bank

The goal of {bank} is to provide alternative backends for caching with {memoise} & {shiny}.

Installation

# install.packages("remotes")
remotes::install_github("thinkr-open/bank")

About

You’re reading the doc about version : 0.0.0.9002

This README has been compiled on the

Sys.time()
#> [1] "2023-03-27 14:58:39 CEST"

Here are the test & coverage results :

devtools::check(quiet = TRUE)
#> ℹ Loading bank
#> ── R CMD check results ──────────────────────────────────── bank 0.0.0.9002 ────
#> Duration: 40.3s
#> 
#> 0 errors ✔ | 0 warnings ✔ | 0 notes ✔

covr::package_coverage()
#> bank Coverage: 68.00%
#> R/postgres.R: 60.87%
#> R/redis.R: 66.67%
#> R/mongo.R: 81.01%

Some things to know before starting

Caching scope

When using {bank} backends with {shiny}, caching will done at the app-level, in other words the cache is stored across sessions. Be aware of this behavior if you have sensitive data inside your app, as this might imply data leakage.

See ?shiny::bindCache

With an app-level cache scope, one user can benefit from the work done for another user’s session. In most cases, this is the best way to get performance improvements from caching. However, in some cases, this could leak information between sessions. For example, if the cache key does not fully encompass the inputs used by the value, then data could leak between the sessions. Or if a user sees that a cached reactive returns its value very quickly, they may be able to infer that someone else has already used it with the same values.

Cache flushing

As with any {cachem} compatible objects, the cache can be manually flushed using the $reset() method – this will call drop() on MongoDb, FLUSHALL in Redis, & DBI::dbRemoveTable() + DBI::dbCreateTable() with Postgres.

library(bank)
mongo_cache <- cache_mongo$new(
  db = "bank",
  url = "mongodb://localhost:27066",
  prefix = "sn"
)

mongo_cache$reset()

As {bank} relies on external backends, it’s probably better to let the DBMS handle the flushing of cache. For example, in redis.conf, you can set :

maxmemory 2mb
maxmemory-policy allkeys-lru

LRU (least recently used) will allow redis to flush the key based on when they were used. See https://redis.io/topics/lru-cache.

MongoDB doesn’t come with a LRU mechanism, but you can set data to be ephemeral with TTL index inside your collection.

{bank} also tries to help with that by updating a lastAccessed date metadata field in Mongo whenever you $get() the key, meaning that you can implement your own caching strategy to evict least recently used cached objects.

Postgre limitation

Postgre bytea column can only store up to 1GB elements, so you can’t write a cache that’s > 1GB.

Backends

Note that if you want to use {bank} in a {shiny} app:

renderCachedPlot() require {shiny} version 1.5.0 or higher
bindCache() require {shiny} version 1.6.0 or higher

For now, the following backends are supported:

MongoDB
Redis
Postgre

Mongo

Launching a container with mongo.

docker run --rm --name mongobank -d -p 27066:27017 -e MONGO_INITDB_ROOT_USERNAME=bebop -e MONGO_INITDB_ROOT_PASSWORD=aloula mongo:4

With `{memoise}`

First, the cache_mongo can be used

library(memoise)
library(bank)
# Create a mongo cache.
# The arguments will be passed to mongo::gridfs
mongo_cache <- cache_mongo$new(
  db = "bank",
  url = "mongodb://bebop:aloula@localhost:27066",
  prefix = "sn"
)
#> Loading required namespace: mongolite

f <- function(x) {
  sample(1:1000, x)
}

mf <- memoise(f, cache = mongo_cache)
mf(5)
#> [1]  31 609 671 766 219
mf(5)
#> [1]  31 609 671 766 219

Inside `{shiny}`

Here is a first simple application that shows you the basics :

library(shiny)
ui <- fluidPage(
  # Creating a slider input that will be used as a cache key
  sliderInput("nrow", "NROW", 1, 32, 32),
  # Plotting a piece of mtcars
  plotOutput("plot")
)

server <- function(input, output, session) {
  output$plot <- renderCachedPlot(
    {
      # Pretending this takes a long time
      Sys.sleep(2)
      plot(mtcars[1:input$nrow, ])
    },
    cacheKeyExpr = list(
      # Defining the cache key
      input$nrow
    ),
    # Using our mongo cache
    cache = mongo_cache
  )
}
shinyApp(ui, server)

As you can see, the first time you set the slider to a given value, it takes a little bit to compute. Then it’s almost instantaneous.

