Salesforce Architect Notes

1/16

Salesforce Interview Notes · Lead / Architect

★ARCHITECTURE & DESIGN PATTERNS★

Skew, big objects, trigger frameworks, LDV, migration

⚠️Data Skew

Skew = too many child/related records tied to ONE parent record
Causes record-locking errors + slow queries + sharing recalc pain
Account skew: >10,000 child records (Contacts/Cases/Opps) on one Account
Ownership skew: >10,000 records of one object owned by one user
Lookup skew: >10,000 records pointing at same lookup target value
Classic symptom: Approval Process waiting on manager ⟶ parent row lock
Parent lock cascades: updating child locks the parent record
Fixes: distribute across parents, "bucket" accounts, integration user w/ no role
Ownership skew fix: put owner OUTSIDE role hierarchy, no public groups
Also: reduce batch size to 1, sort by ParentId to avoid concurrent locks

🗄️Big Objects

Stores billions of records outside standard object storage
Standard big object = FieldHistoryArchive (field history archival)
Custom big object suffix __b; defined by index at creation time
Index: up to 5 fields, queried LEFT-TO-RIGHT, cannot skip fields
Index is IMMUTABLE after deploy ⟶ redesign = new object
No triggers, flows, validation rules, process builder on big objects
Insert via Apex insertImmediate(), Bulk API, or CSV import
Read via SOQL on index fields only; Async SOQL for bulk analytics
Use for: audit trails, IoT/event history, 360 archive, compliance

🧩Trigger Frameworks

Goal: ONE trigger per object, zero logic in the trigger body
Trigger Handler pattern: Apex class holds all logic, trigger just delegates
Virtual base class: overridable beforeInsert(), afterUpdate() hooks
Interface-based: handlers implement common interface (rare in practice)
Handler dispatches on Trigger.isBefore / isAfter / isInsert context
Recursion control: static Boolean / Set<Id> of processed record Ids
Bypass switch: Custom Metadata or hierarchy Custom Setting per handler
Keep logic bulkified: no SOQL/DML inside for loops
Popular: Kevin O'Hara sfdc-trigger-framework, fflib Domain layer

⚖️Trigger vs Declarative

Default: Flow first — declarative is the platform direction
Record-Triggered Flow replaced Workflow Rules + Process Builder (retired)
Use Apex trigger for: complex logic Flow cannot do
Apex needed for: custom error handling, redirects, callouts pre-commit
Apex needed for: cross-object rollups w/o master-detail, recursive logic
Apex needed for: large-volume bulk processing, complex sorting/matching
Don't mix Flow + trigger on same object/same operation ⟶ order chaos
Before-save Flow is ~10x faster than Process Builder (no extra DML)

📈Large Data Volumes

Batch Apex + Scheduled Apex to process records asynchronously
Batch scope default 200, max 2000; 5 concurrent batch jobs
Selective SOQL is mandatory: filter on indexed field
Standard index selective: <30% of first 1M rows, <15% after, cap 1M
Custom index selective: <10% of rows, hard cap 333,333 rows
Skinny tables: read-only denormalized copy, max 100 columns, no joins
Avoid NULL / != / NOT / leading-wildcard LIKE ⟶ kills index
Divisions partition very large orgs; Custom Indexes via Salesforce support
PK Chunking for Bulk API extracts of huge objects
Defer sharing recalculation during mass loads
Change Data Capture to propagate changes out, not nightly full pulls
Archive cold data to Big Objects or external store (Salesforce Connect)

🚚One-Time Migration

Tooling: Data Loader / ETL (Informatica, MuleSoft) over Bulk API
Bulk API 2.0: 150M records / 24 hrs; auto-chunked server-side
Load PARENTS before children; use External Id as match key
External Id upsert avoids extra lookup queries + duplicate rows
Deactivate triggers, flows, validation rules, workflow during load
Load in parallel batches; serial mode if you hit lock contention
Sort file by parent Id to prevent concurrent parent locks
Defer sharing calc + turn off role hierarchy grants if possible
Re-enable automation, then reconcile record counts + error files
Dry-run in Full sandbox first; capture success/error CSVs

🛡️Exception Framework

Custom logging object e.g. ACE_Exception_Log__c stores exception details
Capture: class, method, line, getStackTraceString(), record Ids, user
Wrap logic in try/catch; never swallow silently, always log
Log DML with allOrNone=false ⟶ Database.SaveResult errors captured
Publish Platform Event for logging so log survives a rollback
Custom exception: class MyException extends Exception
addError() on record for user-facing validation failure
Dashboard/report on log object + alert on severity

🏛️MVC & Conventions

Model = objects, fields, relationships (the data layer)
View = LWC, Aura, Visualforce page, Lightning App Builder
Controller = Apex class holding business logic
Naming: CamelCase Java style; PascalCase classes, camelCase methods
Suffix by role: XController, XService, XSelector, XHandler, XTest
Data Warehouse = consolidate transactional data centrally for analysis
SF is source system ⟶ warehouse (Snowflake/BigQuery) for cross-system BI
Integration design phases: discover ⟶ data model ⟶ pattern ⟶ security ⟶ volume ⟶ error handling ⟶ cutover

⇄ Quick Comparison ⇄

Skew Types & Thresholds

Skew Type	Threshold	Primary Fix
Account (data) skew	>10,000 children per Account	Distribute across parent buckets
Ownership skew	>10,000 records one owner	Owner outside role hierarchy
Lookup skew	>10,000 records same lookup value	Spread values / reduce automation
Symptom	UNABLE_TO_LOCK_ROW	Batch size 1, sort by ParentId

