Performance

When to Use

Optimize Salesforce sync performance when experiencing high API usage, slow entity saves, or large queue backlogs. The primary levers are: caching object metadata, async queue processing, and selective field/record queries.

Decision

Scenario	Strategy
High volume entity saves	Enable `async = TRUE` on all mappings
Need low latency sync	Use `async = FALSE` with retry on `PUSH_FAIL`
Large Salesforce objects	Use specific field lists, not all fields
Repeated schema inspection	Metadata is cached — do not disable caching
High pull frequency	Increase `pull_frequency` to rate-limit API calls

Caching

Object Metadata: - Cache: salesforce:objects, salesforce:object:[name] - Lifetime: short_term_cache_lifetime (default: 300s) - Clear: drush cache:rebuild or Salesforce admin UI

API Versions: - Cache: salesforce:versions - Lifetime: long_term_cache_lifetime (default: 86400s)

Record Types: - Cache: salesforce:record_types - Lifetime: short_term_cache_lifetime

Queue Processing Strategy

High Volume Sites: - Enable async = TRUE on all mappings - Use standalone endpoints with custom scheduling - Increase global_push_limit and push_limit - Monitor queue size and API limits

Low Latency Requirements: - Use async = FALSE for critical mappings - Accept blocking during entity save - Implement retry logic via PUSH_FAIL event

API Call Optimization

Pull Queries: - Use specific field lists (not SELECT *) - Implement WHERE clauses to limit records - Use pull_trigger_date for incremental sync - Avoid unnecessary subqueries

Push Operations: - Batch queue processing vs real-time - Use upsert when appropriate - Minimize field mappings to required fields

Also worth applying: use pull_record_type_filter to reduce pull volume for multi-RecordType objects.

Common Mistakes

Wrong: Disabling caching to force fresh metadata on every API call → Right: The 300-second short-term cache is appropriate; cache miss on every call wastes API quota
Wrong: Using async = FALSE for bulk import operations → Right: Always use async = TRUE for bulk operations; process the resulting queue via drush sfpushq in batch