Save and reuse verified LMTP task results

A caller defines each task's complete scientific and runtime identity. A completed result is reused only after checking that identity and the payload checksum. Computation and installation hold exclusive local ownership. An existing ownership directory requires inspection; locks never expire automatically. These functions do not define estimators or schedule workers.

Usage

margot_lmtp_checkpoint_read(directory, key, identity)

margot_lmtp_checkpoint_compute(directory, key, identity, compute)

margot_lmtp_checkpoint_inventory(directory, identities, load = TRUE)

Arguments

directory

Private directory for these tasks.

key

Unique task name using lowercase letters, digits and underscores.

identity

Nonempty list identifying all result-affecting inputs, ordered participants, histories, folds, weights, settings and code/builds. Environments, closures and external pointers are refused; pass a fingerprint of a function instead of the function.

compute

Zero-argument function computing a missing task.

identities

Named list of identities for the complete expected task set.

load

For inventory, whether to return each verified value. With load = FALSE, every file is still verified but records hold metadata only (path and checksums), so the inventory does not hold the saved results in memory. The default TRUE keeps the earlier behaviour.

Value

read returns a verified record or NULL; compute returns a record containing value, path, reused and checksums; inventory returns completion status, a task table and verified records, which hold value only when load = TRUE.

Details

Identities are compared in canonical form (see margot_lmtp_checkpoint_identity()), so attributes that record where an input came from, such as the file path RcppTOML attaches to a parsed study file, do not change a task's identity.