Flow vs Apex Trigger

Record-Triggered Flow	Criterion	Apex Trigger
Declarative, admin-owned	Skill needed	Code, dev + test class
Simple field updates, rollups	Best for	Complex, bulk, recursive logic
Limited, per-element faults	Error handling	Full try/catch + custom errors
Slower at high volume	LDV performance	Bulkified, best for millions
Before-save = fastest option	Speed	Fast, but needs discipline

Easy way to remember

SKEW = 10K is the magic number for all three types
ONE trigger per object, ZERO logic inside it
Migration: Parents first, automation OFF, External Ids as keys

Interview tips

Q. What number defines data skew?

A. 10,000 — child records per parent, records per owner, or per lookup value.

Q. Name a standard Big Object.

A. FieldHistoryArchive. Index is set at creation and can never be changed.

Q. Trigger or Flow?

A. Flow by default; Apex for complex bulk logic, callouts, and real error handling.

Takeaway

Design for the 10,000-record cliff — skew, selectivity, and one clean trigger handler are what separate Lead from Senior.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

2/16

Salesforce Interview Notes · Lead / Architect

★SALESFORCE INTEGRATION★

Patterns, APIs, auth, callout limits, events

🔀6 Integration Patterns

Remote Process Invocation - Request & Reply: SF calls out, waits, tracks state
Remote Process Invocation - Fire & Forget: remote acks, control returns immediately
Batch Data Synchronization: bulk both directions, scheduled, not real time
Remote Call-In: external system does CRUD on SF data (inbound)
UI Update Based on Data Changes: SF UI auto-refreshes on data change
Data Virtualization: real-time external read, no persistence, no reconciliation
Virtualization = Salesforce Connect + external objects (OData 2.0/4.0 or Apex adapter)
Fire & Forget best implemented with Platform Events, not a callout

⏱️Callout Limits

100 callouts per Apex transaction (sync + async combined)
Default timeout 10 seconds per callout; max settable 120 seconds
120 seconds cumulative timeout for ALL callouts in one transaction
API calls: ~250,000 / 24h on Professional + Enterprise (buy more)
@ReadOnly doubles query rows to 1,000,000 - read-only endpoints
SOQL in REST endpoints: LIMIT 50000 rows max
Messaging.sendEmail 10x per transaction; 5,000 single emails/day
225,000 workflow emails/day - avoid emailing from integration code

🔐Named Creds vs Remote Site

Named Credential = endpoint URL + auth (user/pass, OAuth token) in one
Remote Site Setting = whitelist only, no credentials stored
Callout syntax: req.setEndpoint('callout:My_Named_Cred/services/data/v58.0/...')
Named Credential ⟹ no secrets in code, no Remote Site entry needed
Remote Site still required for raw https:// endpoints + WSDL2Apex stubs
Endpoint in a Custom Label = per-environment switching without deploy

🔑OAuth + Auth Flows

Flows: Web Server, User-Agent, JWT Bearer, Username-Password, Refresh Token, Device
OAuth = grant access without sharing passwords; used by Connected Apps
Connected App: SAML + OAuth, SSO, tokens, IP relaxation, refresh token policy
JWT Bearer = server-to-server, no user interaction, cert on connected app
JWT: POST to /services/oauth2/token, grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer
Returns access_token + instance_url ⟶ always use returned instance_url
Pass JWT in Authorization: Bearer header (preferred) not query string
Query-string tokens leak into logs, browser history, referrer headers
Postman password flow: username, password+token, client_id, client_secret, grant_type=password

📥Apex REST (Inbound)

@RestResource(urlMapping='/MyService/*') - class must be global, path mandatory
URI always /services/apexrest/<mapping>/
@HttpGet select | @HttpPost create/upsert | @HttpPut insert-or-replace
@HttpPatch update | @HttpDelete delete
Each verb annotation may appear only ONCE per class
RestRequest req = RestContext.request; RestContext.response for output
RestRequest holds requestBody, params, headers, requestURI
Response: RestContext.response.responseBody = Blob.valueOf(jsonStr)
REST in SF is stateless, XML + JSON supported, JSON is default

📤Outbound Callout

Http h = new Http(); HttpRequest req; h.send(req) ⟶ HttpResponse
setEndpoint / setMethod / setHeader('Content-Type','application/json') / setBody
Build body with JSON.serialize(wrapper), never string concatenation
JSON.deserializeUntyped ⟶ Map<String,Object>; JSON.deserialize(body, Wrapper.class) when typed
No callouts inline from a trigger - callout blocks the transaction
Use @future(callout=true) or Queueable + Database.AllowsCallouts
@future takes primitives only - pass Ids as Strings, re-query inside
Static Boolean flag stops recursion when callout result updates same records
Bulkify: collect Ids in trigger, ONE async call, not one per record
Test callouts with a mock class holding the canned response

📡Events + Streaming

Pub/sub on the event bus; publisher + subscriber fully decoupled
Publish via EventBus.publish(), Flow, or external API - a publish is NOT a callout
Subscribe: Apex triggers, Flow, LWC empApi, external CometD clients
Platform Event: retained 24-72 hours, replayable by ReplayId
CDC: near real-time create/update/delete/undelete, retained 3 days
CDC channel: /data/AccountChangeEvent, /data/Employee__ChangeEvent, /data/ChangeEvents (all)
CDC perms: View All Data + View All Users for /data/ChangeEvents
Streaming API PushTopic = notifications from a SOQL query you define
Don't use events for synchronous response or guaranteed ordered request/reply

