gRPC Authentication with Cloud Run


Update 5/10/20: now contains OIDC tokens support; use that instead of

A couple months back I was happy to assist the Cloud Run (managed) team in validating gRPC support on that platform. The testing/validation covered writing a simple deployable gRPC client and server that also performed OpenIDConnect (OIDC) Authentication over gRPC (i.,e. Cloud Run Authentication). In the course of developing that, i gained an understanding of how gRPC authentication headers are handled and manged directly with gRPC. This article explains how to connect to a secure gRPC service running on Cloud Run using native gRPC library constructs.

The links cited in the Reference section discusses gRPC on Cloud Run but these do not cover either authentication at all or do not specify authentication using gRPC-centric constructs with Google Cloud Auth client libraries.

This article covers a simple client-server you can deploy on Cloud run that includes gRPC authentication using google cloud credentials . We specifically use ServiceAccount Credentials but the library cited below will work while running on GCE, GKE or even on Cloud RUn itself.


You can find the source here

OIDC Basics for GCP Services and gRPC

This article does not go into details about OpenID Connect tokens and how to use them with gRPC and GCP. However, as background please see

The links above shows how to acquire Google OIDC tokens that you can use for a variety of GCP services or even standalone. One thing noted that not all google cloud auth libraries provide interfaces to get OIDC tokens. At the time of writing, only google-auth libraries for java and python are supported with nodejs pending and golang yet to be officially implemented. Furthermore, most of those languages that do support acquiring OIDC token happen to also support automatic injection into gRPC calls.

For golang (as in the code in this article), it does not yet 1) support getting google OIDC tokens and 2) using those tokens in a library directly with gRPC. This article provides an unsupported implementation of both (from google’s official perspective atleast):

Both library sets above for golang also implements a specific TokenSource that uses a source google identity to get its OIDC token as yet another standard TokenSource or Credentials. For implementation details, see IdTokenSource.

Furthermore, that oauth.TokenSource{} also implements the interfaces that gRPC understands natively. What that means is gRPC clients if given that tokens source will automatically acquire, use, refresh and manage the lifecycle of the OIDC token!

For details, the specific interface that does that for gRPC is:

// NewIDTokenRPCCredential returns a crdential object for use with gRPC clients
func NewIDTokenRPCCredential(ctx context.Context, tokenSource oauth2.TokenSource) (credentials.PerRPCCredentials, error) 

// GetRequestMetadata gets the request metadata as a map from a TokenSource.
func (ts TokenSource) GetRequestMetadata(ctx context.Context, uri ...string) (map[string]string, error)

// RequireTransportSecurity indicates whether the credentials requires transport security.
func (ts TokenSource) RequireTransportSecurity() bool

As an example of direct usage of an IDToken with grpc native constructs like grpc.WithPerRPCCredentials():

import ""
	idTokenSource, err := idtoken.NewTokenSource(ctx, targetAudience, idtoken.WithCredentialsFile(serviceAccount))
	if err != nil {
		log.Fatalf("unable to create TokenSource: %v", err)
	tok, err := idTokenSource.Token()
	if err != nil {

    ce := credentials.NewTLS(&tlsCfg)
		conn, err = grpc.Dial(*address,

For equivalent samples in other languages see gRPC Authentication with Google OpenID Connect tokens.

Anyway, lets go directly into the details

I’m assuming you have Cloud Run setup and relatively above with gRPC and the auth concepts cited above

Setup Env Vars

    export PROJECT_ID=`gcloud config get-value core/project`
    gcloud config set run/region us-central1
    gcloud config set run/platform managed

Build and deploy gRPC Server Image

    docker build -t$PROJECT_ID/grpc_run_serve -f Dockerfile.server .
    docker push$PROJECT_ID/grpc_run_serve
    gcloud beta run deploy grpc --image$PROJECT_ID/grpc_run_serve --no-allow-unauthenticated

Create Client SA

Now create the service account that will have access to invoke the Cloud Run service

    mkdir -p certs
    gcloud iam service-accounts create grpc-client-account --display-name "gRPC Client Service Account"
    gcloud iam service-accounts keys create certs/grpc_client.json --iam-account=grpc-client-account@$

Set IAM Permission for roles/run.invoker:

cat <<EOT >> iam_policy.json
- members:
  - serviceAccount:grpc-client-account@$
  role: roles/run.invoker
version: 1

gcloud beta run services set-iam-policy grpc iam_policy.json 

Build Client

At this point, the gRPC service is secure by default and would require an OIDC token with the correct audience field and IAM permissions to get through

The audience filed for cloud run needs to be the fully qualified name with the protocol (custom domain aud fields is currently not supported)

    export AUDIENCE=`gcloud beta run services describe grpc --format="value(status.url)"`
    export ADDRESS=`echo $AUDIENCE |  awk -F[/:] '{print $4}'`
    echo $AUDIENCE
    echo $ADDRESS

On the root folder of this repo, run:

docker build -t$PROJECT_ID/grpc_run_client -f Dockerfile.client .

