Limits And Scalability Planning

Use this skill when designing a new Salesforce implementation for scale, auditing an existing org for governor limit exposure before a major release, or troubleshooting a production incident caused by limit violations. This skill provides a structured approach to understanding, measuring, and planning around Salesforce's governor limit architecture. It does not cover individual code optimization techniques — for code-level CPU and heap tuning, use apex-cpu-and-heap-optimization.

Before Starting

• What execution context is relevant? Synchronous user transactions, asynchronous jobs (Batch, Queueable, Future), or scheduled platform events all have different limit profiles.
• What are the current record counts on key objects? Objects with more than 1 million records require different query and DML strategies than smaller objects.
• Is there an active incident or is this preventive planning? Mode 1 (planning), Mode 2 (audit), and Mode 3 (troubleshoot) each have a different entry point.
• What is the org edition? Enterprise and Unlimited Edition orgs have higher metadata limits (800 custom objects) than Professional Edition orgs.

Core Concepts

How Governor Limits Work

Salesforce enforces governor limits per transaction to ensure fair resource sharing across all tenants on a shared infrastructure. Every Apex execution — whether triggered by a user action, a scheduled job, a platform event trigger, or an inbound API call — runs inside a transaction that has hard resource ceilings. When any ceiling is hit, the transaction fails immediately with a System.LimitException. There is no warning; the only prevention is design.

Limits are not shared across transactions. A Batch Apex job that processes 10,000 records in 10 batches of 1,000 gets a fresh governor limit reset for each batch execution. This is the fundamental reason Batch Apex exists: it provides a controlled way to work around single-transaction limits by chunking work.

Synchronous Transaction Limits

These apply to all Apex executed in the context of a synchronous user action (button click, page load, inbound REST/SOAP call, platform event trigger in synchronous mode):

Asynchronous Transaction Limits

Future methods, Queueable Apex, and scheduled Apex run asynchronously. They receive the same SOQL and DML limits as synchronous code, but CPU and heap are higher:

The expanded CPU limit (6x synchronous) is why callout-heavy or compute-heavy operations should be moved to @future, Queueable, or Batch Apex.

Batch Apex Limits

Batch Apex (Database.Batchable) is the correct pattern for processing large data sets that exceed single-transaction limits. Key limits:

Scope size tuning: The default scope size when not specified in Database.executeBatch() is 200. Smaller scopes (50–100) reduce heap and CPU pressure per chunk at the cost of more execute() calls and more governor limit cycles. Larger scopes (1,000–2,000) maximize throughput but risk heap overflow if each record produces a large object graph. Tune scope size based on record complexity and the operations in execute().

Queue contention: The limit of 5 active/queued batch jobs applies across all jobs in the org. If multiple teams submit batch jobs simultaneously — data migrations, nightly rollups, integration sync jobs — they will queue behind each other. Plan batch job schedules to avoid collision. Apex Flex Queue (capacity: 100 jobs) can hold additional jobs in Holding state before they enter the active queue.

Org-Wide Metadata Limits

These apply to the entire org and do not reset per transaction:

The 500 custom fields per object limit is a shared pool. Long text area fields, formula fields, and lookup fields all count toward the same 500. Orgs that model a single object (e.g., Account) as a multi-purpose record type with many field sets risk exhausting this limit. Review field usage regularly and archive or deprecate unused fields.

Data Volume Planning

Data volume directly affects query performance, sharing recalculation, and storage. Plan around these thresholds:

Query performance thresholds:

• Objects with fewer than 100,000 records: standard indexes cover most queries.
• Objects with 1–5 million records: selective queries with indexed filters perform well; non-selective queries begin to cause full table scans.
• Objects with 5+ million records: consider skinny tables (contact Salesforce support to create) for frequently queried field subsets. Skinny tables are internal read-optimized copies of selected fields and eliminate the need for the query engine to join to the main record table.

SOQL selectivity rules: A query is selective when it uses indexed fields and the result set is a small fraction of total records. The selectivity thresholds that Salesforce uses internally:

• For objects with fewer than 333,333 records: result set must be fewer than 10% of total records.
• For objects with 333,333–1,000,000 records: result set must be fewer than 33,333 records.
• For objects with more than 1,000,000 records: result set must be fewer than 333,000 records, or fewer than 10% — whichever is smaller. Non-selective queries bypass indexes and scan the full object table. On large objects this causes CPU timeout and "query took too long" errors.

Data skew — sharing recalculation: When a single user or a small set of users owns a disproportionate number of records (more than 10,000 records per owner is the threshold for skew concern), sharing recalculation after ownership changes becomes slow and can lock the org for extended periods. Common causes: integration users who own all imported records, queue-based ownership without ownership reassignment, single admin owning all accounts. Mitigation: distribute ownership, use public read/write sharing where data sensitivity allows, or use territory management.

Platform Event Limits

Platform events are not a substitute for high-frequency streaming. At 250,000 deliveries per 24 hours, a single integration sending events every second hits the limit in roughly 70 hours of sustained operation. Batch or aggregate events where possible, and monitor usage against the limit in Setup > Platform Events.

Storage Limits

Data storage counts every record in every standard and custom object, plus chatter activity, email logs, and archived records. File storage counts attachments, files, documents, and Content records. When either limit is approached, record creation fails. Monitor storage in Setup > Storage Usage and plan archival or deletion strategies before limits are reached.

Mode 1: Plan for Scale (New Build)

Use this mode when designing a new org, new feature, or new integration before development begins.

