Introduction

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
library(GCSConnection)
gcs_cloud_auth(json_file = NULL)

Credentials

Auto authentication

You need to have a service account credentials to authenticate with Google Cloud Storage API. The credentials determines which data sources are accessable to you. Please follow the instructions on Google Authentication to create your credentials file and download it to your local machine(Note that it can be done in the Cloud Console). The package will search for the credentials file from the environment variable GOOGLE_APPLICATION_CREDENTIALS when it is loaded. If it fails to find the credentials, it will look for the environment variable GCS_AUTH_FILE instead. If both environment variables are empty, you will only be able to access public data. If you set the environment variables after the package is loaded, you can rerun the auto authentication by gcs_cloud_auth().

Manual authentication

The function gcs_cloud_auth provides two ways to authenticate with Google Cloud. When you have a service account credentials, you can authenticate with Google Cloud by calling gcs_cloud_auth(credentials_file_path). The second method is to get credentials from Cloud SDK. The package is able to interact with the command-line tool gcloud. Once the account has been set in the gcloud program, the package can be authenticated by gcs_cloud_auth(gcloud = TRUE). By default, the package will use the first account in gcloud to authenticate with Google. If there are multiple accounts in gcloud, you can switch the account by gcs_cloud_auth(gcloud = TRUE, email = "your email address").

Anouymous access

Not all the cloud buckets require a credentials to access. You can read public data without authenticating with Google. By default, if the package fails to find the credentials from the environment variables, it will use anouymous access method to query data from google cloud. Otherwise, you can switch to the anouymous mode by calling gcs_cloud_auth(json_file = NULL).

Once the credentials is set, you can check your token and authenticate type via:

gcs_get_cloud_auth()

Manually create connections

Creating a connection to a file in a bucket is simple, you can simply provide an URL to gcs_connection. Below is an example to create a read connection to a public dataset on Google Cloud Platform.

## equivalent:
## file <- "genomics-public-data/NA12878.chr20.sample.DeepVariant-0.7.2.vcf"
file <- "gs://genomics-public-data/NA12878.chr20.sample.DeepVariant-0.7.2.vcf"
con <- gcs_connection(description = file, open = "r")
readLines(con, n = 2L)
close(con)

Note that you do not need a credentials for accessing the public data. the character "r" in the parameter open is an abbreviation of read, please see ?gcs_connection for more details.

It is also possible to specify file and bucket name separately. For example:

file_name <- "NA12878.chr20.sample.DeepVariant-0.7.2.vcf"
bucket_name <- "genomics-public-data"
con <- gcs_connection(
    description = file_name, open = "r", bucket = bucket_name
)
readLines(con, n = 2L)
close(con)

The above code can create the same connection as the previous example. Note that you cannot provide both URI and bucket name to gcs_connection.

For the write connection, it can be made by specifying an appropriate open mode in the gcs_connection function. Please note that due to the limitation of the Google Cloud Storage, the write connection is not seekable, which means you cannot use the seek function to

navigate through the file. When the write connection is created. It always starts in the beginning of the file. If there exists a file with the same name, the old file will be deleted. After the write connection is closed, the file will become immutable and no further change on the file can be made.

create connections from the file manager

Besides manually creating a connection, the package provides a simple and convinent S4 class to manage files. You can list all files in a bucket/folder by

## These are equivalent
## x <- gcs_dir("gs://genomics-public-data/clinvar/")
## x <- gcs_dir("genomics-public-data/clinvar/")

x <- gcs_dir("gs://genomics-public-data/clinvar/")
x

Note that for listing files in a bucket, the trailing slash does not make any differences. However, if you want to list files in a folder inside a bucket, it is recommended to explicitly add a / at the end. The trailing slash is not mandatory, but if no trailing slash presents, the package need one extra communication with the cloud to determine whether the path is a file or a folder, which can double your time cost.

In fact, there is no hierarchical structure in a bucket, all files are in the same level. By using the slash delimiter, the cloud makes the files appear as though they are stored in folders. Though the cloud does not have any restriction on the use of / in the file name, a good practice is to not use / at the end of a file name for it can cause unnecessary confusion.

Once obtaining a list of files, you can changes your current directory or view the detail of a file through $ or [[ operator

## equivalent: x$README.txt
myfile <- x[["README.txt"]]
myfile

You can also use .. to go to the parent folder, ~ to go to the bucket root.

## equivalent: myfile$`..`
myfile[[".."]]

## myfile$`~`
myfile[["~"]]

The connection can be made by

## Equivalent: gcs_connection(myfile)
con <- myfile$get_connection(open = "r")
con
close(con)

Note that you can pass a file object to the function gcs_connection. The function gcs_cp supports both file and folder objects. You can query file info, copy, delete files through via the file manager

## Get file name
myfile$file_name

## copy file
## For the destination, you can specify a path to the file, 
## or a path to a folder.
destination <-tempdir()
myfile$copy_to(destination)
file.exists(file.path(destination,myfile$file_name))

## Delete file, the function is excutable 
## only when you have the right to delete the file.
## Use `quiet = TRUE` to suppress the confirmation before deleting the file.
# x$README.txt$delete(quiet = FALSE)

Note that you cannot delete a folder since beforementioned file hierarchy issue. Once all files in a folder has been deleted, the folder will be removed.

Requester pays

Some buckets have requester pays enabled, which means you are responsible for the cost of accessing the data in the bucket. Therefore, you must have a valid billing project for reading the data in the bucket. By default, if you use JSON file to authenticate with Google Cloud, the billing project is the project listed in the JSON file, which is the project that you use to generate the JSON file. If you use gcloud, the billing project is the default project in your config file. You can view and change the default billing project via

gcs_get_billing_project()
gcs_set_billing_project("your project ID")

For accessing the data in the bucket, you also need to pass the argument billing_project = TRUE to the gcs functions that you want to use. For example, if uri is a path to a file in a requester pays bucket, you can view its information via gcs_dir(uri, billing_project = TRUE). Please keep in mind that you will get charged if you send a billing project to a bucket which do not have requester pays enabled. Therefore, the default behavior of all gcs functions is to access the bucket without a billing project. If you want your billing project being involved in every gcs function, you can alter this setting by

gcs_set_requester_pays(TRUE)
gcs_get_requester_pays()

Buffer size

For reducing the number of network requests and speeding up the performance of the connection, there is a buffer associated with each connection. The default size is 1Mb for a buffer. You can set or get the buffer size via gcs_read_buff, gcs_write_buff, gcs_get_read_buff and gcs_get_write_buff functions:

gcs_get_read_buff()
gcs_get_write_buff()

Please note the minimum buffer size for a write connection is 256 Kb. Creating a connection with small buffer size may impact the performance.



Try the GCSConnection package in your browser

Any scripts or data that you put into this service are public.

GCSConnection documentation built on Nov. 8, 2020, 5:30 p.m.