🛡️Error Handling

Savepoint sp = Database.setSavepoint(); Database.rollback(sp) in catch
Database.insert(list, false) / Database.upsert() = partial success allowed
Loop SaveResult[], isSuccess() false ⟶ log to API_Error_Log__c
ALWAYS set a response body - success path AND failure path
Wrap the whole method in try/catch; return JSON error, never a raw stack
Log every call (request + response + status) to an Integration_Tracker__c
Flow fault path ⟶ Integration_Log__c + notify integration team
Test endpoints from Workbench REST Explorer or Postman

🏗️MuleSoft + Volume

MuleSoft = Java ESB: routing, JDBC, DataWeave, API management, connectors
Connects to SF via Anypoint Platform Salesforce Connector
Heroku = PaaS for hosting apps, NOT an integration platform - you code it yourself
Huge volumes: Bulk API (async, batched) over row-by-row REST/SOAP
Big Objects for storage without standard data storage; Batch Apex to process
Data Loader uses Bulk API for insert/update/upsert/delete/export
View-only external data ⟶ Salesforce Connect, zero storage consumed

⇄ Quick Comparison ⇄

REST vs SOAP API

REST API	Aspect	SOAP API
URI / URL resources	Addressing	WSDL contract
JSON (default) + XML	Format	XML only
Stateless, loosely typed	Typing	Strongly typed (Enterprise WSDL)
GET POST PATCH PUT DELETE HEAD	Verbs	Single POST envelope
Mobile, web, lightweight	Best for	Legacy, middleware, contract-driven
Defined by Roy Fielding, 2000	Origin	Simple Object Access Protocol

Platform Event vs CDC

Platform Event	Aspect	Change Data Capture
Custom __e payload you define	Payload	Record deltas + changed fields
You publish explicitly	Trigger	Automatic on DML
24-72 hours replay	Retention	3 days replay
Custom app-to-app messaging	Use case	Keep external system in sync

Easy way to remember

R-F-B-R-U-D = Request/reply, Fire&forget, Batch sync, Remote call-in, UI update, Data virtualization
100 callouts, 120s total, 10s default
Named Cred = URL + secret; Remote Site = URL only

Interview tips

Q. Why can't a trigger call out directly?

A. Callout would block the transaction; use @future(callout=true) or Queueable with AllowsCallouts.

Q. Enterprise vs Partner WSDL?

A. Enterprise is strongly typed to your org schema; Partner is loosely typed, introspected, multi-org.

Q. Full CRUD on warehouse data, native UI, no storage?

A. Salesforce Connect with writeable external objects via OData or Apex adapter.

Takeaway

Pick the pattern first, then the tech - real time means callout or Connect, decoupled means Platform Events, volume means Bulk API.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

3/16

Salesforce Interview Notes · Lead / Architect

★SECURITY & SHARING MODEL★

OWD, sharing, profiles, perm sets, Apex enforcement

🔒OWD Baseline

OWD = most restrictive floor; everything else only OPENS access
Private / Public Read Only / Public Read-Write
Public Read/Write/Transfer ⟶ Lead + Case ONLY
Public Full Access ⟶ Campaigns only
Controlled by Parent ⟶ detail side of master-detail, NOT editable
Detail object has no Owner ⟶ no OWD, no sharing rules on it
3 columns: internal | external | guest
External access can NEVER exceed internal access
Internal Private ⟶ external must be Private too
Guest users: Read only, free licence, own guest profile

🏛️Role Hierarchy

Shares UPWARD only: manager sees records of users below
Peers do NOT see each other ⟶ use sharing rules / public groups
Needs OWD stricter than Public Read/Write + Grant Access Using Hierarchies
Grant Access Using Hierarchies: locked ON for standard objects
Can only be unchecked for CUSTOM objects
Role is OPTIONAL; profile is MANDATORY (exactly one)
User can change own role; CANNOT change own profile
Deleting a role ⟶ reassign users first, sharing recalculation runs

🔁Sharing Rules

Owner-based: by owner user / role / role+subordinates / public group
Criteria-based: by field values, ignores owner
Only work when OWD = Private or Public Read Only
Grant ONLY - can never restrict below OWD
Apply to existing AND new records, automatically
Access levels: Read Only or Read/Write
Limit: 300 sharing rules per object = 250 owner + 50 criteria
Share with: Roles, Roles & Subordinates, Public Groups
Public Group = "perm set for roles"; group types: Public + Personal
Delete a group/queue ⟶ references shown, recalculation, access lost

🤝Manual + Apex Sharing

3 kinds: managed (SF), user managed (manual), Apex managed
Manual: Sharing button; owner, Full Access user, or admin
Manual only when OWD more restrictive than Public Read/Write
Share object: AccountShare, OpportunityShare, MyObj__Share
__Share auto-created; NOT if OWD = Public Read/Write
Custom fields: ParentId, UserOrGroupId, AccessLevel, RowCause
Standard: OpportunityAccessLevel = 'Read'|'Edit'|'All' (never 'Read/Write')
Custom RowCause (Apex sharing reason) survives owner change
Custom RowCause maintained only by Modify All Data
Database.insert(shares, false) ⟶ partial success

