frostfs-node/pkg/network/cache/multi.go
Evgenii Stratonikov 8377372a40 [] go.mod: Update api-go
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
2024-07-26 16:44:19 +03:00

378 lines
8.4 KiB
Go

package cache
import (
"context"
"errors"
"fmt"
"sync"
"time"
rawclient "git.frostfs.info/TrueCloudLab/frostfs-api-go/v2/rpc/client"
clientcore "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/core/client"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/network"
metrics "git.frostfs.info/TrueCloudLab/frostfs-observability/metrics/grpc"
tracing "git.frostfs.info/TrueCloudLab/frostfs-observability/tracing/grpc"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client"
objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
"google.golang.org/grpc"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
type singleClient struct {
sync.RWMutex
client clientcore.Client
lastAttempt time.Time
}
type multiClient struct {
mtx sync.RWMutex
clients map[string]*singleClient
// addrMtx protects addr field. Should not be taken before the mtx.
addrMtx sync.RWMutex
addr network.AddressGroup
opts ClientCacheOpts
}
const defaultReconnectInterval = time.Second * 30
func newMultiClient(addr network.AddressGroup, opts ClientCacheOpts) *multiClient {
if opts.ReconnectTimeout <= 0 {
opts.ReconnectTimeout = defaultReconnectInterval
}
return &multiClient{
clients: make(map[string]*singleClient),
addr: addr,
opts: opts,
}
}
func (x *multiClient) createForAddress(ctx context.Context, addr network.Address) (clientcore.Client, error) {
var c client.Client
prmInit := client.PrmInit{
DisableFrostFSErrorResolution: true,
}
if x.opts.Key != nil {
prmInit.Key = *x.opts.Key
}
prmDial := client.PrmDial{
Endpoint: addr.URIAddr(),
GRPCDialOptions: []grpc.DialOption{
grpc.WithChainUnaryInterceptor(
metrics.NewUnaryClientInterceptor(),
tracing.NewUnaryClientInteceptor(),
),
grpc.WithChainStreamInterceptor(
metrics.NewStreamClientInterceptor(),
tracing.NewStreamClientInterceptor(),
),
},
}
if x.opts.DialTimeout > 0 {
prmDial.DialTimeout = x.opts.DialTimeout
}
if x.opts.StreamTimeout > 0 {
prmDial.StreamTimeout = x.opts.StreamTimeout
}
if x.opts.ResponseCallback != nil {
prmInit.ResponseInfoCallback = x.opts.ResponseCallback
}
c.Init(prmInit)
err := c.Dial(ctx, prmDial)
if err != nil {
return nil, fmt.Errorf("can't init SDK client: %w", err)
}
return &c, nil
}
// updateGroup replaces current multiClient addresses with a new group.
// Old addresses not present in group are removed.
func (x *multiClient) updateGroup(group network.AddressGroup) {
// Firstly, remove old clients.
cache := make([]string, 0, group.Len())
group.IterateAddresses(func(a network.Address) bool {
cache = append(cache, a.String())
return false
})
x.addrMtx.RLock()
oldGroup := x.addr
x.addrMtx.RUnlock()
if len(oldGroup) == len(cache) {
needUpdate := false
for i := range oldGroup {
if cache[i] != oldGroup[i].String() {
needUpdate = true
break
}
}
if !needUpdate {
return
}
}
x.mtx.Lock()
defer x.mtx.Unlock()
loop:
for a := range x.clients {
for i := range cache {
if cache[i] == a {
continue loop
}
}
x.clients[a].invalidate()
delete(x.clients, a)
}
// Then add new clients.
