cluster and multiprocess with future and analogsea

cluster_multi_future.R

library(analogsea)
library(tidyverse)
library(furrr)

cluster_tag <- 'r_cluster' 
cluster_prefix <- 'node-'
number_of_nodes <- 2
# Path to private SSH key that matches key uploaded to DigitalOcean
# it looks like ~ does not work here... not sure why
ssh_private_key_file <- "/Users/jal/.ssh/id_rsa"

my_droplets <- docklets_create(names = paste0(cluster_prefix, as.character(1:number_of_nodes)),
                               region = "sfo2",
                               size = "16gb",
                               tags = cluster_tag)

# pull the ip addresses for the droplets
ips <- droplets(tag=cluster_tag) %>%
  map_chr(~.x$networks$v4[[1]]$ip_address) 

# Connect and create a cluster
cl <- makeClusterPSOCK(
  # vector of IPs in our cluster
  workers = ips,

  # DigitalOcean droplets use root for user
  user = "root",

  # use the key connected to Digital Ocean
  rshopts = c(
    "-o", "StrictHostKeyChecking=no",
    "-o", "IdentitiesOnly=yes",
    "-i", ssh_private_key_file
  ),

  # run Rscript in the tidyverse docker
  rscript = c("sudo","docker","run","--net=host","rocker/tidyverse","Rscript"),
  
  rscript_args = c(
  # Install furrr (future too) 
  "-e", shQuote("install.packages('furrr')")
  ), 
  verbose=TRUE
)

simulate_pi <- function(n_iterations) {
  rand_draws <- matrix(runif(2 * n_iterations, -1, 1),  ncol = 2)
  num_in <- sum(sqrt(rand_draws[,1]**2 + rand_draws[,2]**2) <= 1)
  pi_hat <- (num_in / n_iterations) * 4
  return(pi_hat)
}

library(tictoc)
draw_list <- as.list(rep(2500000, 400))


plan(cluster, workers = cl)

tic('simulate pi - parallel  - no multiprocess on nodes')
sims_list <- future_map(draw_list, simulate_pi)
toc()

plan(list(tweak(cluster, workers = cl), multiprocess))

tic('simulate pi - parallel - with multiprocess on nodes')
sims_list <- future_map(draw_list, simulate_pi)
toc()