🪪Profiles & Perm Sets

Profile = functional access; 1 per user, many users per profile
Cannot build a profile from scratch ⟶ clone an existing one
Standard profiles not fully editable / not deletable
Perm sets ADD only; never restrict. Many per user
Licence-based perm set ⟶ only users of that licence; "none" ⟶ anyone
Only Permission Set Group MUTING can remove a permission
PSG statuses: Updated / Outdated / Updating / Failed
Permissions are cumulative = union of profile + all perm sets
Read/Edit profile + Edit/Delete perm set ⟶ Read+Edit+Delete
Minimum Access - Salesforce = modern recommended baseline profile

🧱Object + Field Security

4 layers: Org ⟶ Object ⟶ Field ⟶ Record
Org: login IP ranges, login hours, password policies, MFA
Object: CRED + View All + Modify All (profile or perm set)
FLS set on profiles/perm sets; applies to layouts, list views, reports, API
Admin profile checkbox lets you edit read-only fields
Transfer Record perm ⟶ change owner with only Read access
App perm "View and Edit Converted Leads" for converted leads
Reports/dashboards/documents: access = FOLDER access
Users are never deleted ⟶ Freeze (keeps licence + records) or Deactivate (releases licence)
Login IP Ranges set ⟶ "Reset Security Token" option disappears

⚙️Apex Sharing Keywords

with sharing ⟶ enforce running user's RECORD-level sharing
without sharing ⟶ system mode, sees all records
inherited sharing ⟶ takes caller's mode; defaults to with sharing if entered directly
inherited sharing = safest for reusable service classes
Keywords control record level ONLY - not FLS, not object CRUD
Flow: System Context bypasses OLS/FLS ⟶ run in user context to respect it
Lightning/Aura: also need Apex Class Access + component visibility filters

🛡️SOQL Security + Access Check

WITH SECURITY_ENFORCED ⟶ checks OLS+FLS in SELECT and FROM
Throws System.QueryException instead of returning data
WITH USER_MODE + stripInaccessible are the newer alternatives
Security.stripInaccessible(AccessType.READABLE, recs) ⟶ strips, no throw
AccessType.CREATABLE ⟶ sanitize before insert; decision.getRecords()
Describe check: Schema.sObjectType.Contact.fields.Email.isAccessible()
UserRecordAccess (API v24.0+) ⟶ HasEditAccess / HasReadAccess
UserRecordAccess: max 200 record Ids, must be a LIST not a SET
Credentials live in Named Credentials, never in Apex
Auth: SSO via SAML/OAuth 2.0, Social Sign-On, MFA (TOTP/SMS)

⇄ Quick Comparison ⇄

Profile vs Permission Set

Profile	Feature	Permission Set
Exactly one, mandatory	Per user	Many, optional
Grant AND restrict	Effect	Grant only (add)
Clone only, no new	Creation	Create new freely
Tied to licence type	Licence	Licence-based or none
Baseline access	Purpose	Extra rights on top

Enforcing FLS in Apex

WITH SECURITY_ENFORCED	Feature	stripInaccessible()
Inline in SOQL	Where	Method on result/DML list
Throws QueryException	On violation	Silently removes fields
SELECT + FROM clauses	Scope	READABLE / CREATABLE / UPDATABLE
Read queries	Best for	Sanitising records before DML

Easy way to remember

OWD is the FLOOR - hierarchy, rules, manual, Apex only build UP
Sharing keywords = records; SECURITY_ENFORCED = objects + fields
300 sharing rules = 250 owner + 50 criteria

Interview tips

Q. Can a sharing rule restrict access?

A. No. Sharing only widens access above OWD; nothing lowers the OWD floor.

Q. Which sharing declaration for a reusable service class?

A. inherited sharing - takes caller's mode, defaults to with sharing if entered directly.

Q. Does with sharing enforce FLS?

A. No. Record level only. Use WITH SECURITY_ENFORCED, USER_MODE or stripInaccessible.

Takeaway

OWD sets the floor, everything else opens it up - and Apex sharing keywords protect records, not fields.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

4/16

Salesforce Interview Notes · Lead / Architect

★ASYNCHRONOUS APEX★

Future, Queueable, Batch, Scheduled - limits and gotchas

🧭Pick Your Async

Future ⟶ callout from trigger + fix Mixed DML. Fire-and-forget.
Queueable ⟶ chaining, complex types, need a job Id.
Batch ⟶ millions of records, fresh limits per chunk.
Scheduled ⟶ run any of the above at a set time.
Rule: >1 batch of records ⟶ Batch; else Queueable.
Queueable = modern replacement for @future. Default to it.
All four get higher (often doubled) governor limits.
All four can be started from a trigger.
Flow: Request ⟶ Enqueue ⟶ Persistence ⟶ Dequeue.
Timing never guaranteed: server + queue depth decide.

⏭️@future

MUST be static + return void. No return value ever.
Params: primitives / arrays / collections of primitives only.
No sObjects: value may go stale + overwrite newer data.
Workaround: pass List<Id>, re-query inside. Or JSON.serialize wrapper.
@future(callout=true) or CalloutException. Default is false.
50 calls per transaction; 250,000 per 24h (or licences x 200).
Cannot chain: future ⟶ future = not allowed.
Cannot be called from Batch start/execute/finish.
Callable from trigger, Apex class, Schedulable. 1 from Queueable.
No job Id ⟶ cannot monitor. Query AsyncApexJob by MethodName/JobType.
Banned in VF constructor, getMethodName(), setMethodName().
Order not guaranteed; two can run concurrently ⟶ record locking.
Process Builder? Wrap it in an @InvocableMethod.

