Feature/refresh sync out 3645

[SandipBajracharya]

No description provided.

[linear-code]

OUT-3645

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
dropbox-integration	Ready	Preview, Comment	May 29, 2026 9:44am

[greptile-apps]

Greptile Summary

This PR adds a pending-action queue mechanism to the file_folder_sync table, enabling retry-tracked outbound sync operations toward Assembly or Dropbox, along with two partial unique indexes to enforce per-channel deduplication of assembly and Dropbox file IDs.

• Adds pending_action, pending_action_target, pending_action_attempts, pending_action_last_attempt_at, and pending_action_last_error columns to file_folder_sync, with a null-consistency check constraint ensuring action and target are always set or cleared together.
• Adds two partial unique indexes scoped to non-deleted rows; the migration creates these without CONCURRENTLY, which will exclusively lock the table during the build.
• Introduces PendingAction and PendingActionTarget TypeScript enums in constants.ts, keeping the DB enum values in sync with application code.

Confidence Score: 3/5

Safe to merge for schema correctness, but the second migration will exclusively lock file_folder_sync for the full index build, which poses a real availability risk on a production table before that is addressed.

The column additions and check constraint are clean and low-risk. The unique-index migration runs without CONCURRENTLY, taking an ACCESS EXCLUSIVE lock on file_folder_sync for the entire build and blocking all reads and writes on what is likely a busy table. There is also no mechanism to reset pending_action_attempts when a new pending action cycle begins on a row, which could silently exhaust the retry budget.

src/db/migrations/20260526094109_add_unique_indexes_on_portal_channel_assembly_and_dbx_file.sql needs attention for the non-concurrent index creation; src/db/schema/fileFolderSync.schema.ts warrants a second look around the attempts-counter reset semantics.

Important Files Changed

Filename	Overview
src/db/migrations/20260526094109_add_unique_indexes_on_portal_channel_assembly_and_dbx_file.sql	Creates two partial unique indexes without CONCURRENTLY, which will take an exclusive lock on file_folder_sync for the duration of the build; includes helpful pre-deploy audit queries as comments.
src/db/schema/fileFolderSync.schema.ts	Adds pending-action columns, two partial unique indexes, and a null-consistency check constraint to fileFolderSync; attempts counter has no reset guard between pending-action cycles.
src/db/migrations/20260522104036_add_pending_action_columns_to_file_folder_sync.sql	Adds two new enum types and five new nullable/defaulted columns to file_folder_sync with a null-consistency check constraint; straightforward and safe as an ALTER TABLE migration.
src/db/constants.ts	Adds PendingAction and PendingActionTarget enums with matching TypeScript value types; clean and consistent with existing enum pattern.
.gitignore	Adds docs/ directory to .gitignore; no issues.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Event triggers sync change] --> B{Set pending_action + pending_action_target}
    B --> C[file_folder_sync row - check constraint enforced]
    C --> D[Worker polls: WHERE pending_action IS NOT NULL]
    D --> E{Attempt action}
    E -->|Success| F[Clear pending_action and pending_action_target to NULL]
    E -->|Failure| G[Increment pending_action_attempts]
    G --> H{attempts >= MAX?}
    H -->|No| D
    H -->|Yes| I[Dead-letter / give up]
    F --> J[Row retains pending_action_attempts value - no auto-reset]

Loading

_{Reviews (1): Last reviewed commit: "refactor(OUT-3645): unify ON CONFLICT co..." | Re-trigger Greptile}

[greptile-apps]

pendingActionAttempts can carry over stale counts to a new pending cycle

There is no constraint or default reset that zeros pending_action_attempts when pending_action is cleared and then re-set on the same row. If application code checks attempts >= MAX_RETRIES before processing, a row that previously failed 3 times and was cleared will start its next pending action cycle with attempts = 3 and could be immediately abandoned.

[greptile-apps]

No index to efficiently query rows with a pending action

The worker or poller that picks up rows needing processing will query something like WHERE pending_action IS NOT NULL. Without an index on pending_action, every such scan is a full table sweep on file_folder_sync. Since this column is sparse (most rows will have NULL), a partial index on (pending_action, pending_action_target) where pending_action IS NOT NULL would keep the working set tiny and make polling very cheap.