Here we describe how to do auth with a package that uses gargle, without requiring any user interaction. This comes up in a wide array of contexts, ranging from simple rendering of a local R Markdown document to deploying a data product on a remote server.

We assume the wrapper package uses the design described in How to use gargle for auth in a client package. Examples include:

Full details on gargle::token_fetch(), which powers this strategy, are given in How gargle gets tokens.

Provide a token or pre-authorize token discovery

The main principle for auth that does not require user interaction:

Provide a token directly or take advance measures that indicate you want a token to be discovered.

We present several ways to achieve this, basically in order of preference.

Provide a service account token directly

When two computers are talking to each other, possibly with no human involvement, the most appropriate type of token to use is a service account token.

This requires some advance preparation, but that tends to pay off pretty quickly, in terms of having a much more robust auth setup.

Step 1: Get a service account and then download a token. Described in the gargle article How to get your own API credentials, specifically in the Service account token section.

Step 2: Call the wrapper package’s main auth function proactively and provide the path to your service account token. Example using googledrive:

If this code is running on, e.g., a continuous integration service and you need to use an encrypted token, see the gargle article Managing tokens securely.

For certain APIs, service accounts are inherently awkward, in which case you might have to resort to an OAuth user token. Gmail is a good example. In this case, you might want the service account to act on your behalf, e.g., to send email as you. This is described as “impersonation”, which should tip you off that Google does not really encourage this workflow. This requires granting the service account domain-wide authority, which requires that your account is associated with a G Suite domain and also approval from the domain’s administrator. If this is impossible or unappealing, use an OAuth user token, as described below.

Rig a service account token as Application Default Credentials

Wrapper packages that use gargle::token_fetch() in the recommended way have access to the token search strategy known as Application Default Credentials.

You need to put your service token in a very specific location or, alternatively, record the location of your service token in a specific environment variable.

Full details are in the credentials_app_default() section of the gargle article How gargle gets tokens.

If you have your token rigged properly, you do not need to do anything else, i.e. you do not need to call PACKAGE_auth() explicitly. Your token should just get discovered upon first need.

For troubleshooting purposes, you can temporarily toggle a gargle option to see verbose output about the execution of gargle::token_fetch():

options(gargle_quiet = FALSE)

Provide an OAuth token directly

If you somehow have the OAuth token you want to use as an R object, you can provide it directly to the token argument of the main auth function. Example using googledrive:

gargle caches each OAuth user token it obtains to an .rds file, by default. If you know the filepath to the token you want to use, you could use readRDS() to read it and provide as the token argument to the wrapper’s auth function. Example using googledrive:

# googledrive
drive_auth(token = readRDS("/path/to/your/oauth-token.rds"))

How would you know this filepath? That requires some attention to the location of gargle’s OAuth token cache folder, which is described in the next section.

Full details are in the credentials_byo_oauth2() section of the gargle article How gargle gets tokens.

Arrange for an OAuth token to be re-discovered

This is the least recommended strategy, but it appeals to many users, because it doesn’t require creating a service account. Just remember that the perceived ease of using the token you already have (an OAuth user token) is quickly cancelled out by the greater difficulty of managing such tokens for non-interactive use. You might be forced to use this strategy with certain APIs, such as Gmail, that are difficult to use with a service account.

Two main principles:

  1. Take charge of – or at least notice – the folder where OAuth tokens are being cached.
  2. Make sure exactly one cached token will be identified and pre-authorize its use.

There are many ways to do this. We’ll work several examples using that convey the range of what’s possible.

I just want my .Rmd to render

Step 1: Get that first token. You must run your code at least once, interactively, do the auth dance, and allow gargle to store the token in its cache.

Step 2: Revise your code to pre-authorize the use of that token next time. Now your .Rmd can be rendered or your .R script can run, without further interaction.

You have two choices to make:

  • Set the gargle_oauth_email option or call PACKAGE_auth(email = ...).
    • The option-based approach can be implemented in each .Rmd or .R or in a user-level or project level .Rprofile startup file.
  • Authorize the use of the “matching token”, in general, (email is TRUE) or give a specific identity (email is an actual address).

This sets an option that allows gargle to use cached tokens whenever there’s a unique match:

options(gargle_oauth_email = TRUE)

This sets an option to use tokens associated with a specific email address:

options(gargle_oauth_email = "jenny@example.org")

This gets a token right now and allows the use of a matching token, using googledrive as an example:

This gets a token right now, for the user with a specific email address:

Project-level OAuth cache

This is like the previous example, but with an added twist: we use a project-level OAuth cache. This is good for deployed data products.

Step 1: Obtain the token intended for non-interactive use and make sure it’s cached in a (hidden) directory of the current project. Using googledrive as an example:

Do this setup once per project.

Another way to accomplish the same setup is to specify the desired cache location directly in the call to the auth function:

If you are doing setup in a web-based environment, such as RStudio Server, you may also need to request out-of-band auth, whenever you are first acquiring a token. That is a separate issue, which is explained in Auth when using R in the browser.

Step 2: In all downstream use, announce the location of the cache and pre-authorize the use of a suitable token discovered there. Continuing the googledrive example:

Setting the option gargle_oauth_email = TRUE says that googledrive is allowed to use a token that it finds in the cache, without interacting with a user, as long as it discovers EXACTLY one matching token. This option-setting code needs to appear in each script, .Rmd, or app that needs to use this token non-interactively. Depending on the context, it might be suitable to accomplish this in a startup file, e.g. project-level .Rprofile.

Here’s a variation where we say which token to use by explicitly specifying the associated email. This is handy if there’s a reason to have more than one token in the cache.

Here’s another variation where we specify the necessary info directly in an auth call, instead of in options:

Here’s one last variation that applicable when the local cache could contain multiple tokens:

Be very intentional about paths and working directory. Personally I would use here::here(".secrets)" everywhere above, to make things more robust.

For troubleshooting purposes, you can temporarily toggle a gargle option to see verbose output about the execution of gargle::token_fetch():

options(gargle_quiet = FALSE)

For a cached token to be considered a “match”, it must match the current request with respect to user’s email, scopes, and OAuth app (client ID or key and secret). By design, these settings have very low visibility, because we usually want to use the defaults. If your token is not being discovered, consider if any of these fields might explain the mismatch.