
),
)),
grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient(
grpc_prometheus.StreamClientInterceptor,
grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()),
)),
}
// GetClient returns manager client.
func GetClient(target string, options ...grpc.DialOption) (Client, error) {
conn, err := grpc.Dial(
target,
append(defaultDialOptions, options...)...,
)
if err != nil {
return nil, err
}
return &client{
ManagerClient: managerv1.NewManagerClient(conn),
conn: conn,
}, nil
}
// GetClientByAddr returns manager client with addresses.
func GetClientByAddr(netAddrs []dfnet.NetAddr, opts ...grpc.DialOption) (Client, error) {
for _, netAddr := range netAddrs {
ipReachable := reachable.New(&reachable.Config{Address: netAddr.Addr})
if err := ipReachable.Check(); err == nil {
logger.Infof("use %s address for manager grpc client", netAddr.Addr)
return GetClient(netAddr.Addr, opts...)
}
logger.Warnf("%s manager address can not reachable", netAddr.Addr)
}
return nil, errors.New("can not find available manager addresses")
}
// Client is the interface for grpc client. // Client is the interface for grpc client.
type Client interface { type Client interface {
// Update Seed peer configuration. // Update Seed peer configuration.
@ -80,100 +132,34 @@ type client struct {
conn *grpc.ClientConn conn *grpc.ClientConn
} }
// GetClient returns manager client.
func GetClient(target string, options ...grpc.DialOption) (Client, error) {
dialOptions := []grpc.DialOption{
grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithBlock(),
grpc.WithConnectParams(grpc.ConnectParams{
Backoff: backoff.Config{
BaseDelay: backoffBaseDelay,
Multiplier: backoffMultiplier,
Jitter: backoffJitter,
MaxDelay: backoffMaxDelay,
},
}),
grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient(
grpc_prometheus.StreamClientInterceptor,
grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()),
)),
}
dialOptions = append(dialOptions, options...)
conn, err := grpc.Dial(
target,
dialOptions...,
)
if err != nil {
return nil, err
}
return &client{
ManagerClient: managerv1.NewManagerClient(conn),
conn: conn,
}, nil
}
// GetClientByAddr returns manager client with addresses.
func GetClientByAddr(netAddrs []dfnet.NetAddr, opts ...grpc.DialOption) (Client, error) {
for _, netAddr := range netAddrs {
ipReachable := reachable.New(&reachable.Config{Address: netAddr.Addr})
if err := ipReachable.Check(); err == nil {
logger.Infof("use %s address for manager grpc client", netAddr.Addr)
return GetClient(netAddr.Addr, opts...)
}
logger.Warnf("%s manager address can not reachable", netAddr.Addr)
}
return nil, errors.New("can not find available manager addresses")
}
// Update SeedPeer configuration. // Update SeedPeer configuration.
func (c *client) UpdateSeedPeer(req *managerv1.UpdateSeedPeerRequest) (*managerv1.SeedPeer, error) { func (c *client) UpdateSeedPeer(req *managerv1.UpdateSeedPeerRequest) (*managerv1.SeedPeer, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.UpdateSeedPeer(context.Background(), req)
defer cancel()
return c.ManagerClient.UpdateSeedPeer(ctx, req)
} }
// Get Scheduler and Scheduler cluster configuration. // Get Scheduler and Scheduler cluster configuration.
func (c *client) GetScheduler(req *managerv1.GetSchedulerRequest) (*managerv1.Scheduler, error) { func (c *client) GetScheduler(req *managerv1.GetSchedulerRequest) (*managerv1.Scheduler, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.GetScheduler(context.Background(), req)
defer cancel()
return c.ManagerClient.GetScheduler(ctx, req)
} }
// Update scheduler configuration. // Update scheduler configuration.
func (c *client) UpdateScheduler(req *managerv1.UpdateSchedulerRequest) (*managerv1.Scheduler, error) { func (c *client) UpdateScheduler(req *managerv1.UpdateSchedulerRequest) (*managerv1.Scheduler, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.UpdateScheduler(context.Background(), req)
defer cancel()
return c.ManagerClient.UpdateScheduler(ctx, req)
} }
// List acitve schedulers configuration. // List acitve schedulers configuration.
func (c *client) ListSchedulers(req *managerv1.ListSchedulersRequest) (*managerv1.ListSchedulersResponse, error) { func (c *client) ListSchedulers(req *managerv1.ListSchedulersRequest) (*managerv1.ListSchedulersResponse, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.ListSchedulers(context.Background(), req)
defer cancel()
return c.ManagerClient.ListSchedulers(ctx, req)
} }
// Get object storage configuration. // Get object storage configuration.
func (c *client) GetObjectStorage(req *managerv1.GetObjectStorageRequest) (*managerv1.ObjectStorage, error) { func (c *client) GetObjectStorage(req *managerv1.GetObjectStorageRequest) (*managerv1.ObjectStorage, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.GetObjectStorage(context.Background(), req)
defer cancel()
return c.ManagerClient.GetObjectStorage(ctx, req)
} }
// List buckets configuration. // List buckets configuration.
func (c *client) ListBuckets(req *managerv1.ListBucketsRequest) (*managerv1.ListBucketsResponse, error) { func (c *client) ListBuckets(req *managerv1.ListBucketsRequest) (*managerv1.ListBucketsResponse, error) {
ctx, cancel := context.WithTimeout(context.Background(), contextTimeout) return c.ManagerClient.ListBuckets(context.Background(), req)
defer cancel()
return c.ManagerClient.ListBuckets(ctx, req)
} }
// List acitve schedulers configuration. // List acitve schedulers configuration.

