tetra: use the builtin gRPC retry backoff mechanism

gRGC A6 - gRPC Retry Design (a.k.a. built in backoff retry)
https://github.com/grpc/proposal/blob/master/A6-client-retries.md was
implemented by grpc/grpc-go#2111 but unusable
for a long time since maxAttempts was limited to hardcoded 5
(grpc/grpc-go#4615), recent PR fixed that
grpc/grpc-go#7229.

It's transparent to the user, to see it in action, make sure the gRPC
server is unreachable (do not start tetragon for example), run tetra
with: GRPC_GO_LOG_SEVERITY_LEVEL=warning <tetra cmd>

Note that logs don't always have the time to be pushed before exit so
output might be a bit off but the number of retries is respected (you
can debug or synchronously print in the grpc/stream.c:shouldRetry or
:withRetry to verify).

Also note that the final backoff duration is completely random and
chosen between 0 and the final duration that was computed via to the
params: https://github.com/grpc/grpc-go/blob/v1.65.0/stream.go#L702

Signed-off-by: Mahe Tardy <[email protected]>

cmd/tetra/common/client.go

@@ -15,6 +15,47 @@ import (
 	"google.golang.org/grpc/credentials/insecure"
 )
 
+// gRGC A6 - gRPC Retry Design (a.k.a. built in backoff retry)
+// https://github.com/grpc/proposal/blob/master/A6-client-retries.md
+// was implemented by https://github.com/grpc/grpc-go/pull/2111 but unusable
+// for a long time since maxAttempts was limited to hardcoded 5
+// (https://github.com/grpc/grpc-go/issues/4615), recent PR fixed that
+// https://github.com/grpc/grpc-go/pull/7229.
+//
+// It's transparent to the user, to see it in action, make sure the gRPC server
+// is unreachable (do not start tetragon for example), run tetra with:
+// GRPC_GO_LOG_SEVERITY_LEVEL=warning <tetra cmd>
+//
+// Note that logs don't always have the time to be pushed before exit so output
+// might be a bit off but the number of retries is respected (you can debug or
+// synchronously print in the grpc/stream.c:shouldRetry or :withRetry to
+// verify).
+//
+// Also note that the final backoff duration is completely random and chosen
+// between 0 and the final duration that was computed via to the params:
+// https://github.com/grpc/grpc-go/blob/v1.65.0/stream.go#L702
+func retryPolicy(retries int) string {
+	if retries < 0 {
+		// gRPC should ignore the invalid retry policy but will issue a warning,
+		return "{}"
+	}
+	// maxAttempt includes the first call
+	maxAttempt := retries + 1
+	// let's not limit backoff by hardcoding 1h in MaxBackoff
+	// since we need to provide a value >0
+	return fmt.Sprintf(`{
+	"methodConfig": [{
+	  "name": [{"service": "tetragon.FineGuidanceSensors"}],
+	  "retryPolicy": {
+		  "MaxAttempts": %d,
+		  "InitialBackoff": "1s",
+		  "MaxBackoff": "3600s",
+		  "BackoffMultiplier": 2,
+		  "RetryableStatusCodes": [ "UNAVAILABLE" ]
+	  }
+	}]}`, maxAttempt)
+}
+
 func CliRunErr(fn func(ctx context.Context, cli tetragon.FineGuidanceSensorsClient), fnErr func(err error)) {
 	c, err := NewClientWithDefaultContextAndAddress()
 	if err != nil {
@@ -60,7 +101,11 @@ func NewClient(ctx context.Context, address string, timeout time.Duration) (*Cli
 	c.Ctx, _ = signal.NotifyContext(timeoutContext, syscall.SIGINT, syscall.SIGTERM)
 
 	var err error
-	c.conn, err = grpc.NewClient(address, grpc.WithTransportCredentials(insecure.NewCredentials()))
+	c.conn, err = grpc.NewClient(address,
+		grpc.WithTransportCredentials(insecure.NewCredentials()),
+		grpc.WithDefaultServiceConfig(retryPolicy(Retries)),
+		grpc.WithMaxCallAttempts(Retries+1), // maxAttempt includes the first call
+	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create gRPC client with address %s: %w", address, err)
 	}