RUN gRPC Client

Now run the grpc client and specify the serviceAccount json file that is mounted inside the container (note: you should cd to the root of this repo so that the path to certs/ is mounted):

docker run -v `pwd`/certs:/certs -t$PROJECT_ID/grpc_run_client --address $ADDRESS:443 --usetls=true  --servername $ADDRESS --audience $AUDIENCE --serviceAccount /certs/grpc_client.json

or with go cli

``bash go run src/grpc_client.go
–address $ADDRESS:443 –usetls=true
–servername $ADDRESS –audience $AUDIENCE
–serviceAccount certs/grpc_client.json

The output of `grpc_run_client` will show the OIDC token sent to the cloud run instance which you can decode at [](  Note the `aud:`, `email` and `iss` fields 

  "iss": "",
  "aud": "",
  "azp": "",
  "sub": "101659512549165144150",
  "email": "",
  "email_verified": true,
  "iat": 1572983749,
  "exp": 1572987349

The second portion is 5 unary responses back from the GRPC service that displays the K_REVISION env variable from Cloud RUn

2019/11/05 20:52:14 RPC Response: 0 message:"Hello unary RPC msg   from K_REVISION grpc-tnslx" 
2019/11/05 20:52:15 RPC Response: 1 message:"Hello unary RPC msg   from K_REVISION grpc-tnslx" 
2019/11/05 20:52:16 RPC Response: 2 message:"Hello unary RPC msg   from K_REVISION grpc-tnslx" 
2019/11/05 20:52:17 RPC Response: 3 message:"Hello unary RPC msg   from K_REVISION grpc-tnslx" 
2019/11/05 20:52:18 RPC Response: 4 message:"Hello unary RPC msg   from K_REVISION grpc-tnslx" 

The final output is a buffered form Server-side Streaming messages back (i.,e the server sends back two responses back on the single request). At the time of writing 11/5/19, server streaming is not officially supported as its not true streaming but a buffered response anyway

2019/11/05 20:52:18 Stream Header: %!(EXTRA metadata.MD=map[alt-svc:[quic=":443"; ma=2592000; v="46,43",h3-Q049=":443"; ma=2592000,h3-Q048=":443"; ma=2592000,h3-Q046=":443"; ma=2592000,h3-Q043=":443"; ma=2592000] content-type:[application/grpc] streamheaderkey:[val] x-cloud-trace-context:[7928b8ab5aa6b2dc759ca6ff7fa5bb4c] date:[Tue, 05 Nov 2019 20:52:19 GMT] server:[Google Frontend] content-length:[52]])

2019/11/05 20:52:18 Message: %!(EXTRA string=Msg1 Stream RPC msg)

2019/11/05 20:52:18 Stream Header: %!(EXTRA metadata.MD=map[date:[Tue, 05 Nov 2019 20:52:19 GMT] server:[Google Frontend] content-length:[52] alt-svc:[quic=":443"; ma=2592000; v="46,43",h3-Q049=":443"; ma=2592000,h3-Q048=":443"; ma=2592000,h3-Q046=":443"; ma=2592000,h3-Q043=":443"; ma=2592000] content-type:[application/grpc] streamheaderkey:[val] x-cloud-trace-context:[7928b8ab5aa6b2dc759ca6ff7fa5bb4c]])

2019/11/05 20:52:18 Message: %!(EXTRA string=Msg2 Stream RPC msg)

2019/11/05 20:52:18 Stream Trailer:  map[]

Note client->gRPC server side streaming for Cloud Run (managed) is not currently supported

enjoy grpc-ing!



Openssl CA

I added in a sample SSL certificate set into the images which are not used since Cloud Run automatically does SSL management for you. However, if you wanted to run this over Self-Signed certs, here is a setup to create your own CA (note, you will need to edit openssl.conf here and specify the SNI settings (i.,e override value in

    DNS.1 =

This site supports webmentions. Send me a mention via this form.

comments powered by Disqus