x.addrMtx.Lock()
x.addr = group
x.addrMtx.Unlock()
}
var errRecentlyFailed = errors.New("client has recently failed, skipping")
func (x *multiClient) iterateClients(ctx context.Context, f func(clientcore.Client) error) error {
var firstErr error
x.addrMtx.RLock()
group := x.addr
x.addrMtx.RUnlock()
group.IterateAddresses(func(addr network.Address) bool {
select {
case <-ctx.Done():
firstErr = context.Canceled
return true
default:
}
var err error
c, err := x.client(ctx, addr)
if err == nil {
err = f(c)
}
// non-status logic error that could be returned
// from the SDK client; should not be considered
// as a connection error
var siErr *objectSDK.SplitInfoError
var eiErr *objectSDK.ECInfoError
success := err == nil || errors.Is(err, context.Canceled) || errors.As(err, &siErr) || errors.As(err, &eiErr)
if success || firstErr == nil || errors.Is(firstErr, errRecentlyFailed) {
firstErr = err
}
if err != nil {
x.ReportError(err)
}
return success
})
return firstErr
}
func (x *multiClient) ReportError(err error) {
if errors.Is(err, errRecentlyFailed) {
return
}
if status.Code(err) == codes.Canceled || errors.Is(err, context.Canceled) {
return
}
// non-status logic error that could be returned
// from the SDK client; should not be considered
// as a connection error
var siErr *objectSDK.SplitInfoError
var eiErr *objectSDK.ECInfoError
if errors.As(err, &siErr) || errors.As(err, &eiErr) {
return
}
// Dropping all clients here is not necessary, we do this
// because `multiClient` doesn't yet provide convenient interface
// for reporting individual errors for streaming operations.
x.mtx.RLock()
for _, sc := range x.clients {
sc.invalidate()
}
x.mtx.RUnlock()
}
func (s *singleClient) invalidate() {
s.Lock()
if s.client != nil {
_ = s.client.Close()
}
s.client = nil
s.Unlock()
}
func (x *multiClient) ObjectPutInit(ctx context.Context, p client.PrmObjectPutInit) (res client.ObjectWriter, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectPutInit(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectPutSingle(ctx context.Context, p client.PrmObjectPutSingle) (res *client.ResObjectPutSingle, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectPutSingle(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectDelete(ctx context.Context, p client.PrmObjectDelete) (res *client.ResObjectDelete, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectDelete(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectGetInit(ctx context.Context, p client.PrmObjectGet) (res *client.ObjectReader, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectGetInit(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectRangeInit(ctx context.Context, p client.PrmObjectRange) (res *client.ObjectRangeReader, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectRangeInit(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectHead(ctx context.Context, p client.PrmObjectHead) (res *client.ResObjectHead, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectHead(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectHash(ctx context.Context, p client.PrmObjectHash) (res *client.ResObjectHash, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectHash(ctx, p)
return err
})
return
}
func (x *multiClient) ObjectSearchInit(ctx context.Context, p client.PrmObjectSearch) (res *client.ObjectListReader, err error) {
err = x.iterateClients(ctx, func(c clientcore.Client) error {
res, err = c.ObjectSearchInit(ctx, p)
return err
})
return
}
func (x *multiClient) ExecRaw(func(client *rawclient.Client) error) error {
panic("multiClient.ExecRaw() must not be called")
}
func (x *multiClient) Close() error {
x.mtx.RLock()
{
for _, c := range x.clients {
if c.client != nil {
_ = c.client.Close()
}
}
}
x.mtx.RUnlock()
return nil
}
func (x *multiClient) RawForAddress(ctx context.Context, addr network.Address, f func(client *rawclient.Client) error) error {
c, err := x.client(ctx, addr)
if err != nil {
return err
}
err = c.ExecRaw(f)
if err != nil {
x.ReportError(err)
}
return err
}
func (x *multiClient) client(ctx context.Context, addr network.Address) (clientcore.Client, error) {
strAddr := addr.String()
x.mtx.RLock()
c, cached := x.clients[strAddr]
x.mtx.RUnlock()
if cached {
c.RLock()
if c.client != nil {
cl := c.client
c.RUnlock()
return cl, nil
}
if x.opts.ReconnectTimeout != 0 && time.Since(c.lastAttempt) < x.opts.ReconnectTimeout {
c.RUnlock()
return nil, errRecentlyFailed
}
c.RUnlock()
} else {
var ok bool
x.mtx.Lock()
c, ok = x.clients[strAddr]
if !ok {
c = new(singleClient)
x.clients[strAddr] = c
}
x.mtx.Unlock()
}
c.Lock()
defer c.Unlock()
if c.client != nil {
return c.client, nil
}
if x.opts.ReconnectTimeout != 0 && time.Since(c.lastAttempt) < x.opts.ReconnectTimeout {
return nil, errRecentlyFailed
}
cl, err := x.createForAddress(ctx, addr)
if err != nil {
c.lastAttempt = time.Now()
return nil, err
}
c.client = cl
return cl, nil
}