data-skew-and-sharing-performance

Data Skew and Sharing Performance

Use this skill when users report slow sharing recalculations, group membership lock errors, or degraded performance when updating records owned by a small number of users or parented under a single account. This skill diagnoses ownership skew and parent-child skew and recommends targeted mitigations.

---

Before Starting

Gather this context before working on anything in this domain:

• Identify which objects are suspected: note API names and approximate record counts.
• Find the ownership distribution: run a report grouped by Record Owner on the suspect object, sorted descending. Flag any user or queue owning more than 10,000 records of a single object.
• For parent-child skew: run a report on the child object grouped by parent (Account, Case, etc.). Flag any parent that has more than 10,000 children.
• Know the current OWD for affected objects: Private OWD combined with a role hierarchy amplifies recalculation cost.
• Know whether the org uses sharing rules sourced from roles or public groups — these are the triggers for recalculation fan-out.

---

Core Concepts

Ownership Data Skew

Ownership data skew occurs when a single user or queue owns more than 10,000 records of a single object. This is the most common performance trap for orgs that park records under a catch-all user (e.g., an "Unassigned Leads" queue or a single integration user that owns all migrated records).

Why it causes problems: When a user moves in the role hierarchy — or is added to or removed from a public group that is the source of a sharing rule — Salesforce must update the sharing table entries for every record that user owns. With 50,000 records owned by one user, a single role change triggers 50,000 sharing table recalculations. This can produce long-running background jobs, "Group membership operation already in progress" errors, and lock contention that blocks other sharing operations.

The 10,000-record threshold: Salesforce documents this explicitly in the Designing Record Access for Enterprise Scale guide. Below 10,000 records per owner, most sharing operations complete quickly. Above this threshold, any change that triggers recalculation for that owner becomes a risk.