📦Queueable

implements Queueable ⟶ one method: execute(QueueableContext).
System.enqueueJob() returns AsyncApexJob Id ⟶ monitorable + abortable.
Accepts non-primitive members: sObjects, custom Apex types.
50 enqueueJob per SYNC transaction; only 1 in async context.
Exceeded ⟶ LimitException "Too many queueable jobs added to the queue".
Chaining: 1 child per parent. Stack depth 5 in Dev/Trial orgs.
Callouts ⟶ implements Database.AllowsCallouts.
Non-callout job CAN chain into an AllowsCallouts job.
Check Limits.getQueueableJobs() vs getLimitQueueableJobs() before enqueue.
Need >1 queueable from batch execute? Schedule them instead.
No chaining inside tests ⟶ guard with Test.isRunningTest().
Batches processed / total batches always 0.

🗂️Batch Apex

implements Database.Batchable<sObject>. 3 methods, all required.
start(BatchableContext) ⟶ Database.QueryLocator OR Iterable<sObject>.
execute(BatchableContext, List<sObject> scope) ⟶ the work.
finish(BatchableContext) ⟶ email, chain next job.
start + finish run ONCE. execute runs ceil(records / scope size).
Default scope 200, max 2,000. Iterable scope = no upper limit.
QueryLocator = 50M rows (bypasses row limit). Iterable = 50,000.
Fresh limits per execute: 200 SOQL, 50k rows, 150 DML, 10k DML rows.
Rename execute ⟶ execute1 and the class will not compile.
Database.executeBatch(cls, scope) returns Id ⟶ System.abortJob().
FOR UPDATE banned in QueryLocator - locking is implied.
Aggregate queries fail (no queryMore) ⟶ Iterable<AggregateResult>.
External objects ⟶ must use Iterable, not QueryLocator.
Only one batch start() runs at a time per org.

💾State + Failure

Batch is STATELESS by default: fresh object per execute.
Database.Stateful ⟶ instance member vars survive between chunks.
Cost: serialize/deserialize state every execute. Tiny scope = pain.
1 record fails ⟶ whole 200 fails, but next batch still runs.
Committed batches are NEVER rolled back by a later failure.
Database.update(list, false) ⟶ partial commit, only bad row fails.
Database.RaisesPlatformEvents ⟶ fires BatchApexErrorEvent on error.
Subscribers handle the event ⟶ retry/log without touching the batch.
Track results via successRecord/failRecord Sets + email in finish().

⏰Scheduled + Cron

implements Schedulable ⟶ one method: execute(SchedulableContext sc).
System.schedule('name', cronExp, obj) returns a CronTrigger Id.
System.scheduleBatch() = schedule a batch with no Schedulable class.
Cron = 7 fields: sec min hr day_of_month month day_of_week [year].
Sec/Min 0-59, Hours 0-23, DOM 1-31, Month 1-12, DOW 1-7/SUN-SAT, Year 1970-2099.
DOM and DOW cannot BOTH be concrete - one must be ?.
'0 0 0 ?' = daily midnight. '0 10 17 ? * MON-FRI' = 5:10pm weekdays.
'0 30 0 1 1 ? *' = 00:30 on Jan 1 every year.
100 scheduled jobs active; 250,000 scheduled executions / 24h.
Lightning UI cannot set minutes/seconds - only a cron expression can.
No SYNC callout from Schedulable.execute(). Delegate to batch/queueable/@future.
@future CAN be called from a scheduler. Batch too.

📊Flex Queue + Monitor

5 batch jobs queued or active concurrently, per org.
Flex queue holds 100 more in Holding status. 101st ⟶ LimitException.
Statuses: Holding, Queued, Preparing, Processing, Aborted, Completed, Failed.
FIFO, but reorder via System.FlexQueue.moveBeforeJob() / moveAfterJob().
Batch + Queueable ⟶ query AsyncApexJob (Status, NumberOfErrors, JobItemsProcessed, TotalJobItems, ExtendedStatus).
Scheduled ⟶ query CronTrigger (TimesTriggered, NextFireTime) + CronJobDetail.
ctx.getTriggerId() gives the running scheduled job's Id.
UI: Setup ⟶ Monitoring ⟶ Apex Jobs. Flex order: Setup ⟶ Apex Flex Queue.
Tests: async inside startTest/stopTest runs sync + not counted.

⚠️Who Can Call What

Batch from another batch: finish() ONLY.
From start/execute ⟶ System.AsyncException: Database.executeBatch cannot be called...
Batch from trigger: legal, bad practice - 5-job limit blows up fast.
Future from batch: NEVER. Enqueue a Queueable from finish() instead.
Loophole: batch execute ⟶ web service ⟶ that service calls @future.
Cursors: start = 15 open per user; execute + finish = 5 each.
Callouts: 100 per transaction; in batch, 100 per execute().
Mixed DML = setup (User, Group, Queue, PermSet) + non-setup in one txn.
Fix: push the setup-object DML into a @future method.

