Data Import and Export

Patrick Miller, Keith Ingersoll

2017-08-14

Passing Data Back and Forth

Often the simplest, but most useful operation when working with the Platform is to move data in and out. From the perspective of the R client, we call moving data from the Platform to the local machine reading. Likewise, moving data from the local machine to the Platform is called writing.

The civis client handles data imports and exports in two basic ways:

  1. Moving data directly between the R workspace and the Platform (the most common use case).
  2. Moving data between the Platform and local csv files (this is useful for large data that doesn’t fit into memory).

Data can be stored on Platform in two places:

  1. Amazon Redshift, a SQL database.
  2. Amazon S3, also referred to as the ‘files’ endpoint.

Tables in Redshift are accessed and modified using SQL queries. Tables in Redshift can be easily shared and used in multiple workflows by multiple people. However, importing and exporting even small files on Redshift can be slow.

R objects and arbitrary files can be stored on Amazon S3, and are accessed using a numeric file id. Data frames are uploaded as CSVs for portability, and arbitrary R objects are serialized using saveRDS for speed and efficiency.

Reading Data Into R From Platform

The main workhorse for getting data from Platform is read_civis. This function is designed to work similarly to the built in function read.csv, returning a dataframe from a table in Platform. For more flexibility, read_civis can download files from Redshift using an SQL query, or download a file from S3 (‘the files endpoint’) using a file id.

To read from a table in Platform, simply provide the name of the schema, table within the schema, and the database:

df <- read_civis("schema.tablename", database = "my-database")

For convenience, a default database can be set in the package options, and not specified in further calls to any IO function. If there is only one database available, this database will automatically be used as the default. In the examples that follow, we assume that a default database has been set.

options(civis.default_db = "my-database")
df <- read_civis("schema.tablename")

read_civis accepts SQL queries when more flexibility is needed. This is accomplished by wrapping sql(...) around a string containing the query. With read_civis, queries are always read only, and always return a data.frame.

query <- "SELECT * FROM table JOIN other_table USING id WHERE var1 < 23"
df <- read_civis(sql(query))

Finally, read_civis accepts a file id as the first argument to read in files from S3 as data frames. IDs are obtained from write_civis_file.

data(iris)
id <- write_civis_file(iris)
df <- read_civis(id)

For maximum flexibility, read_civis accepts parameters from read.csv which can be used to define data types when the defaults are not appropriate. For instance, when numbers should be read in as characters or when strings shouldn’t be read in as factors.

query <- "SELECT * FROM table JOIN other_table USING id WHERE var1 < 23"
df <- read_civis(sql(query), colClasses = "character")
df2 <- read_civis(sql(query), as.is = TRUE)

Uploading Data to a Database

The complement to reading data into the R workspace is writing data to the Platform. The function write_civis uploads data frames or csv files to an Amazon Redshift database. The function write_civis_file uploads R objects and arbitrary files to Amazon S3 (the files endpoint).

When creating a new table, write_civis relies on Platform to determine data types. Distkeys and sortkeys can optionally be set to improve query performance. Again, we set a default database in these examples for convenience.

options(civis.default_db = "my_database")
df <- data.frame(x = rnorm(100), y = rnorm(100), z = rnorm(100))
write_civis(df, tablename = "schema.tablename",
            distkey = "id", sortkey1 = "date", sortkey2 = "type")

By default, write_civis will fail if the table passed in tablename already exists. Optionally, write_civis can append to an existing table. It may also delete all rows and then append (truncate). If specific datatypes are required, a table may first be created with a SQL CREATE TABLE command and then data can be inserted with write_civis.

write_civis(df, tablename = "schema.tablename", if_exists = "append")
write_civis(df, tablename = "schema.tablename", if_exists = "truncate")

If a csv file is saved to disk but not loaded in the R workspace, write_civis will upload the csv to Platform without needing first load the csv into RAM. This can save time when a file is large. Uploading a csv directly to Platform is done by simply passing the file name and path to write_civis as the first argument:

write_civis("~/path/to/my_data.csv", tablename="schema.tablename")

Uploading Data to S3

Finally, write_civis_file uploads data frames, R objects and files to Amazon S3, which is also referred to as the ‘files endpoint.’ Data frames are uploaded as CSVs. R objects saved to the files endpoint and are serialized using saveRDS.

Data frames and R objects can be loaded back into memory by passing the file id to read_civis, and an appropriate using argument.

# Upload a data frame
data(iris)
id <- write_civis_file(iris)
iris2 <- read_civis(id)

# Upload an arbitrary R object
farm <- list(chickens = 1, ducks = 4, pigs = 2, cows = 1)
id <- write_civis_file(farm)
farm2 <- read_civis(id, using = readRDS)

When passed a file name and path, write_civis_file will upload the file to S3 as-is. To read the file back into memory, an appropriate function to convert the file to a data frame must be provided to the using argument of read_civis. For example, a JSON file can be read back into R using jsonlite::fromJSON.

id <- write_civis_file("path/to/my_data.json")
read_civis(id, using = jsonlite::fromJSON)

Downloading Large Data Sets from Platform.

Occasionally, a table may be too large to store in memory. download_civis can be used in place of read_civis to download data straight to disk from Platform.

Like read_civis, download_civis can download files from Amazon Redshift by passing schema.tablename, or sql(...) as the first argument. Files can be downloaded from Amazon S3 by passing the file id to download_civis.

query <- "SELECT * FROM table JOIN other_table USING id WHERE var1 < 23"
download_civis(sql(query), file = "path/to/my_file.csv")
download_civis("schema.tablename", file = "path/to/my_file.csv")

id <- write_civis_file(iris)
download_civis(id, file = "path/to/my_iris.rds")

Running Queries on Platform

Arbitrary queries can be run on Redshift using query_civis, which returns the meta-data of the query.

q_res <- query_civis("GRANT ALL ON schema.my_table TO GROUP admin")

Existing queries can be re-run by passing the query id to query_civis:

id <- q_res$id
query_civis(id)

Common Errors

Civis API key not properly set or has expired.

Often an improper API key will return an error like below:

 Error in api_key() : 
  The environmental variable CIVIS_API_KEY is not set. Add this to your .Renviron or call Sys.setenv(CIVIS_API_KEY = '<api_key>') 

However, there may be cases where the errors are less straightforward. It is a good idea to test that API credentials are properly set with a simple call such as civis::users_list_me(). See the README to set up API keys correctly.

Query does not return any results.

This may happen if a table is empty or when no rows match a WHERE statement. To fix, double check that the query is correct or the table is not empty.

read_civis(sql("SELECT * FROM schema.tablename WHERE 1 = 0"))
Error in download_script_results(run$script_id, run$run_id) : 
  Query produced no output. 

Database not set correctly.

For both read_civis and write_civis, the database must be set to the correct, case sensitive name (not hostname) of the database.

 Error in get_db(database) : 
  Argument database is NULL and options("civis.default_db") not set. Set this option using options(civis.default_db = "my_database") 

To see a complete list of database names, run:

sapply(databases_list(), function(x) x$name)