Trade-offs and platform limits of integration options
The limits and trade-offs that decide whether an integration survives production: long-running request and callout limits, Salesforce Connect and External Services constraints, event retention, CDC entity limits, Bulk API 2.0 transactions, query cursors, outbound message retries, partial success behind HTTP 200 and Well-Architected trade-offs.
Given a use case, identify the trade-offs, limitations, and constraints that meet the proposed solution.
~8 min read · 17 practice questions ·
What the exam expects
- Diagnose a failing design from its symptom: REQUEST_LIMIT_EXCEEDED, invalid query locators, callouts rejected from triggers, records missing behind HTTP 200.
- State the constraints of each option: External Services, Salesforce Connect, platform events and Change Data Capture, Bulk API 2.0, outbound messaging, Heroku Connect, MuleSoft Anypoint.
- Know which limits no longer bite (callouts and the long-running request limit) and which do (120-second callout timeout).
- Frame trade-offs with the Well-Architected pillars instead of maximizing every control.
Key terms: long-running request (an Apex request over 5 seconds, or an inbound API call of 20 seconds or longer, each with its own limit); retention window (72 hours of replayable high-volume events); replay ID (increasing event position a subscriber resumes from); partial success (some records fail inside a well-formed request); pillar trade-off (spending on one quality dimension at another's expense).
Key facts
- Detail
- Excluded since Winter '20; timer pauses during the callout
- Detail
- Over 5 s; 1 per 100 licenses, min 10, max 50
- Detail
- Default 10 s; max 120 s per callout and cumulative per transaction
- Detail
- 25 of 20 s or longer in production and sandboxes
- Detail
- No limit since API 56.0, all versions; 2-day lifetime (the catalog's 10 per user is stale)
- Detail
- 72 hours; no Salesforce-side retry
- Detail
- 5
- Detail
- Retried until 24 hours old, then dropped
- Detail
- Parallel, no serial mode, each its own transaction
- Detail
- 500 nodes; each graph all-or-nothing
- Detail
- Extra cost in Enterprise, Performance, Unlimited
Sources for this section (10)
- Continuations (LWC)developer.salesforce.com (opens in a new tab)
- Callout Limitsdeveloper.salesforce.com (opens in a new tab)
- Execution Governors and Limitsdeveloper.salesforce.com (opens in a new tab)
- API Request Limitsdeveloper.salesforce.com (opens in a new tab)
- Integration Patterns Overviewarchitect.salesforce.com (opens in a new tab)
- API Query Cursor Limitsdeveloper.salesforce.com (opens in a new tab)
- Select Objects for Change Notificationsdeveloper.salesforce.com (opens in a new tab)
- Considerations for Outbound Messageshelp.salesforce.com (opens in a new tab)
- Using Composite Graphsdeveloper.salesforce.com (opens in a new tab)
- Salesforce Platform Features Supported by Salesforce Connecthelp.salesforce.com (opens in a new tab)
- Do three middleware apps sharing one integration user still exhaust open query cursors?
- When does Bulk API 2.0 make a parent-before-child sorted file unsafe?
- Why is the cross-org adapter a poor fit for a pipeline total across four orgs?
- What happens to an outbound message whose endpoint stays down?
The questions are free to read. Answers with their sources and self-grading are included with the pack.
Keep reading with the Integration Architect pack
- Core concepts
- Exam traps
- Worked scenario
The pack includes:
- All 25 lessons of the Integration Architect guide, with reading progress
- Interactive decision walks and step-by-step sequence steppers
- Inline practice questions and one-click topic practice (17 questions for this lesson)
- An official source cited for every claim
- An AI tutor that answers from the guide
- Unlimited timed practice exams