⇄ Quick Comparison ⇄

Future vs Queueable

@future	Aspect	Queueable
Static void method	Shape	Class, execute(QueueableContext)
Primitives only	Params	sObjects + custom types OK
None returned	Job Id	AsyncApexJob Id from enqueueJob
Impossible	Chaining	1 child, depth 5 (Dev/Trial)
@future(callout=true)	Callouts	implements Database.AllowsCallouts
50 per transaction	Txn limit	50 sync / 1 async
Not monitorable	Monitoring	Setup + SOQL + abortable
Mixed DML, legacy	Best for	Everything else

Hard Limits

Limit	Applies To	Value
Async executions / 24h	All async	250,000 or licences x 200
Concurrent batch jobs	Batch	5 queued or active
Flex queue (Holding)	Batch	100
Scope size	Batch execute	200 default, 2,000 max
Records returned	QueryLocator vs Iterable	50,000,000 vs 50,000
Active scheduled jobs	Schedulable	100
Query cursors	start vs execute/finish	15 vs 5
Callouts	Per txn / per execute	100

Easy way to remember

FQBS = Fire, Queue, Bulk, Schedule
5 batches, 50 futures, 100 scheduled, 200 scope
50M QueryLocator, 50K Iterable

Interview tips

Q. Can a batch call a future method?

A. No. Use System.enqueueJob() from finish() instead.

Q. Why no sObjects in @future?

A. Record can change before it runs - stale values would overwrite newer data.

Q. Batch or Queueable?

A. More than one batch of records ⟶ Batch. Otherwise Queueable.

Takeaway

Queueable is the default async; Batch only when one transaction's limits truly cannot hold the data.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

5/16

Salesforce Interview Notes · Lead / Architect

★APEX CORE, TRIGGERS & GOVERNOR LIMITS★

Context vars, order of execution, limits, bulkification

⚡Trigger Context Vars

Trigger.new ⟶ List of new sObjects; insert/update/undelete only
Trigger.old ⟶ List of old versions; update/delete only; ALWAYS read-only
Trigger.newMap / oldMap ⟶ Map<Id, sObject>; oldMap = update+delete only
Trigger.newMap NULL in before insert ⟶ no Ids yet ⟶ NPE
Trigger.new editable ONLY in before triggers; read-only in after
Trigger.new list itself read-only ⟶ can't add/remove, only set fields
isExecuting, isInsert, isUpdate, isDelete, isUndelete, isBefore, isAfter
Trigger.size + Trigger.operationType (System.TriggerOperation enum)
7 events: before ins/upd/del + after ins/upd/del + after undelete
NO before undelete event exists
upsert fires 4 events: before+after insert, before+after update
merge ⟶ delete events on losers + update events on winner
Lead convert DOES fire insert triggers on Account/Contact/Opportunity

⏱Before vs After

BEFORE ⟶ validate + default fields on SAME record, no extra DML
BEFORE ⟶ never issue insert/update on Trigger.new ⟶ recursion/error
AFTER ⟶ record saved (not committed); Id, CreatedDate, LastModifiedDate available
AFTER ⟶ create/update related + child records, callouts (async), emails
Need the Id or newMap? ⟶ after insert
Lookup child cascade delete ⟶ after delete trigger, query by parent Id
addError() on record or field ⟶ custom trigger error message

🔢Order of Execution

1. Load record from DB / init from submitted values
2. New values overwrite old
3. System validation (required, format, length) + layout rules
4. BEFORE triggers
5. Custom validation + duplicate rules + system validation again
6. Save to DB, NOT committed
7. AFTER triggers
8-9. Assignment rules ⟶ auto-response rules
10. Workflow rules; field update ⟶ before/after update triggers refire ONCE
11-12. Escalation ⟶ Processes / flows (after-save record-triggered)
13-14. Entitlement ⟶ roll-up summary recalc, parent re-saves, then grandparent
15. Criteria-based sharing recalculated
16. COMMIT all DML
17. Post-commit: email send, @future + queueable enqueued

🚦Limits & Bulkification

Limits reset at start of each transaction; multi-tenant fairness
LimitException is UNCATCHABLE ⟶ try/catch will not save you
"Too many SOQL queries: 101" ⟶ hard limit, support cannot raise it
@future daily cap throws System.AsyncException ⟶ also uncatchable
Limits.getQueries(), Limits.getDMLRows() ⟶ check consumption pre-emptively
NEVER SOQL or DML inside a for loop ⟶ collect Ids in Set, query once, DML once
Bulkify helper methods too, not just the trigger
>50,000 rows ⟶ Batch Apex; QueryLocator returns up to 50 million, scope 200
SOQL for loop for (Account a : [SELECT...]) chunks 200 ⟶ saves heap, not rows
Move heavy work to Batch / Queueable / @future for higher limits
No element cap on List/Set/Map ⟶ heap size is the real ceiling
Test with 200 records, positive AND negative paths

🔁Recursion Control

Static Boolean flag ⟶ simple, but only safe for <200 records
200+ records = multiple batches ⟶ flag blocks batches 2..n silently
PREFERRED: static Set<Id> of processed Ids ⟶ bulk-safe
Compare oldMap vs new field value ⟶ run only on real change
Static Map<Id, value> to detect genuine field deltas
Unchecked recursion ⟶ CPU time-out errors
Watch Apex + Flow/Process re-triggering each other

💾DML vs Database Methods

insert/update/delete/upsert/merge/undelete ⟶ ALL-or-nothing, throws DmlException
Database.insert(list, false) ⟶ partial success, no exception thrown
allOrNone defaults to TRUE = same behaviour as plain DML
Results: SaveResult (insert/update), UpsertResult, MergeResult, DeleteResult, UndeleteResult
Loop results ⟶ isSuccess(), getId(), getErrors() ⟶ getStatusCode/getMessage/getFields
merge only on Lead, Contact, Account ⟶ NOT custom objects
Hard delete ⟶ delete list; then Database.emptyRecycleBin(list)
Row locking ⟶ SELECT ... FOR UPDATE
Rollback ⟶ Database.setSavepoint() / rollback
Mixed DML: setup (User, Group, GroupMember, PermissionSetAssignment, Queue) + non-setup in one txn ⟶ MIXED_DML_OPERATION; split via @future/Queueable, System.runAs in tests

