
Are this host's DB primitives fresh enough to model against?
Source:R/lnk_preflight_vintage.R
lnk_preflight_vintage.Rdsnapshot_bcfp.sh loads four primitives from public sources into each
host's local fwapg. Every cypher reloads them during prep; the
dispatcher does not, and nothing checks. On 2026-08-30 the dispatcher's
cabd.dams was 2026-05-23 and fresh.modelled_stream_crossings
2026-05-26 — a run started that day would have modelled one bucket on
May inputs and three on August inputs, produced one consolidated table
set, and said nothing about it anywhere (link#246).
Usage
lnk_preflight_vintage(
conn = NULL,
max_age_days = 7,
tables = .lnk_vintage_primitives(),
now = Sys.time(),
vintage = .lnk_vintage_read(conn, tables),
stamp = .lnk_vintage_stamp(conn, tables),
heads = .lnk_vintage_heads(stamp),
quiet = FALSE
)Arguments
- conn
A DBI::DBIConnection to the host's local fwapg, or
NULLwhenvintageis supplied directly.- max_age_days
Maximum acceptable age of the oldest primitive.
- tables
Fully-qualified table names to check. Defaults to the snapshot-loaded set.
- now
Reference time. Injectable so tests are not clock-dependent.
- vintage
A data frame with
table_nameandlast_analyze, and optionallytable_exists. Defaults to readingconn; pass directly to test, or to judge a stamp collected on another host. Rows in a frame withouttable_existsare taken to exist.- stamp
A data frame with
table_nameandloaded_at: this host's load stamps. Defaults to readingconn;NULL(the default whenconnisNULL) means "not consulted" and leaves the statistics to decide. A data frame with no row for a table makes that tableunstamped.- quiet
Suppress the human-readable report.
Value
Invisibly, a list with ok, vintage, stale, absent,
unknown, missing (the union of the last two, kept for callers that
only care that something was wrong), unstamped, judged_by (per
table: "stamp", "pg_stat" or "none"), oldest_days and message.
Details
The FWA tables in .lnk_input_primitives() are bulk-restored
FWA. They are never ANALYZEd, so they carry no vintage at all and are
not an axis this can measure — including them would mean every host
failing forever on data that is not the staleness risk.
Absence is not a pass, but absence and ignorance are different
failures and are reported as such. A table that does not exist is
absent; one that exists but yields no usable timestamp is unknown.
Both fail, in the same direction as a stale one — a query returning
nothing must never read as "nothing is stale" — but they send an operator
to different places. Conflating them cost a pilot run, which reported
never loaded / absent for a table that was present and healthy.
The load stamp is the primary source (link#265, absorbing #247).
snapshot_bcfp.sh records each primitive it loads in
provenance.source_artifact (recorded_by = 'loader', with loaded_at).
When the gate has a connection it judges each table on that stamp, and a
table with no stamp for this host fails (unstamped) — even when the
statistics below say it is fresh. That is #247's point: statistics are an
inference that can read too-new (autoanalyze fires on ordinary writes), and
a host that never recorded a load should say so on its first run rather
than pass on an inference. The statistics age is still shown beside an
unstamped table, as context. A stamp never rescues a table that does not
exist. probe rows are not stamps: a probe did not load anything.
Without a connection (a vintage frame judged offline) there is no stamp
to read, the statistics decide, and the message says the stamp was not
consulted.
Two statistics quirks make this fiddlier than it looks. last_analyze
alone is unusable: measured across all ten primitives it is NULL on every
one and only last_autoanalyze is populated, so the query takes
GREATEST of the pair (in Postgres that ignores NULLs and is NULL only
when every argument is). And on a freshly restored database neither is
set — statistics are collected by (auto)analyze, so a table that was just
loaded has rows and no stats at all. The query therefore falls back to the
relation file's mtime, which exists for any real table. That fallback is
why unknown should be unreachable in practice.
Examples
# Judge a vintage table without touching a database:
now <- as.POSIXct("2026-08-30 12:00:00", tz = "UTC")
v <- data.frame(
table_name = link:::.lnk_vintage_primitives(),
last_analyze = now - c(1, 2, 1, 99) * 86400)
res <- lnk_preflight_vintage(vintage = v, now = now, max_age_days = 7,
quiet = TRUE)
res$ok
#> [1] FALSE
res$stale
#> [1] "whse_fish.pscis_assessment_svw"