Introduction to arkdb

Carl Boettiger

2021-04-05

arkdb

Package rationale

Increasing data sizes create challenges for the fundamental tasks of publishing, distributing, and preserving data. Despite (or perhaps because of) the diverse and ever-expanding number of database and file formats, the humble plain text file such as comma or tab-separated-values (e.g. .csv or .tsv files) remains the gold standard for data archiving and distribution. These files can read on almost any platform or tool and can be efficiently compressed using long-standing and widely available standard open source libraries like gzip or bzip2. In contrast, database storage formats and dumps are usually particular to the database platform used to generate them, and will likely not be compatible between different database engines (e.g. PostgreSQL -> SQLite) or even between different versions of the same engine. Researchers unfamiliar with these databases will have difficulty accessing such data, and these dumps may also be in formats that are less efficient to compress.

Working with tables that are too large for working memory on most machines by using external relational database stores is now a common R practice, thanks to ever-rising availability of data and increasing support and popularity of packages such as DBI, dplyr, and dbplyr. Working with plain text files becomes increasingly difficult in this context. Many R users will not have sufficient RAM to simply read in a 10 GB .tsv file into R. Similarly, moving a 10 GB database out of a relational data file and into a plain text file for archiving and distribution is similarly challenging from R. While most relational database back-ends implement some form of COPY or IMPORT that allows them to read in and export out plain text files directly, these methods are not consistent across database types and not part of the standard SQL interface. Most importantly for our case, they also cannot be called directly from R, but require a separate stand-alone installation of the database client. arkdb provides a simple solution to these two tasks.

The goal of arkdb is to provide a convenient way to move data from large compressed text files (e.g. *.tsv.bz2) into any DBI-compliant database connection (see DBI), and move tables out of such databases into text files. The key feature of arkdb is that files are moved between databases and text files in chunks of a fixed size, allowing the package functions to work with tables that would be much to large to read into memory all at once. This will be slower than reading the file into memory at one go, but can be scaled to larger data and larger data with no additional memory requirement.

Installation

You can install arkdb from GitHub with:

# install.packages("devtools")
devtools::install_github("cboettig/arkdb")

Tutorial

library(arkdb)

# additional libraries just for this demo
library(dbplyr)
library(dplyr)
library(nycflights13)
library(fs)

Creating an archive of an existing database

First, we’ll need an example database to work with. Conveniently, there is a nice example using the NYC flights data built into the dbplyr package.

tmp <- tempdir() # Or can be your working directory, "."
db <- dbplyr::nycflights13_sqlite(tmp)
#> Caching nycflights db at /tmp/RtmpliOIif/nycflights13.sqlite
#> Creating table: airlines
#> Creating table: airports
#> Creating table: flights
#> Creating table: planes
#> Creating table: weather

To create an archive, we just give ark the connection to the database and tell it where we want the *.tsv.bz2 files to be archived. We can also set the chunk size as the number of lines read in a single chunk. More lines per chunk usually means faster run time at the cost of higher memory requirements.

dir <- fs::dir_create(fs::path(tmp, "nycflights"))
ark(db, dir, lines = 50000)
#> Exporting airlines in 50000 line chunks:
#>  ...Done! (in 0.003139734 secs)
#> Exporting airports in 50000 line chunks:
#>  ...Done! (in 0.01213717 secs)
#> Exporting flights in 50000 line chunks:
#>  ...Done! (in 6.469425 secs)
#> Exporting planes in 50000 line chunks:
#>  ...Done! (in 0.01893973 secs)
#> Exporting weather in 50000 line chunks:
#>  ...Done! (in 0.4542961 secs)

We can take a look and confirm the files have been written. Note that we can use fs::dir_info to get a nice snapshot of the file sizes. Compare the compressed sizes to the original database:

fs::dir_info(dir) %>% 
  select(path, size) %>%
  mutate(path = fs::path_file(path))
#> # A tibble: 5 x 2
#>   path                    size
#>   <chr>            <fs::bytes>
#> 1 airlines.tsv.bz2         260
#> 2 airports.tsv.bz2      28.13K
#> 3 flights.tsv.bz2        4.85M
#> 4 planes.tsv.bz2        11.96K
#> 5 weather.tsv.bz2      278.84K

fs::file_info(fs::path(tmp,"nycflights13.sqlite")) %>% 
  pull(size)
#> 44.9M

Unarchive

Now that we’ve gotten all the database into (compressed) plain text files, let’s get them back out. We simply need to pass unark a list of these compressed files and a connection to the database. Here we create a new local SQLite database. Note that this design means that it is also easy to use arkdb to move data between databases.

files <- fs::dir_ls(dir, glob = "*.tsv.bz2")
new_db <- DBI::dbConnect(RSQLite::SQLite(), fs::path(tmp, "local.sqlite"))

As with ark, we can set the chunk size to control the memory footprint required:

unark(files, new_db, lines = 50000)  
#> Importing /tmp/RtmpliOIif/nycflights/airlines.tsv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.006095648 secs)
#> Importing /tmp/RtmpliOIif/nycflights/airports.tsv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.01172686 secs)
#> Importing /tmp/RtmpliOIif/nycflights/flights.tsv.bz2 in 50000 line chunks:
#>  ...Done! (in 3.995563 secs)
#> Importing /tmp/RtmpliOIif/nycflights/planes.tsv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.01855302 secs)
#> Importing /tmp/RtmpliOIif/nycflights/weather.tsv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.1691473 secs)

unark returns a dplyr database connection that we can use in the usual way:

tbl(new_db, "flights")
#> # Source:   table<flights> [?? x 19]
#> # Database: sqlite 3.35.2 [/tmp/RtmpliOIif/local.sqlite]
#>     year month   day dep_time sched_dep_time dep_delay arr_time sched_arr_time
#>    <int> <int> <int>    <int>          <int>     <int>    <int>          <int>
#>  1  2013     1     1      517            515         2      830            819
#>  2  2013     1     1      533            529         4      850            830
#>  3  2013     1     1      542            540         2      923            850
#>  4  2013     1     1      544            545        -1     1004           1022
#>  5  2013     1     1      554            600        -6      812            837
#>  6  2013     1     1      554            558        -4      740            728
#>  7  2013     1     1      555            600        -5      913            854
#>  8  2013     1     1      557            600        -3      709            723
#>  9  2013     1     1      557            600        -3      838            846
#> 10  2013     1     1      558            600        -2      753            745
#> # … with more rows, and 11 more variables: arr_delay <int>, carrier <chr>,
#> #   flight <int>, tailnum <chr>, origin <chr>, dest <chr>, air_time <int>,
#> #   distance <int>, hour <int>, minute <int>, time_hour <dbl>
# Remove example files we created.
DBI::dbDisconnect(new_db)
unlink(dir, TRUE)
unlink(fs::path(tmp, "local.sqlite"))

Pluggable text formats

By default, arkdb uses tsv format, implemented in base tools, as the text-based serialization. The tsv standard is particularly attractive because it side-steps some of the ambiguities present in the CSV format due to string quoting. The IANA Standard for TSV neatly avoids this for tab-separated values by insisting that a tab can only ever be a separator.

arkdb provides a pluggable mechanism for changing the back end utility used to write text files. For instance, if we need to read in or export in .csv format, we can simply swap in a csv based reader in both ark() and unark() methods, as illustrated here:

dir <- fs::dir_create(fs::path(tmp, "nycflights"))

ark(db, dir, 
    streamable_table = streamable_base_csv())
#> Exporting airlines in 50000 line chunks:
#>  ...Done! (in 0.001517534 secs)
#> Exporting airports in 50000 line chunks:
#>  ...Done! (in 0.01243067 secs)
#> Exporting flights in 50000 line chunks:
#>  ...Done! (in 6.767432 secs)
#> Exporting planes in 50000 line chunks:
#>  ...Done! (in 0.02043438 secs)
#> Exporting weather in 50000 line chunks:
#>  ...Done! (in 0.4580312 secs)
files <- fs::dir_ls(dir, glob = "*.csv.bz2")
new_db <- DBI::dbConnect(RSQLite::SQLite(), fs::path(tmp, "local.sqlite"))

unark(files, new_db,
      streamable_table = streamable_base_csv())
#> Importing /tmp/RtmpliOIif/nycflights/airlines.csv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.005546808 secs)
#> Importing /tmp/RtmpliOIif/nycflights/airports.csv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.01324701 secs)
#> Importing /tmp/RtmpliOIif/nycflights/flights.csv.bz2 in 50000 line chunks:
#>  ...Done! (in 4.400694 secs)
#> Importing /tmp/RtmpliOIif/nycflights/planes.csv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.01743388 secs)
#> Importing /tmp/RtmpliOIif/nycflights/weather.csv.bz2 in 50000 line chunks:
#>  ...Done! (in 0.1346788 secs)

arkdb also provides the function streamable_table() to facilitate users creating their own streaming table interfaces. For instance, if you would prefer to use readr methods to read and write tsv files, we could construct the table as follows (streamable_readr_tsv() and streamable_readr_csv() are also shipped inside arkdb for convenience):

stream <- 
   streamable_table(
     function(file, ...) readr::read_tsv(file, ...),
     function(x, path, omit_header)
       readr::write_tsv(x = x, path = path, append = omit_header),
     "tsv")

and we can then pass such a streamable table directly to ark() and unark(), like so:

ark(db, dir, 
    streamable_table = stream)
#> Exporting airlines in 50000 line chunks:
#> Warning: The `path` argument of `write_tsv()` is deprecated as of readr 1.4.0.
#> Please use the `file` argument instead.
#>  ...Done! (in 0.05276942 secs)
#> Exporting airports in 50000 line chunks:
#>  ...Done! (in 0.01055765 secs)
#> Exporting flights in 50000 line chunks:
#>  ...Done! (in 3.917006 secs)
#> Exporting planes in 50000 line chunks:
#>  ...Done! (in 0.01881218 secs)
#> Exporting weather in 50000 line chunks:
#>  ...Done! (in 0.2543302 secs)

Note several constraints on this design. The write method must be able to take a generic R connection object (which will allow it to handle the compression methods used, if any), and the read method must be able to take a textConnection object. readr functions handle these cases out of the box, so the above method is easy to write. Also note that the write method must be able to append, i.e. it should use a header if append=TRUE, but omit when it is FALSE. See the built-in methods for more examples.

A note on compression

unark can read from a variety of compression formats recognized by base R: bzip2, gzip, zip, and xz, and ark can choose any of these as the compression algorithm. Note that there is some trade-off between speed of compression and efficiency (i.e. the final file size). ark uses the bz2 compression algorithm by default, supported in base R, to compress tsv files. The bz2 offers excellent compression levels, but is considerably slower to compress than gzip or zip. It is comparably fast to uncompress. For faster archiving when maximum file size reduction is not critical, gzip will give nearly as effective compression in significantly less time. Compression can also be turned off, e.g. by using ark() with compress="none" and unark() with files that have no compression suffix (e.g. *.tsv instead of *.tsv.gz).

Distributing data

Once you have archived your database files with ark, consider sharing them privately or publicly as part of your project GitHub repo using the piggyback R package. For more permanent, versioned, and citable data archiving, upload your *.tsv.bz2 files to a data repository like Zenodo.org.