Let’s try a more complex application:

# We'll put everything in a function so that it can later be reused with other backends
library(magrittr)

generate_app <- function(cache_object) {
  ui <- fluidPage(
    h1(
      sprintf(
        "Caching in an external DB using %s",
        deparse(
          substitute(cache_object)
        )
      )
    ),
    sidebarLayout(
      sidebarPanel = sidebarPanel(
        # This sliderInput will be the cache key
        # i.e we don't want to recompute the plot everytime
        sliderInput("nrow", "Nrow", 1, 32, 32),
        # Allow to clear the cache
        actionButton("clear", "Clear Cache")
      ),
      mainPanel = mainPanel(
        # Outputing the reactive and a plot
        verbatimTextOutput("txt"),
        plotOutput("plot"),
        # If you care about listing the cache keys
        uiOutput("keys")
      )
    )
  )

  server <- function(input,
                     output,
                     session) {

    # Our plot, cached using the cache object and
    # watching the nrow
    output$plot <- renderCachedPlot(
      {
        showNotification(
          h2("I'm computing the plot"),
          type = "message"
        )

        # Fake long computation
        Sys.sleep(2)

        # Plot
        plot(mtcars[1:input$nrow, ])
      },
      # We cache on the input$nrow
      cacheKeyExpr = list(
        input$nrow
      ),
      # The cache object is used here
      cache = cache_object
    )

    rc <- reactive({
      showNotification(
        h2("I'm computing the reactive()"),
        type = "message"
      )
      # Fake long computation
      Sys.sleep(2)

      input$nrow * 100
    }) %>%
      # Using bindCache() require shiny > 1.6.0
      bindCache(
        input$nrow,
        cache = cache_object
      )

    output$txt <- renderText({
      rc()
    })

    keys <- reactive({
      # Listing the keys
      invalidateLater(500)
      cache_object$keys()
    })

    output$keys <- renderUI({
      tags$ul(
        lapply(keys(), tags$li)
      )
    })

    observeEvent(input$clear, {
      # Sometime you might want to remove everything from the cache
      cache_object$reset()

      showNotification(
        h2("Cache reset"),
        type = "message"
      )
    })
  }

  shinyApp(ui, server)
}

generate_app(mongo_cache)

Flushing MongoDB cache using LRU

All keys registered to MongoDB comes with a metadata.lastAccessed parameter. Using this parameter, you’ll be able to flush old cache if needed.

mongo <- mongolite::gridfs(
  db = "bank",
  url = "mongodb://bebop:aloula@localhost:27066",
  prefix = "sn"
)
get_metadata <- function(mongo) {
  purrr::map(mongo$find()$metadata, jsonlite::fromJSON)
}
Sys.sleep(10)
mf(5)
#> [1]  31 609 671 766 219
get_metadata(mongo)
#> [[1]]
#> [[1]]$key
#> [1] "116acf5d3c7188709a0374305ba3a33747ef5ce323f3e170862551ed523d7a425658d2fb50d11a557250935326669e0a37efb90205d2ba114795359bb55234ca"
#> 
#> [[1]]$lastAccessed
#> [1] "2023-03-27 15:00:26"

Sys.sleep(10)
mf(5)
#> [1]  31 609 671 766 219
get_metadata(mongo)
#> [[1]]
#> [[1]]$key
#> [1] "116acf5d3c7188709a0374305ba3a33747ef5ce323f3e170862551ed523d7a425658d2fb50d11a557250935326669e0a37efb90205d2ba114795359bb55234ca"
#> 
#> [[1]]$lastAccessed
#> [1] "2023-03-27 15:00:36"

Redis

Launching a container with redis.

docker run --rm --name redisbank -d -p 6379:6379 redis:5.0.5 --requirepass bebopalula

With `{memoise}`

# Create a redis cache.
# The arguments will be passed to redux::hiredis
redis_cache <- cache_redis$new(password = "bebopalula")
#> Loading required namespace: redux

f <- function(x) {
  sample(1:1000, x)
}

mf <- memoise(f, cache = redis_cache)
mf(5)
#> [1] 680 901 873 651 605
mf(5)
#> [1] 680 901 873 651 605

Inside `{shiny}`

Here is a first simple application that shows you the basics :

ui <- fluidPage(
  # Creating a slider input that will be used as a cache key
  sliderInput("nrow", "NROW", 1, 32, 32),
  # Plotting a piece of mtcars
  plotOutput("plot")
)

server <- function(input, output, session) {
  output$plot <- renderCachedPlot(
    {
      # Pretending this takes a long time
      Sys.sleep(2)
      plot(mtcars[1:input$nrow, ])
    },
    cacheKeyExpr = list(
      # Defining the cache key
      input$nrow
    ),
    # Using our redis cache
    cache = redis_cache
  )
}
shinyApp(ui, server)

For the larger app:

generate_app(redis_cache)

Postgres

Launching a container with postgres.

docker run --rm --name some-postgres -e POSTGRES_PASSWORD=mysecretpassword -d -p 5433:5432 postgres

With `{memoise}`

# Create a postgres cache.
# The arguments will be passed to DBI::dbConnect(RPostgres::Postgres(), ...)
postgres_cache <- cache_postgres$new(
  dbname = "postgres",
  host = "localhost",
  port = 5433,
  user = "postgres",
  password = "mysecretpassword"
)
#> Loading required namespace: RPostgres

f <- function(x) {
  sample(1:1000, x)
}

mf <- memoise(f, cache = postgres_cache)
mf(5)
#> [1] 758 764 404 689 557
mf(5)
#> [1] 758 764 404 689 557

