Drupal Config Reconcile

Drupal Config Reconcile

A guided workflow for resolving the gap between your tracked config (config/default, plus any split dirs) and a deployed environment's active configuration. Most teams have the mechanics (config:status, cim, cex) but no disciplined way to walk drift item-by-item and decide which side wins. This is that discipline.

The core decision: import vs export vs skip

For every divergent config item, exactly one of three things is true, and you pick a direction:

• Import = disk → DB. The tracked file is the source of truth; it deploys to the environment on the next code deploy + config import. No file change — you keep what's in git.
• Export = DB → disk. The environment's value wins; you write its config into your tracked files (or git rm the file if the environment doesn't have it).
• Skip. Leave the divergence unresolved and move on (the right call for benign uuid/_core-only diffs).

Never run config write operations against production. This workflow only reads from a reference environment and only writes to local files. It never writes to any remote database.

Remote CLI — host-neutral

Examples below use Pantheon's Terminus. The workflow is the same on any host — substitute your platform's remote-drush command. If your platform provides Drush site aliases, the generic drush @<alias> <cmd> form works everywhere and is the simplest baseline.

Auth/link once per platform: Acquia acli auth:login + acli link; Pantheon terminus auth:login; Platform.sh/Upsun platform login / upsun login; Lagoon lagoon login. Never run write/clone operations into production on any of them.

Workflow

1. Pick a reference environment (never live/prod)

Use a non-production environment as the prod stand-in (e.g. Pantheon dev/test, or a CI/staging alias). Refuse to target the live environment — config writes against live are how teams nuke production config.

2. Make the reference DB fresh

The diff is only meaningful if the reference env reflects current production. If the env's DB is stale (commonly >24h), refresh it from live first. On Pantheon:

terminus env:clone-content <site>.live <env> --db-only -y   # Pantheon; see the table above for acli / platform / upsun / lagoon

Cloning modifies a shared environment and takes minutes — confirm with the user first; never do it silently. (Note: most platforms can't reliably tell you when an env was last refreshed — terminus backup:list reports backup recency, not clone-from-live recency. Treat it as a weak hint and ask.)

3. Get the diff on the reference env (not locally)

Run config:status on the reference env so Drupal applies config splits and config_exclude_modules correctly — a local run produces false positives from dev-only modules (devel, views_ui, a prod split, etc.):

terminus drush <site>.<env> -- config:status --format=json   # or your platform's remote-drush form (see table)

States you'll see:

• Only in DB — exists in the env's database, not in tracked config.
• Only in sync — exists in tracked config, not in the env's database.
• Different — present in both, content differs.

Empty list → "in sync, nothing to reconcile" → stop.

4. Fetch the env's value per item with config:get (not a full export)

Do not reach for config:export --destination + scp — config_split commonly fails trying to create its relative split dir under a temp destination, and SSH-command discovery is brittle across Terminus versions. Fetch each item on demand:

terminus drush <site>.<env> -- config:get <name> --format=yaml > /tmp/ref-<name>.yml   # or your platform's remote-drush form

config:get --format=yaml emits the exact file format Drupal exports (no _core key) — byte-identical to tracked sibling files.

Caveat: config:get appends an extra trailing newline. Drupal's phpcs (Drupal.Files.EndFileNewline.TooMany) rejects two newlines at EOF and a pre-commit hook will block the commit. Normalize to exactly one trailing newline when you write the file:

printf '%s\n' "$(cat /tmp/ref-<name>.yml)" > config/default/<name>.yml

5. Walk each item, one at a time

Create a todo list (one entry per config name) so progress is visible. Process in this order — Only in DB, then Different, then Only in sync — and for each:

1. Locate the local file. Usually config/default/<name>.yml. If absent there, check split dirs (config/prod/, config/local/, config/envs/…) before concluding it's missing — split-managed items live outside config/default and must be written back to the same split dir.
2. Show the content diff (label left = local/disk, right = env/DB):
   • Different: diff -u config/default/<name>.yml /tmp/ref-<name>.yml
   • Only in DB: local absent → show the env's full content.
   • Only in sync: env absent → show the local file's full content.
3. Recommend a direction, and say why:
   • Only in DB → Export. Config created in the environment (UI or update hook) that should be captured into git (message templates, view displays, etc.).
   • Only in sync → Import. New config added in code, not yet deployed; keep it and let it deploy.
   • Different:
     • Only uuid and/or _core differ → Skip (benign environment artifact; don't churn files).
     • Env has real intentional-looking value changes → Export.
     • Local file has intentional code changes → Import.
     • Genuinely unclear → present both sides plainly, default to Skip, let the user decide.
4. Ask Import / Export / Skip.
5. Perform it — local files only:
   • Export (Different/Only in DB): write the env's file with the trailing newline normalized (split-managed items → split dir). Only in sync + Export means the env doesn't have it → confirm, then git rm the local file.
   • Import: keep the local file; no change. ⚠️ Only in DB + Import is destructive — the env has config absent from git, so a full drush cim on deploy will delete it from the environment. Warn explicitly and require a second confirmation.
   • Skip: nothing.

6. Verify with the import transformer — NOT config:status

After writing exported files, do not re-run config:status to confirm — it can keep reporting a freshly-written item as Only in DB even though the file is present and valid (a stale display artifact). Verify against the storage layer cim actually uses, the import transformer (which applies config_split exactly as a deploy does):

drush ev '
$t = \Drupal::service("config.import_transformer")->transform(\Drupal::service("config.storage.sync"));
foreach (["<name1>","<name2>"] as $n) {
  print "$n => " . ($t->exists($n) && in_array($n, $t->listAll(), true) ? "RECONCILED (deploy-safe)" : "MISSING — would be deleted on cim") . "\n";
}'

exists() && in listAll() = YES means a real deploy keeps the config. Use this — not the config:status table — as the source of truth for "did my export take?"

7. Summarize and hand off

• Exported (env → git): files created/overwritten/removed.
• Kept for deploy (import): items whose local version reaches the env on next deploy.
• Skipped: unresolved divergences and why.

Then show git status + git diff --stat config/, and do not auto-commit or push — ask first. Remember that Import items only take effect after the code deploys and a config import runs on the environment.

Gotchas

• config:status on the reference env compares that env's sync dir vs its DB — it won't see your uncommitted local edits. Commit/deploy your edits first, then reconcile remaining drift.
• Config splits: prod/local split items live in config/prod / config/local, not config/default. Always write Export results back to the file's real location.
• uuid-only or _core-only diffs are almost always benign — don't churn files over them.