frostfs-node/pkg/services/tree/cache.go
Dmitrii Stepanov 652237500f
All checks were successful
Vulncheck / Vulncheck (push) Successful in 1m7s
Pre-commit hooks / Pre-commit (push) Successful in 1m36s
Build / Build Components (push) Successful in 1m47s
Tests and linters / gopls check (push) Successful in 3m50s
Tests and linters / Tests (push) Successful in 4m4s
Tests and linters / Run gofumpt (push) Successful in 4m0s
Tests and linters / Staticcheck (push) Successful in 4m7s
Tests and linters / Lint (push) Successful in 4m11s
Tests and linters / Tests with -race (push) Successful in 4m46s
OCI image / Build container images (push) Successful in 4m30s
[#1689] tree: Dial tree service after create connection on sync
Background trees sync creates grpc connection with
`grpc.WithDefaultCallOptions(grpc.WaitForReady(true))` option. When grpc
connection created with this option, client will wait until a connection
becomes available or the RPC's deadline is reached. As background sync
has no timeout in context, so in case of client is in TRANSIENT_FAILURE
RPC call will hang forever.

Change-Id: I17c8c1d2779bb81c541f47dd0e558e0b8ed2e7c1
Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
2025-04-24 17:05:43 +03:00

90 lines
1.9 KiB
Go

package tree
import (
"context"
"crypto/ecdsa"
"errors"
"fmt"
"sync"
"time"
internalNet "git.frostfs.info/TrueCloudLab/frostfs-node/internal/net"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/network"
"github.com/hashicorp/golang-lru/v2/simplelru"
"google.golang.org/grpc"
"google.golang.org/grpc/connectivity"
)
type clientCache struct {
sync.Mutex
simplelru.LRU[string, cacheItem]
key *ecdsa.PrivateKey
ds *internalNet.DialerSource
}
type cacheItem struct {
cc *grpc.ClientConn
lastTry time.Time
}
const (
defaultClientCacheSize = 32
defaultClientConnectTimeout = time.Second * 2
defaultReconnectInterval = time.Second * 15
)
var errRecentlyFailed = errors.New("client has recently failed")
func (c *clientCache) init(pk *ecdsa.PrivateKey, ds *internalNet.DialerSource) {
l, _ := simplelru.NewLRU(defaultClientCacheSize, func(_ string, value cacheItem) {
if conn := value.cc; conn != nil {
_ = conn.Close()
}
})
c.LRU = *l
c.key = pk
c.ds = ds
}
func (c *clientCache) get(ctx context.Context, netmapAddr string) (TreeServiceClient, error) {
c.Lock()
ccInt, ok := c.Get(netmapAddr)
c.Unlock()
if ok {
item := ccInt
if item.cc == nil {
if d := time.Since(item.lastTry); d < defaultReconnectInterval {
return nil, fmt.Errorf("%w: %s till the next reconnection to %s",
errRecentlyFailed, d, netmapAddr)
}
} else {
if s := item.cc.GetState(); s == connectivity.Idle || s == connectivity.Ready {
return NewTreeServiceClient(item.cc), nil
}
_ = item.cc.Close()
}
}
var netAddr network.Address
if err := netAddr.FromString(netmapAddr); err != nil {
return nil, err
}
cc, err := dialTreeService(ctx, netAddr, c.key, c.ds)
lastTry := time.Now()
c.Lock()
if err != nil {
c.Add(netmapAddr, cacheItem{cc: nil, lastTry: lastTry})
} else {
c.Add(netmapAddr, cacheItem{cc: cc, lastTry: lastTry})
}
c.Unlock()
if err != nil {
return nil, err
}
return NewTreeServiceClient(cc), nil
}