Inside `{shiny}`

Here is a first simple application that shows you the basics :

ui <- fluidPage(
  # Creating a slider input that will be used as a cache key
  sliderInput("nrow", "NROW", 1, 32, 32),
  # Plotting a piece of mtcars
  plotOutput("plot")
)

server <- function(input, output, session) {
  output$plot <- renderCachedPlot(
    {
      # Pretending this takes a long time
      Sys.sleep(2)
      plot(mtcars[1:input$nrow, ])
    },
    cacheKeyExpr = list(
      # Defining the cache key
      input$nrow
    ),
    # Using our postgres cache
    cache = postgres_cache
  )
}
shinyApp(ui, server)

For the larger app:

generate_app(postgres_cache)

Chosing a cache method

Benchmark

As we are deporting the caching to an external DBMS, the query will of course be slower than using memory cache of disk cache. But this difference in speed comes with a simpler scalability of the caching, as several instances of the app can rely on the same caching backend without the need to be on the same machine.

library(magrittr)
library(bank)
big_iris <- purrr::rerun(100, iris) %>% data.table::rbindlist()
#> Warning: `rerun()` was deprecated in purrr 1.0.0.
#> ℹ Please use `map()` instead.
#>   # Previously
#>   rerun(100, iris)
#> 
#>   # Now
#>   map(1:100, ~iris)
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
#> generated.

nrow(big_iris)
#> [1] 15000
pryr::object_size(big_iris)
#> 542.11 kB

library(cachem)

mem_cache <- cache_mem()

disk_cache <- cache_disk()

mongo_cache <- cache_mongo$new(
  db = "bank",
  url = "mongodb://bebop:aloula@localhost:27066",
  prefix = "sn"
)

redis_cache <- cache_redis$new(password = "bebopalula")

postgres_cache <- cache_postgres$new(
  dbname = "postgres",
  host = "localhost",
  port = 5433,
  user = "postgres",
  password = "mysecretpassword"
)

bench::mark(
  mem_cache = mem_cache$set("iris", big_iris),
  disk_cache = disk_cache$set("iris", big_iris),
  mongo_cache = mongo_cache$set("iris", big_iris),
  redis_cache = redis_cache$set("iris", big_iris),
  postgres_cache = postgres_cache$set("iris", big_iris),
  check = FALSE,
  iterations = 100
)
#> # A tibble: 5 × 6
#>   expression          min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>     <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 mem_cache        8.04µs   9.18µs   94621.     2.23KB   956.  
#> 2 disk_cache        3.3ms    3.4ms     283.    16.19KB     0   
#> 3 mongo_cache       9.6ms  13.02ms      71.1    2.52MB     4.54
#> 4 redis_cache       4.6ms   5.69ms     167.   536.58KB     1.68
#> 5 postgres_cache   1.42ms   1.93ms     419.   626.09KB     0

bench::mark(
  mem_cache = mem_cache$get("iris"),
  disk_cache = disk_cache$get("iris"),
  mongo_cache = mongo_cache$get("iris"),
  redis_cache = redis_cache$get("iris"),
  postgres_cache = postgres_cache$get("iris"),
  iterations = 100
)
#> # A tibble: 5 × 6
#>   expression          min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>     <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 mem_cache        6.11µs   6.52µs  144938.         0B     0   
#> 2 disk_cache     556.12µs 605.16µs    1503.   528.62KB    15.2 
#> 3 mongo_cache      7.85ms  10.01ms      88.3    2.06MB     6.64
#> 4 redis_cache      3.68ms   5.17ms     182.     1.03MB     3.71
#> 5 postgres_cache   3.05ms   3.85ms     194.   566.62KB     3.97

docker stop mongobank redisbank postgresbank

You want another backend?

If you have any other backend in mind, feel free to open an issue here and we’ll discuss the possibility of implementing it in {bank}.

Code of Conduct

Please note that the bank project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

ThinkR-open / bank

readme

bank

Installation

About

Some things to know before starting

Caching scope

Cache flushing

Postgre limitation

Backends

Mongo

With `{memoise}`

Inside `{shiny}`

Flushing MongoDB cache using LRU

Redis

With `{memoise}`

Inside `{shiny}`

Postgres

With `{memoise}`

Inside `{shiny}`

Chosing a cache method

Benchmark

You want another backend?

Code of Conduct

ThinkR-open / bank

readme

bank

Installation

About

Some things to know before starting

Caching scope

Cache flushing

Postgre limitation

Backends

Mongo

With {memoise}

Inside {shiny}

Flushing MongoDB cache using LRU

Redis

With {memoise}

Inside {shiny}

Postgres

With {memoise}

Inside {shiny}

Chosing a cache method

Benchmark

You want another backend?

Code of Conduct

With `{memoise}`

Inside `{shiny}`

With `{memoise}`

Inside `{shiny}`

With `{memoise}`

Inside `{shiny}`