Retry policy without perTryTimeout in VirtualService makes grpc request timeout unexpectlly #40299
Description
Bug Description
Retry policy without perTryTimeout in VirtualService will cause some UT request for grpc.
The default retry policy in RDS is
"route": {
"cluster": "outbound|1600||grpc-service.default.svc.cluster.local",
"timeout": "0s",
"retryPolicy": {
"retryOn": "connect-failure,refused-stream,unavailable,cancelled,retriable-status-codes",
"numRetries": 2,
"retryHostPredicate": [
{
"name": "envoy.retry_host_predicates.previous_hosts"
}
],
"hostSelectionRetryMaxAttempts": "5",
"retriableStatusCodes": [
503
]
},
"maxStreamDuration": {
"maxStreamDuration": "0s",
"grpcTimeoutHeaderMax": "0s"
}
}After I apply the VirtualService
spec:
hosts:
- grpc-service
http:
- retries:
attempts: 1
retryOn: unavailable,cancelled
route:
- destination:
host: grpc-service
port:
number: 1600The RDS config changes to
"route": {
"cluster": "outbound|1600||grpc-service.default.svc.cluster.local",
"timeout": "0s",
"retryPolicy": {
"retryOn": "unavailable,cancelled",
"numRetries": 1,
"retryHostPredicate": [
{
"name": "envoy.retry_host_predicates.previous_hosts"
}
],
"hostSelectionRetryMaxAttempts": "5"
},
"maxGrpcTimeout": "0s"
}It deleted maxStreamDuration and added maxGrpcTimeout, I can't confirm which configuration is causing those UT requests.
But I try to modify the defaultHttpRetryPolicy in mesh config, it will not cause these promblem.
defaultHttpRetryPolicy:
retryOn: unavailable
attempts: 1"route": {
"cluster": "outbound|1600||grpc-service.default.svc.cluster.local",
"timeout": "0s",
"retryPolicy": {
"retryOn": "unavailable",
"numRetries": 1,
"retryHostPredicate": [
{
"name": "envoy.retry_host_predicates.previous_hosts"
}
],
"hostSelectionRetryMaxAttempts": "5"
},
"maxStreamDuration": {
"maxStreamDuration": "0s",
"grpcTimeoutHeaderMax": "0s"
}
}Version
$ istioctl version
client version: 1.13.1
control plane version: 1.13.1
data plane version: 1.13.1 (135 proxies)
Additional Information
No response