Step 1 — Build the Governor Limit Inventory

For each planned transaction type (user action, trigger, integration event, batch job), document:

1. SOQL count: How many queries are needed? Include queries from helper methods, triggers on related objects, and validation rules that use formula fields backed by SOQL.
2. DML count: How many DML statements and rows per transaction? Each insert, update, delete, upsert on a different SObject type is one DML statement.
3. CPU budget: Estimate CPU usage. Complex formula field evaluation, large collection iterations, and string manipulation all consume CPU. If callouts are involved, they consume CPU during response parsing.
4. Heap budget: Estimate peak heap. Large SOQL result sets, deep object graphs, and string concatenation in loops are common heap killers.

Step 2 — Identify Async Offload Candidates

Move any of the following to async execution:

• Operations that do not need to complete before the user sees a response
• Callouts that are not required in the same transaction as a record save
• Rollup calculations over large related lists
• Cross-object field sync that touches more than 200 records
• Integration payloads that require transformation and outbound delivery

Step 3 — Plan Data Volume Milestones

Project record growth for key objects at 1-year, 3-year, and 5-year intervals. Flag any object projected to exceed 1 million records and design queries to use indexed fields from day one. Large volume objects need:

• A selective WHERE clause using indexed fields on every query
• No SELECT * in Apex — select only the fields needed
• Skinny table consideration when 5 million records is in scope within 3 years

Step 4 — Plan Storage

Calculate data storage: estimated records × average record size (roughly 2KB per standard record, more for objects with many long text fields). Add file storage for attachments, content versions, and emails. If projected storage exceeds 80% of allocated within 2 years, include an archival strategy in the design (Big Objects for cold archival, data export + deletion for compliance retention, or a Salesforce Shield archive strategy).

Step 5 — Check Metadata Headroom

Count existing custom objects, fields per object, and active flows. Calculate headroom against org limits. Flag any object approaching 400 custom fields (80% of the 500 limit) for a field audit before adding new fields.

Mode 2: Audit an Existing Org for Limit Risk

Use this mode when reviewing an org before a major release, after an acquisition, or during a health check engagement.

Audit Checklist

Governor limit exposure:

• Review Apex debug logs for LIMIT_USAGE_FOR_NS entries — look for transactions consuming more than 70% of any governor limit.
• Check for SOQL queries in loops — any [SELECT ...] inside a for loop is an immediate risk.
• Review Batch Apex scope sizes — any job with scope > 500 should be validated against heap usage.
• Check for synchronous callouts in triggers — callouts from before/after triggers run in the same user transaction; slow external systems extend CPU time.
• Verify that future method chains do not exist — @future methods cannot call other @future methods. This is a hard runtime error.

Org-wide metadata headroom:

• Custom objects used vs. 800 limit — flag anything above 700 (87.5%).
• Custom fields per object — review any object above 400 fields.
• Active flows — flag anything above 3,500 active flows (87.5% of 4,000).

Data volume risks:

• Identify objects with more than 1 million records using record counts in Setup > Storage Usage.
• Review sharing model on large-volume objects — private sharing on an object with 10M records and complex role hierarchy is a recalculation risk.
• Check for data skew — any owner with more than 10,000 records on a private-sharing object.

Platform event throughput:

• Review platform event usage in Setup > Platform Events — check the 24-hour delivery count against the 250,000 limit.
• Identify any event subscriber triggers that perform high-SOQL or high-DML operations — they share the same governor limits as other async Apex.

Storage:

• Check current data and file storage consumption in Setup > Storage Usage.
• Identify the top storage consumers — which objects hold the most records? Which users have the most file storage?

Risk Rating Model

Mode 3: Troubleshoot a Limit-Related Incident

Use this mode when a production incident involves System.LimitException, timeout errors, or unexpected job failures.

Diagnostic Table

Limit Incident Response Steps

1. Capture a debug log at FINEST level from the failing user or job. Look for LIMIT_USAGE_FOR_NS entries at the end — this shows the exact limits consumed when the failure occurred.
2. Identify the worst-offending limit — the one closest to its ceiling.
3. Trace the caller — which method, trigger, or flow caused the limit spike?
4. Apply the shortest-path fix first — if SOQL is in a loop, extract it. If heap is too large, reduce the SELECT field list or chunk the data set.
5. Validate the fix in a full-volume sandbox — limit errors that only occur at scale cannot be reliably caught in developer orgs with small data sets.
6. Add monitoring — insert System.debug(Limits.getQueries() + ' / ' + Limits.getLimitQueries()) checkpoints in high-risk transactions to catch future creep before it becomes an incident.

Planning Strategies Reference

Related Skills

• apex/batch-apex — for Batch Apex authoring patterns and trigger mechanics
• apex/async-patterns — for Queueable, Future, and scheduling patterns
• apex-cpu-and-heap-optimization — for code-level CPU and heap tuning (NOT this skill)
• data/large-data-volumes — for LDV indexing, skinny tables, and archival strategies
• architect/solution-design-patterns — for automation layer selection

Recommended Workflow

Step-by-step instructions for an AI agent or practitioner activating this skill:

1. Gather context — confirm the org edition, relevant objects, and current configuration state
2. Review official sources — check the references in this skill's well-architected.md before making changes
3. Implement or advise — apply the patterns from Core Concepts and Common Patterns sections above
4. Validate — run the skill's checker script and verify against the Review Checklist below
5. Document — record any deviations from standard patterns and update the template if needed