nordstatExtras

Shared SQLite cache backend for the nordstat family of R packages: rKolada, rTrafa, and pixieweb.

Why

The three nordstat packages each ship a per-package .rds cache that works fine for single-user scripts but falls short in multi-user web applications:

• No concurrent-write safety
• Cache granularity is the whole query, not the individual value — two overlapping queries refetch the same cells
• Metadata is cached but actual data values (get_values(), get_data()) are not

nordstatExtras replaces this with a single SQLite database shared across processes. Values are stored at cell granularity, deduplicated across overlapping queries, and expire based on the age of the cell itself — a refresh by one query propagates to every query that references the same cell.

Install

# install.packages("devtools")
devtools::install_github("LCHansson/nordstatExtras")

Use

Open a cache once per deployment, then hand it to the source packages via their cache_location argument:

library(nordstatExtras)

handle <- nxt_open("cache.sqlite")

# rKolada
kolada_vals <- rKolada::get_values(
  kpi = c("N03700", "N03701"),
  municipality = c("0180", "1480"),
  period = 2020:2024,
  cache = TRUE,
  cache_location = handle
)

# rTrafa
trafa_vals <- rTrafa::get_data(
  "t10011", "itrfslut",
  ar = c("2023", "2024"),
  cache = TRUE,
  cache_location = handle
)

# pixieweb
scb <- pixieweb::px_api("scb")
px_vals <- pixieweb::get_data(
  scb, "BE0101N1",
  Region = c("0180", "1480"),
  Tid = px_top(5),
  cache = TRUE,
  cache_location = handle
)

nxt_close(handle)

All three calls store their data in the same SQLite file. On a cache hit the source package skips the HTTP fetch entirely; on a miss it fetches, normalizes to cell format, and UPSERTs.

Design

• Cells: the cells table stores one row per statistical datum with a composite UNIQUE key (source, api_alias, entity_id, variable, period, dims_hash, lang). Overlapping queries deduplicate.
• Dimensions: arbitrary package-specific dimensions live in a sidecar cell_dims table, keyed by a hash of the dimension vector.
• Queries: the queries table plus a query_cells junction lets the handler look up which cells a given call originally produced, so it can reconstruct the source package's expected tibble shape on load.
• TTL (cell-level): a query is fresh iff every cell it expects is present and the oldest cell is younger than max_age (default 30 days). Cross-query refreshes propagate automatically.
• Async writes: optional mirai-backed background flushing via nxt_write_async() + nxt_flush(). Falls back to sync when mirai is not installed.
• Multi-process: SQLite runs in WAL mode with synchronous = NORMAL, safe for concurrent readers and serial writers.

Maintenance

handle <- nxt_open("cache.sqlite")

# Drop a source entirely
nxt_clear(handle, source = "kolada")

# Delete stale cells (default 30-day TTL)
nxt_gc(handle)

nxt_close(handle)