Push to Prod

Level Up with Shiny for R

posit::conf(2024)

2024-08-12

Deploying Shiny Apps

Let’s Deploy Our App!

We’re going to start by deploying _exercises/15_deploy/15_01_app.R.

First, we’ll just get it onto Connect.

Then, we’ll switch to a programmatic deployment using rsconnect::deployApp().

We’ll work together

I’ll walk through each step in the process. Choose your adventure:

1. Follow along and work through this with me

2. Just watch and absorb

I’ll work in the same file, but there are snapshot of the progress we’ll make, e.g. 15_02_app.R, etc.

Connect with Connect

The easiest way to deploy your app is with RStudio’s Publish App button, right next to the Run App button.

Follow the instructions to get connected with Posit Connect.

Logging

Now that other people can use it

…they’re probably doing something you didn’t expect.

Find the logs of the app we just deployed.

What happens when things go wrong?

Share your app with a small group first. Someone who will tell you what they think and reach out if the app breaks.

What happens when an app breaks in the middle of a user session?

Your Base-R Options are limited

x <- 42
cat("The value of x is", x)

The value of x is 42

message("x is now ", x)

x is now 42

warning("x is ", x, " and that might be bad")

Warning: x is 42 and that might be bad

stop("x cannot be ", x)

Error in eval(expr, envir, enclos): x cannot be 42

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Log Levels

Intro to lgr

library(lgr)

lgr$info("Loading data from customers.csv")

INFO  [17:16:32.792] Loading data from customers.csv

lgr$warn("Missing values detected in column `%s`", "salary")

WARN  [17:16:32.824] Missing values detected in column `salary`

lgr$error("Model validation failed", accuracy = 0.459)

ERROR [17:16:32.831] Model validation failed {accuracy: `0.459`}

lgr$fatal("object of type 'closure' is not subsettable")

FATAL [17:16:32.832] object of type 'closure' is not subsettable

Appenders - glue messages

INFO  [17:16:32.966] Loading data from customers.csv

WARN  [17:16:32.997] Missing values detected {columns: (salaries, age)}

ERROR [17:16:32.999] Model validation failed {accuracy: `0.459`, threshold: `0.5`}

FATAL [17:16:33.000] object of type 'closure' is not subsettable

Without logging

Warning: Error in dispatch: Invalid map parameter
  100: expandLimits
   98: addMarkers
   97: ::
htmlwidgets
shinyRenderWidget [/Users/garrick/teach/level-up-shiny/_examples/12-deploy/02_log_app.R#247]
# ... more stuff

Task: Add logging to the app to diagnose this error.

With logging

DEBUG [10:04:39.334] Rendering map with color mode {color_mode: `light`}
DEBUG [10:04:39.357] Choosing OpenStreetMap.Mapnik tiles
DEBUG [10:04:42.767] Rendering map with color mode {color_mode: `dark`}
Warning: Error in dispatch: Invalid map parameter
  100: expandLimits
   98: addMarkers
   97: ::
htmlwidgets
shinyRenderWidget [/Users/garrick/teach/level-up-shiny/_examples/12-deploy/02_log_app.R#247]

See it in action

Leads to fixed app

Task: Fix and re-deploy the app.

The problem is that input$color_mode() returns "light" or "dark", not "night".

_exercises/15_deploy/15_02_app.R

Config

Overview of {config}

config.yml

default:
  log_level: debug

rsconnect:
  log_level: info

Local

config::get("log_level")

[1] "debug"

Posit Connect

config::get("log_level")

[1] "info"

https://rstudio.github.io/config/

A quick word about environment variables

Have you ever used environment variables? Set GITHUB_PAT somewhere? Or another secret or token?

Envvvars are like options…

Except they are…

• More system-wide than within-process
• Often be set by the controlling process
• Vey-value pairs, but values are always character
• Stickier

How to use environment variables

• Sys.getenv(name, default) to get the value of an envvar
• edit_r_environ() to edit the .Renviron file
• Common to store secrets in .env files
  • Load with {dotenv}
  • Don’t check them into version control

What’s the value of the R_CONFIG_ACTIVE envvar?

Common Posit config names

Service	R_CONFIG_ACTIVE
Posit Connect	rsconnect
Post Cloud	rstudio_cloud
shinyapps.io	shinyapps
Shinylive	shinylive

Task: Set log level

• Use config to set the log level in the app.
• Use a second envvar to override the default log level.
• Deploy the app and test the envvar override.

_exercises/15_deploy/15_03_app.R

Databases

Get Connected

You: Hi, how do I get connected to the database?

IT: Sure thing, here’s the connection string.

postgres://dbuser:$3cur3C0d3!@ec2-54-123-45-67.compute-1.amazonaws.com:5432/mydatabase

host: 
port: 
dbname: 
user: 
password: 

Hello, Boilerplate

con <- DBI::dbConnect(
  RPostgres::Postgres(),
  host = "ec2-54-123-45-67.compute-1.amazonaws.com",
  port = 5432,
  dbname = "mydatabase",
  user = "dbuser",
  password = "$3cur3C0d3!"
)

Hello, {dotenv}

db.env

DB_HOST=ec2-54-123-45-67.compute-1.amazonaws.com
DB_PORT=5432
DB_DATABASE=mydatabase
DB_USER=dbuser
DB_PASSWORD=$3cur3C0d3!

dotenv::load("db.env")

con <- DBI::dbConnect(
  RPostgres::Postgres(),
  host = Sys.getenv("DB_HOST"),
  port = Sys.getenv("DB_PORT"),
  dbname = Sys.getenv("DB_DATABASE"),
  user = Sys.getenv("DB_USER"),
  password = Sys.getenv("DB_PASSWORD")
)

Task: Use the database

• Get credentials from secrets/db-staging.env or secrets/db-prod.env
• Preview score/scorecard tables at the start of the app
• Don’t forget to close the connection
• Add logging throughout the app

• Some updates:
  • arrange() –> window_order()
  • use !! in the between() calls
  • collect() the data in the schools reactive
  • maybe we could cache the data at this point?

_exercises/15_deploy/15_04_app.R

In the project

library(dplyr)

staging_env <- here::here("secrets/db-staging.env")
prod_env <- here::here("secrets/db-prod.env")

dotenv::load_dot_env(staging_env)

con <- DBI::dbConnect(
  RPostgres::Postgres(),
  host = Sys.getenv("DB_HOST"),
  port = Sys.getenv("DB_PORT"),
  dbname = Sys.getenv("DB_DATABASE"),
  user = Sys.getenv("DB_USER"),
  password = Sys.getenv("DB_PASSWORD")
)

tbl(con, "school")
tbl(con, "scorecard")

Now that it works

Make life easier for future you

renv

Is it working now with these packages? Good, lock it down. This is a good time for renv.

renv::init()
# => renv.lock

Later, or on someone else’s machine:

renv::restore()  # restore the same packages
renv::snapshot() # update the lockfile

https://rstudio.github.io/renv/

shinytest2

Now that it’s working, let’s test it.
Does that sound backwards?

Testing shiny apps isn’t like testing package code.

Your tests are most helpful when they keep you from breaking something that was working without knowing it.

https://rstudio.github.io/shinytest2/

shinytest2

First, record the test.

shinytest2::record_test("app-dir/")

Then, or much later, run the test.

shinytest2::test_app("app-dir/")

Frameworks