⚡ Optimized comment count subquery to avoid full table scan

[9larsons]

ref https://linear.app/ghost/issue/ONC-1509/

The OR condition between parent_id and in_reply_to_id columns in the direct_replies count subquery defeats MySQL index usage, causing a full table scan per comment row. Split into two separate indexed COUNT subqueries summed together, allowing MySQL to use the existing foreign key indexes.

This could easily end up reading several million rows for a single query to get the top 20 comments+replies.

[coderabbitai]

Walkthrough

The comment direct-replies counting was changed from a single OR-based subquery to two correlated, indexed subqueries: r1 counts direct root replies where parent_id matches and in_reply_to_id IS NULL; r2 counts replies-to-child where in_reply_to_id matches. Both subqueries apply status filtering via duplicated parameterized placeholders and their results are summed into count__direct_replies. A comment notes indexing and avoidance of the OR path. No public exports or method signatures were changed; only the internal countRelations.direct_replies implementation was modified.

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: optimizing a comment count subquery to avoid full table scans by splitting an OR-based query into separate indexed subqueries.
Description check	✅ Passed	The description is clearly related to the changeset, explaining the performance issue, the solution approach, and referencing a specific issue tracker link.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch comments-perf-fix

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[9larsons]

EXPLAIN before/after (15K comments, dev dataset)

Before — OR condition defeats index usage, full table scan per comment:

table	type	key	rows	Extra
replies	ALL	NULL	15,369	Range checked for each record

After — two separate indexed subqueries:

table	type	key	rows	Extra
r1	ref	comments_parent_id_foreign	17	Using where
r2	ref	comments_in_reply_to_id_foreign	4	Using where

~730x fewer rows scanned per comment (15,369 → ~21).

For a page load on the busiest post (40 comment objects needing count__direct_replies): 614,760 → 840 total rows scanned. Both /members/api/comments/post/:id/ and /ghost/api/admin/comments/post/:id/ go through the same model code path.

On the reported production dataset (77K comments), this would reduce from ~8.9M row scans per request to ~2,400.

[EvanHahn]

Love having to work around the query planner...

[9larsons]

It's a harsh mistress.

[coderabbitai]

🧹 Nitpick comments (1)

ghost/core/core/server/models/comment.js (1)

266-282: Please add/confirm a regression test for count.direct_replies vs count.replies semantics.

Given this logic rewrite, it would be good to lock behavior with a case that includes descendants and mixed statuses, asserting direct-only counts remain distinct from total descendant counts.

Based on learnings: In Ghost comments API, count.replies is the backward-compatible total-descendants alias, while count.direct_replies is direct-only.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@ghost/core/core/server/models/comment.js` around lines 266 - 282, Add a
regression test that verifies count.direct_replies returns only immediate child
comments while count.replies (the backward-compatible alias) returns total
descendant counts; create test data with a comment tree (parent -> child ->
grandchild) and mixed statuses (hidden, deleted, published) and assert
count__direct_replies (as produced by the direct_replies query/alias) equals
only the number of immediate children while count.replies equals all descendants
excluding statuses per the excludedCommentStatuses logic used in direct_replies;
place the test near existing comment model/query tests and reference the
direct_replies behavior (count__direct_replies) and the legacy count.replies
alias to lock expected semantics.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@ghost/core/core/server/models/comment.js`:
- Around line 266-282: Add a regression test that verifies count.direct_replies
returns only immediate child comments while count.replies (the
backward-compatible alias) returns total descendant counts; create test data
with a comment tree (parent -> child -> grandchild) and mixed statuses (hidden,
deleted, published) and assert count__direct_replies (as produced by the
direct_replies query/alias) equals only the number of immediate children while
count.replies equals all descendants excluding statuses per the
excludedCommentStatuses logic used in direct_replies; place the test near
existing comment model/query tests and reference the direct_replies behavior
(count__direct_replies) and the legacy count.replies alias to lock expected
semantics.

---

ℹ️ Review info

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 0991ddb and bef5455.

📒 Files selected for processing (1)

• ghost/core/core/server/models/comment.js