📦Collections & Classes

List = ordered, indexed, duplicates allowed
Set = unordered, unique; primitives + sObjects, NOT collections
Map = key-value; null key allowed; duplicate key OVERWRITES; String keys case-sensitive
Dedupe ⟶ Set.addAll(list) then new List<>(set); or SOQL GROUP BY
Collections passed by REFERENCE; primitives passed by VALUE
static ⟶ initialised once at class load, belongs to class, not in VF view state
static method cannot use instance vars; instance method can use both
private < protected < public < global; private is the default
global needed for @RestResource, webservice, managed-package API; can't be removed once packaged
virtual class CAN be instantiated, abstract CANNOT; abstract methods need override
Interface = 100% abstraction; abstract class = 0-100%
with sharing = enforce user's sharing rules; without sharing = system context
inherited sharing = take caller's context; safe default for @AuraEnabled/VF

💣Gotchas

NO callouts from a trigger ⟶ use @future(callout=true) or Queueable + Database.AllowsCallouts
SOSL in triggers IS now allowed
Return type mandatory on every method (void if none); constructors exempt
Field default values NOT applied by new ⟶ sObjectType.newSObject(rtId, true)
VF controller constructor cannot do DML or callouts ⟶ move to action method
"List has no rows for assignment to SObject" ⟶ query returned zero rows
Custom exception class name must end in 'Exception' and extend Exception
Long = 64-bit (-2^63..2^63-1); Integer = 32-bit
@InvocableMethod: public/global + static, ONE per class, ONE List param, no triggers
Blob.toString() / Blob.valueOf(); System.Label.X; UserInfo.getUserId()
Safe navigation ?. beats chained null checks
One trigger per object; zero logic in trigger, all in handler; no hardcoded Ids

⇄ Quick Comparison ⇄

Context Var Availability

Event	Trigger.new / newMap	Trigger.old / oldMap
Before Insert	new YES / newMap NO	NO / NO
After Insert	YES / YES	NO / NO
Before Update	YES / YES	YES / YES
After Update	YES / YES	YES / YES
Before Delete	NO / NO	YES / YES
After Delete	NO / NO	YES / YES
After Undelete	YES / YES	NO / NO

Governor Limits Per Transaction

Limit	Synchronous	Asynchronous
SOQL queries	100	200
Rows retrieved by SOQL	50,000	50,000
SOSL queries (2,000 rows each)	20	20
DML statements	150	150
Records processed by DML	10,000	10,000
CPU time	10,000 ms	60,000 ms
Heap size	6 MB	12 MB
Callouts	100	100
@future calls	50	50
Queueable jobs enqueued	50	50
Email invocations	10	10
Mobile push calls	10	10

Easy way to remember

100 SOQL / 50k rows / 150 DML / 10k rows / 6MB / 10s -- async doubles heap + SOQL, 6x CPU
BEFORE = same record, no Id. AFTER = other records, has Id
newMap is null in before insert

Interview tips

Q. Why is Trigger.newMap null in before insert?

A. Records have no Ids until saved; use after insert if you need Ids.

Q. Best way to stop trigger recursion?

A. Static Set<Id> of processed Ids in a helper class -- bulk-safe unlike a static Boolean.

Q. Plain DML vs Database.insert?

A. DML is all-or-nothing and throws DmlException; Database.insert(list,false) allows partial success via SaveResult.

Takeaway

One trigger per object, logic in a handler, collections outside loops -- every governor limit is per transaction and every LimitException is uncatchable.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

6/16

Salesforce Interview Notes · Lead / Architect

★SOQL, SOSL & DATA MODEL★

Queries, relationships, indexes and governor limits

🔍SOQL Core

Salesforce Object Query Language ⟶ retrieve records of one sObject + relatives
Syntax: SELECT f1, f2 FROM Object [WHERE cond]
No SELECT * ⟶ must list every field
Returns List<sObject>, single sObject, or Integer (COUNT())
Static SOQL = inline in [ ] + bind vars :myVar / IN :myIds
Dynamic SOQL = Database.query('...') built at runtime
Clauses: WHERE, GROUP BY, HAVING, ORDER BY, LIMIT, OFFSET
OFFSET 8 ⟶ returns from record 9 onward; max OFFSET 2,000
ALL ROWS ⟶ includes recycle bin + archived activities
Semi-join IN (SELECT...) / anti-join NOT IN ⟶ SOQL has no real JOINs

🔎SOSL Core

Salesforce Object Search Language ⟶ full-text search across many objects
FIND {Acc1} IN ALL FIELDS RETURNING Account(Name), Contact(Name) LIMIT 4
Apex uses quotes 'map'; SOAP/REST API uses braces {map}
Scopes: ALL / NAME / EMAIL / PHONE / SIDEBAR FIELDS
Wildcards: * = many chars, ? = single char; OR combines terms
Returns List<List<sObject>> in the order objects were listed
No match ⟶ empty list for that sObject, not null
Cast out: Account[] a = (List<Account>)searchList[0];
Allowed in triggers + classes since API v20.0 (Winter '11)
No DML directly on search results; text must be indexed to be found

