The target path

The target path is the folder DRE writes its generated files to. It has nothing to do with a run’s target (--target, the environment every profile uses). It’s target/ in the project unless set, highest first, by:

1. --target-path <path> on compile, validate, run, clean and ls;
2. DRE_TARGET_PATH;
3. target_path: in dre_project.yml.

A relative path is relative to the project root, whichever of the three sets it; ~ is your home directory. Compiled SQL, run outputs, schema snapshots, run_results.json and the manifest move together. The path must be local or mounted: object storage URLs (s3://, gs://, abfss://) are refused, but a filesystem mount that supports renames works, such as a Databricks Unity Catalog Volume (/Volumes/... on Databricks compute), an NFS/EFS share or a gcsfuse mount. DRE refuses a target path that is the project root, contains the project, or sits inside reports/, macros/, lookups/ or dre_deps/. A path elsewhere inside the project is skipped when DRE reads the project; add it to .gitignore (a new project’s .gitignore covers target/ only).

Schema-drift detection compares each run with the snapshot the last successful run left in the target path, so on an ephemeral runner (a job cluster, a CI runner) point it at a folder that outlives the run:

dre run --schedule close_monthly --target-path /mnt/shared/dre/target

In a Databricks job, a Volume keeps the snapshots and the manifest between runs. Use a Volume (or local disk), not Workspace files (/Workspace/...): those don’t reliably keep the renames behind DRE’s atomic writes and run pointer, and dre validate warns about them.

export DRE_TARGET_PATH=/Volumes/main/reporting/dre/target
dre run --schedule close_monthly

Runs, and runs that overlap

Each run of a Binding writes into its own folder, and a small current file says which run is the latest that finished:

target/run/<report>/<set or default>/
  current          the id of the latest finished run
  lock             only while a run is going on: its run id, host, process and start time
  runs/
    20261009T060000Z-k3f9/   one folder per run: its files and run_results.json

A run id is the run’s UTC start time and four random characters. Each run’s log, dre.log, is in its folder too (other commands write to the console only). dre history <report> lists a report’s runs and which is current; dre history <report> --latest --path prints the folder with the latest files, for scripts.

By default only the latest run is kept. To keep a history (an audit trail of exactly what was sent, and when), set how many runs of each report and Set to keep, of any status; the current run always stays:

flags:
  keep_runs: 30

dre run --keep-runs and DRE_KEEP_RUNS override it. Older runs are removed after each run, and dre clean --prune removes them without deleting anything else (dre clean alone deletes the whole folder). Recommended on servers and anything scheduled; mind the size of large xlsx or parquet outputs.

So several runs can share one target path safely:

• The same report and Set never run twice at once. A second dre run of it finds the lock, stops at once and changes nothing (no files deleted, nothing delivered, the drift snapshot left alone); its error names the run in progress. Different reports run in parallel freely. A lock left by a process on this machine that has gone (a crash, a kill) is taken over with a warning; one from another machine (a shared folder) is never guessed about: dre unlock <report> [--binding <set>] shows it and removes it.
• A crash leaves the previous run current, and only an unfinished folder behind, which the next run removes.
• An older run never replaces a newer one. A rerun for an earlier instant (DRE_RUN_AT) than the current run’s finishes in its own folder but doesn’t become current, and doesn’t touch the drift snapshot.
• Each file DRE reads back (current, run_results.json, the schema snapshots, the manifest, dre.lock) is written whole: a temporary file next to it, flushed to disk, then renamed into place, so a run stopped mid-write leaves the previous file or the new one, never a broken one.

Before 0.4 a Binding’s files were straight in target/run/<report>/<set or default>/. The first 0.4 run moves them into a runs/<time>-legacy/ folder (and the drift snapshot carries on). Don’t run 0.3 and 0.4 on one target path.

Cleaning

Inspect and prune a report’s history with explicit storage and Binding names:

dre history monthly --binding domestic --output json
dre history monthly --binding domestic --latest --path
dre clean --prune --keep-runs 3

--binding names a Set, or default for a report without Sets. --latest selects the current finished run and --path prints only its directory. The current run may have failed: inspect its run_results.json before treating its files as delivered.

For a lock from a stopped process on another machine, inspect its holder before unlocking:

dre unlock monthly --binding domestic --target-path /mnt/shared/dre/target

--yes skips the confirmation in an automated recovery procedure. Use it only after that procedure has confirmed the holder stopped; removing a live lock permits overlapping runs. run-in-progress means a lock still exists, not that the output is corrupt. target-path-unwritable means permissions, a mount or disk capacity needs fixing before another run.

The monthly finance project provides a complete report to inspect. See CLI options, error codes and the target path and Binding terms.

dre clean deletes the target folder only if DRE created it (it leaves a .dre_target file there) or it’s the project’s own target/, so a mistyped --target-path can’t delete anything else.