fix(maintenance): export and import Experiments and Variants in starter (#36337)

[erickgonzalez]

Proposed Changes

Fixes #36337.

The starter export endpoints (/api/v1/maintenance/_downloadStarter and /_downloadStarterWithAssets) silently omitted Experiments and their Variants, and the import path never restored them. ExportStarterUtil only dumps Hibernate-mapped tables plus a few hand-coded "special" tables; the experiment and variant tables are persisted via DotConnect (not Hibernate) and were in neither list — so experiment data was lost in starter-based backups and migrations.

Export (ExportStarterUtil)

• In getAdditionalDataAsJSON() (shared by both starter endpoints), emit:
  • Variant.json — all rows including archived and DEFAULT, via VariantTransformer.
  • Experiment.json — all experiments via ExperimentsFactory.list(empty filter).

Import (StarterEntity + ImportStarterUtil)

• Register Variant.json early (before contentlet version info / multi-tree, which reference variant_id) and Experiment.json last (experiment page_id references a contentlet identifier).
• Restore via the variant/experiment factories:
  • Variants: skip the always-present DEFAULT, and use the factory (not VariantAPI, which rejects archived variants) so archived experiment variants restore.
  • Experiments: ExperimentsFactory.save() upserts by id and binds JSONB columns via addJSONParam.
• Deserialize with the configured dotCMS ObjectMapper (JavaTime + Jdk8 modules) so Instant/Optional and JSONB fields round-trip correctly — BundlerUtil's bare mapper cannot.

Out of scope

• The pg_dump path (_pgDumpAvailable / _downloadDb) already includes experiment/variant (not in its --exclude-table-data deny-list) and is left unchanged.

Checklist

• Compiles (./mvnw compile -pl :dotcms-core)
• No REST annotation changes (no openapi.yaml regeneration needed)
• Manual round-trip verification (export → import into clean instance; confirm experiment + variants restored)

How to Test

1. Create an A/B Experiment on a page (writes experiment row + dotexperiment-… variant rows).
2. GET /api/v1/maintenance/_downloadStarterWithAssets (and _downloadStarter); unzip and confirm Experiment.json and Variant.json are present and populated.
3. Import the starter into a clean instance and confirm the experiment and its variants exist and the experiment opens correctly in the UI.

🤖 Generated with Claude Code

This PR fixes: #36337

[claude]

Claude finished @wezell's task in 1m 43s —— View job

---

Rollback Safety Analysis

• Read rollback-unsafe categories reference
• Get full PR diff
• Analyze changes against unsafe categories
• Post results and add appropriate label

---

Result: ✅ Safe To Rollback

After checking every category in the rollback-unsafe reference, this PR matches none of the unsafe categories.

What the PR changes

File	What it does
ExportStarterUtil.java	Adds Variant.json and Experiment.json to the starter export ZIP
ImportStarterUtil.java	Reads and restores those files during starter import
StarterEntity.java	Registers the two new entities in the ordered import list

Category-by-category verdict

Category	Verdict	Reason
C-1 Structural Data Model Change	✅ Safe	No schema changes. Tables variant and experiment already exist; code only reads/writes via existing factories.
C-2 ES Mapping Change	✅ Safe	No Elasticsearch mapping touched. No putMapping, ESMappingAPIImpl, or reindex calls.
C-3 Content JSON Model Version Bump	✅ Safe	CURRENT_MODEL_VERSION in ImmutableContentlet is not touched.
C-4 DROP TABLE / DROP COLUMN	✅ Safe	No DDL of any kind.
H-1 One-Way Data Migration / Backfill	✅ Safe	No UPDATE … SET across existing rows. The export/import path only reads existing rows and writes them back on import.
H-2 RENAME TABLE / RENAME COLUMN	✅ Safe	No renames.
H-3 PK / Unique Constraint Restructuring	✅ Safe	No constraint changes.
H-4 New ContentType Field Type	✅ Safe	No new field types registered.
H-5 Binary Storage Provider Change	✅ Safe	Storage layer untouched.
H-6 DROP PROCEDURE / FUNCTION	✅ Safe	No stored procedures touched.
H-7 NOT NULL Column Without Default	✅ Safe	No ALTER TABLE at all.
H-8 VTL Viewtool Contract Change	✅ Safe	No viewtool classes or returned DTO types modified.
M-1 Column Type Change	✅ Safe	No DDL.
M-2 Push Publishing Bundle Format	✅ Safe	No bundler/handler changes.
M-3 REST / GraphQL API Contract	✅ Safe	No REST endpoints or response shapes changed.
M-4 OSGi Plugin API Breakage	✅ Safe	No public interface signatures changed.

Summary

Rolling back to N-1 after this change is deployed is completely safe. The only behavioral delta is that starter ZIPs produced by N contain two additional JSON files (Variant.json, Experiment.json) that N-1 does not know how to import — N-1 would simply ignore them (as it does for any unrecognized file in the starter ZIP). No database schema, ES index, data format, or API contract was altered.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[github-actions]

🤖 dotBot Review (Bedrock)

Reviewed 3 file(s); 5 candidate(s) → 3 confirmed, 0 uncertain (unverified, kept for review).

Confirmed findings

• 🟡 Medium dotCMS/src/main/java/com/dotmarketing/util/starter/ImportStarterUtil.java:519 — Missing transaction around variant/experiment imports
  The importVariants() and importExperiments() methods save multiple entities via factory.save() in loops without transactional boundaries. If a failure occurs mid-import (e.g., DB error after partial saves), variants/experiments would be partially persisted, leaving the system in an inconsistent state. Variants are referenced in contentlet_version_info and multi_tree, so partial imports could orphan those references. Confirmed via code inspection: no @WrapInTransaction on importStarterData or its callers, and each save is auto-committed individually.
• 🟡 Medium dotCMS/src/main/java/com/dotmarketing/util/starter/ImportStarterUtil.java:519 — Experiment save() may overwrite existing records in non-clean instances
  The code uses ExperimentsFactory.save() which performs an upsert by experiment ID. When importing into a non-clean environment, this will overwrite existing experiments with matching IDs. While Variant processing skips the DEFAULT variant, there's no similar duplicate check for experiments. This violates the starter's expected behavior of additive restoration and risks data loss if reused in existing environments.
• 🟡 Medium dotCMS/src/main/java/com/dotmarketing/util/starter/ImportStarterUtil.java:519 — Test gap: Missing automated tests for Variant/Experiment import with archived states
  The PR adds export/import logic for Experiments and Variants but lacks automated tests verifying the round-trip of archived entries. Manual testing was mentioned, but no test files (e.g., ImportStarterUtilTest) were found via grep, indicating a test coverage gap for critical data integrity scenarios.

---

_{us.deepseek.r1-v1:0 · Run: #28271403691 · tokens: in: 24273 · out: 9586 · total: 33859 · calls: 9 · est. ~$0.085}