View File

@ -24,9 +24,9 @@ import (
grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware" grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware"
grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap" grpc_zap "github.com/grpc-ecosystem/go-grpc-middleware/logging/zap"
grpc_retry "github.com/grpc-ecosystem/go-grpc-middleware/retry"
grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus" grpc_prometheus "github.com/grpc-ecosystem/go-grpc-prometheus"
"google.golang.org/grpc" "google.golang.org/grpc"
"google.golang.org/grpc/backoff"
"google.golang.org/grpc/credentials/insecure" "google.golang.org/grpc/credentials/insecure"
commonv1 "d7y.io/api/pkg/apis/common/v1" commonv1 "d7y.io/api/pkg/apis/common/v1"
@ -39,13 +39,52 @@ import (
) )
const ( const (
backoffBaseDelay = 1 * time.Second // maxRetries is maximum number of retries.
backoffMultiplier = 1.2 maxRetries = 3
backoffJitter = 0.2
backoffMaxDelay = 120 * time.Second // backoffWaitBetween is waiting for a fixed period of
minConnectTime = 3 * time.Second //fast fail, leave time to try other scheduler // time between calls in backoff linear.
backoffWaitBetween = 500 * time.Millisecond
// perRetryTimeout is GRPC timeout per call (including initial call) on this call.
perRetryTimeout = 3 * time.Second
) )
// defaultDialOptions is default dial options of manager client.
var defaultDialOptions = []grpc.DialOption{
grpc.WithDefaultServiceConfig(balancer.BalancerServiceConfig),
grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithUnaryInterceptor(grpc_middleware.ChainUnaryClient(
grpc_prometheus.UnaryClientInterceptor,
grpc_zap.UnaryClientInterceptor(logger.GrpcLogger.Desugar()),
grpc_retry.UnaryClientInterceptor(
grpc_retry.WithPerRetryTimeout(perRetryTimeout),
grpc_retry.WithMax(maxRetries),
grpc_retry.WithBackoff(grpc_retry.BackoffLinear(backoffWaitBetween)),
),
)),
grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient(
grpc_prometheus.StreamClientInterceptor,
grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()),
)),
}
// GetClient get scheduler clients using resolver and balancer,
func GetClient(options ...grpc.DialOption) (Client, error) {
conn, err := grpc.Dial(
resolver.SchedulerVirtualTarget,
append(defaultDialOptions, options...)...,
)
if err != nil {
return nil, err
}
return &client{
conn,
schedulerv1.NewSchedulerClient(conn),
}, nil
}
// NewBeginOfPiece creates begin of piece. // NewBeginOfPiece creates begin of piece.
func NewBeginOfPiece(taskID, peerID string) *schedulerv1.PieceResult { func NewBeginOfPiece(taskID, peerID string) *schedulerv1.PieceResult {
return &schedulerv1.PieceResult{ return &schedulerv1.PieceResult{
@ -69,41 +108,6 @@ func NewEndOfPiece(taskID, peerID string, finishedCount int32) *schedulerv1.Piec
} }
} }
// GetClient get scheduler clients using resolver and balancer,
func GetClient(options ...grpc.DialOption) (Client, error) {
dialOptions := []grpc.DialOption{
grpc.WithDefaultServiceConfig(balancer.BalancerServiceConfig),
grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithConnectParams(grpc.ConnectParams{
Backoff: backoff.Config{
BaseDelay: backoffBaseDelay,
Multiplier: backoffMultiplier,
Jitter: backoffJitter,
MaxDelay: backoffMaxDelay,
},
MinConnectTimeout: minConnectTime,
}),
grpc.WithStreamInterceptor(grpc_middleware.ChainStreamClient(
grpc_prometheus.StreamClientInterceptor,
grpc_zap.StreamClientInterceptor(logger.GrpcLogger.Desugar()),
)),
}
dialOptions = append(dialOptions, options...)
conn, err := grpc.Dial(
resolver.SchedulerVirtualTarget,
dialOptions...,
)
if err != nil {
return nil, err
}
return &client{
conn,
schedulerv1.NewSchedulerClient(conn),
}, nil
}
// Client is the interface for grpc client. // Client is the interface for grpc client.
type Client interface { type Client interface {
// RegisterPeerTask registers a peer into task. // RegisterPeerTask registers a peer into task.

View File

@ -33,6 +33,7 @@ import (
const ( const (
// SeedTag Default value of tag label for seed peer. // SeedTag Default value of tag label for seed peer.
SeedTag = "d7y/seed" SeedTag = "d7y/seed"
// SeedApplication Default value of application label for seed peer. // SeedApplication Default value of application label for seed peer.
SeedApplication = "d7y/seed" SeedApplication = "d7y/seed"
) )