Splitting a data frame into chunk sizes of X to pass to an API

I have a data frame of values which varies in size each week when I run my code but it's typically ~3000 rows

I need to pass each of these values into a function I've created (a whole world of pain for me!) to get data back from an API.

I've found that I get an "parse error: premature EOF" error if I pass too many values to the API at once so I've broken the original data frame of values into chunks of size 500 (not a good idea, I know).

Is there a more elegant solution to passing all ~3000 values to the API so that I don't overload the API and get a HTTP 429 message?

Here is a reprex to help explain the problem, however in this example lets assume the API limit is 2 and I have 10 values i wish to pass to the API call.

install.packages("pacman")
library(pacman)

pacman::p_load(tidyverse,data.table,httr,jsonlite)

values <- c(1598727430,
 1083632731,
 1710983663,
1033159769,
1659312593,
1720037021,
1538669445,
1508835828,
1033154026,
1003819475,
1811299944) 

df <- data.frame(values)

Here is the function that I created:

# URL OF THE NPI API
path <- "https://npiregistry.cms.hhs.gov/api/?"

# CREATE A FUNCTION TO PULL NPI INFORMATION FROM THE NPI REGISTRY
getNPI <- function(object) {
  request <- httr::GET(
    url = path,
    query = list(
      version = "2.0",
      number = object
    )
  )
  Sys.sleep(0.25)

  warn_for_status(request)

  npi_details <- content(request,
    as = "text",
    encoding = "UTF-8"
  ) %>%
    fromJSON(.,
      flatten = TRUE
    ) %>%
    data.frame()
  # IF THE API THROWS BACK A RESULT WHERE THE COLUMN NAMES CONTAIN 'ERROR'
  # THEN ASSIGN ALL THE ROW VALUES TO NA AND ADD THE NPI VALUE TO THE FIRST
  # COLUMN
  if (any(grepl("ERROR", toupper(colnames(npi_details))))) {
    npi_details <- as.data.frame(matrix("error", ncol = 6, nrow = 1), stringsAsFactors = FALSE) %>%
      dplyr::rename(
        `NPI NUMBER` = V1,
        `CMS REF ADDRESS 1` = V2,
        `CMS REF ADDRESS 2` = V3,
        `CMS REF CITY` = V4,
        `CMS REF STATE` = V5,
        `CMS REF ZIP` = V6
      ) %>% as_tibble()
    
    npi_details[1,1] <- as.character(object)
    return(npi_details)

    # ELSE IF THE DATA FRAME DOES NOT CONTAIN 'ERROR' THEN RUN THIS CHUNK
  } else {
    select(npi_details, contains(c("addresses", "number"))) %>%
      unnest(c(contains("address"))) %>%
      filter(address_purpose == "MAILING") %>%
      rename_all(.funs = toupper) %>%
      select(
        `NPI NUMBER` = RESULTS.NUMBER,
        -COUNTRY_CODE,
        -COUNTRY_NAME,
        -ADDRESS_PURPOSE,
        -ADDRESS_TYPE,
        `CMS REF ADDRESS 1` = ADDRESS_1,
        `CMS REF ADDRESS 2` = ADDRESS_2,
        `CMS REF CITY` = CITY,
        `CMS REF STATE` = STATE,
        `CMS REF ZIP` = POSTAL_CODE
      ) %>% 
      mutate(`NPI NUMBER` = as.character(`NPI NUMBER`))
  }
}

Right now this is how I run my function against my values:

# apply the getNPI function to each chunk of the data frame
data1_2 <- apply(df[1:2,], 1, function(x) list(getNPI(x)))
data3_4 <- apply(df[3:4], 1, function(x) list(getNPI(x)))
....
data9_10 <- apply(df[9:10], 1, function(x) list(getNPI(x)))

# convert the list of lists to a single list
npi_list <- do.call(c,list(data1_2 ,
                           data3_4 ,
                           data5_6 ,
                           data7_8 ,
                           data9_10))

# convert data table
dt_list <- map(npi_list, as.data.table)

# bind them together to get a data frame
provider_npi_data <- rbindlist(dt_list, fill = TRUE, idcol = FALSE) %>% 
  as.data.frame() %>% 
  drop_na()

These could probably be replaced by a single purrr::map() call or slider::slide()

dequer implements a facility to creating a stack from which items can be popped for dispatch to the api, and it should not be too difficult to do a loop to do a pop/wait cycle with tunable arguments to n for pop.

@nirgrahamuk Can you explain how the slider function can be used in the example above? I have tried a number of variants of slide such as:

output <- slider::slide_vec(.x = examples, .f = getNPI, .before = 2,.complete = T)

output <- slide::slide_dfr(.x = examples, .f = getNPI, .before = 2,.complete = T)

purrr::map_dfr(df$values,getNPI)

I guess I don't understand why you had multiple applies in the first place. Its not really doing any chunking/throttling for you per say , at least if your script is run end to end in one shot.
presumably the throttling is happening in your sleep, and that happens for each getNPI regardless of each you do 2 getNPIs and then another 2 getNPI's how is that different to doing 4 gets given your setup ?
please let me know what I've missed.