🔗Relationship Queries

Child ⟶ Parent = dot notation, custom rel uses __r not __c
SELECT Name, Account.Industry FROM Contact
SELECT Position__r.Salary__c FROM Candidate__c
Parent ⟶ Child = subquery on plural relationship name
SELECT Id, (SELECT Id FROM Contacts) FROM Account
Custom child list: (SELECT Id, Name FROM Candidates__r) FROM Position__c
Limits: 20 parent-to-child rels/query, only 1 level of nesting
Limits: 35 child-to-parent rels/query, max 5 levels deep
Contact.Account.Owner.FirstName = legal 3-level traversal
Subquery children come back as a list on each parent record

🧩Relationship Types

Lookup ⟶ loose child-parent, child keeps own owner + sharing, one-to-many relationship.
Master-Detail ⟶ Mandatory field, tight; detail inherits master's ownership/sharing
Many-to-many ⟶ junction object = 2 master-detail fields
First MD field created = primary relationship (drives layout/ownership)
Self relationship = lookup to the same object (Account ⟶ Account)
Hierarchical = special lookup, User object only
Standard object can NEVER be the detail side; master only
Reparenting allowed only if "Allow reparenting" ticked on MD field
Cascade delete: MD always; lookup optional, bypasses validation rules
Delete Account ⟶ its Contacts go to Recycle Bin too

📊Roll-Up & Convert

Roll-up summary = aggregate of children onto the parent record
Master-detail ONLY (plus some standard rels e.g. Account-Opportunity)
Types: COUNT, SUM, MIN, MAX ⟶ read-only field on master
Limit 25 roll-ups per object, raisable to 40 by Support
Formula = within one record; roll-up = across child records
Lookup ⟶ MD conversion: every child must already have a parent
MD ⟶ Lookup conversion: delete all roll-up summaries on parent first

⚡Indexes & Selectivity

Indexed by default: Id, Name, OwnerId, CreatedDate, SystemModstamp
Also indexed: Master-Detail + Lookup FKs, External ID, Unique fields
Selective = filter under threshold; else full table scan / timeout
Standard index: <30% of first 1M rows, <15% after, cap 1,000,000
Custom index: <10% of first 1M rows, <5% after 1M, cap 333,333
Kills index: leading %wildcard, !=, NOT, OR on unindexed, NULL on some
Query plan tool in Dev Console ⟶ cost <1.0 means selective
Skinny tables + custom indexes via Support for Large Data Volumes
>200k rows ⟶ selectivity is mandatory, not optional

🧮Aggregates & Counting

SELECT COUNT() FROM Obj WHERE ... ⟶ server-side, dodges 50k row cap
Aggregates: COUNT(), COUNT(Id), SUM, AVG, MIN, MAX, COUNT_DISTINCT
Non-COUNT() aggregates return List<AggregateResult>
Read via (Decimal)ar.get('expr0') or an alias
GROUP BY (API 18.0+) ⟶ distinct values, e.g. GROUP BY Designation__c
Find dupes: GROUP BY Name HAVING COUNT(Id) > 1
HAVING filters groups; WHERE filters rows before grouping
GROUP BY ROLLUP / CUBE ⟶ subtotals; max 2,000 groups returned

🔒FOR UPDATE & Locks

[SELECT ... FROM Account FOR UPDATE] ⟶ row lock until transaction ends
Prevents lost updates in read-modify-write race conditions
Cannot combine FOR UPDATE with ORDER BY (implicit order by Id)
Not allowed with aggregate/GROUP BY queries, or in SOSL
Waiting txn errors after ~10s: UNABLE_TO_LOCK_ROW
Lock parent before children, in consistent order ⟶ avoids deadlock
MD child insert/update locks the master record implicitly

🚦Limits + External

SOQL: 100 queries sync / 200 async; 50,000 rows per transaction
SOSL: 20 queries per transaction; 2,000 records returned
Batch Apex QueryLocator: 50 million rows; SOQL for-loop chunks of 200
External ID: Text, Number, Email, Auto Number; max 25 per object
External ID = indexed + upsert match key + REST /Account/Ext__c/12345
Unique blocks duplicates only; External ID is NOT automatically unique
External objects (Salesforce Connect) __x ⟶ data stays in remote system
Use External Lookup (parent = external) / Indirect Lookup (match on Ext ID)
External objects: no roll-ups, no triggers, no SOSL, per-query row caps
WhoId = Contact/Lead (Name); WhatId = Account/Opp/Case (Related To)

⇄ Quick Comparison ⇄

SOQL vs SOSL

SOQL	Feature	SOSL
SELECT	Keyword	FIND
One object + related	Scope	Many unrelated objects
query() call	API call	search() call
List / sObject / Integer	Returns	List<List<sObject>>
50,000 rows	Row limit	2,000 rows
100 sync / 200 async	Query limit	20 per transaction
Yes, DML on results	DML	No DML on results
Structured field filters	Best for	Text: name, email, phone
WHERE + indexes	Filtering	Search index, wildcards * ?

Master-Detail vs Lookup

Master-Detail	Feature	Lookup
2 per object	Governor limit	40 per object
Required on detail	Field required	Optional by default
Cascade deletes children	Parent deleted	Clears field or blocks delete
Inherits master's sharing	Security	Independent owner + sharing
Supported	Roll-up summary	Not supported
Only if reparenting allowed	Reparent	Always allowed
Master only, never detail	Standard object	Either side
2 MD fields = junction	Many-to-many	Not possible

Easy way to remember

SOQL SELECTs one tree, SOSL FINDs many trees
MD = Mandatory, Deleted, Inherited, Rolled-up
2 MD, 40 lookups, 50k rows, 2k SOSL

Interview tips

Q. Why does my query time out on 5M rows?

A. Non-selective filter ⟶ full table scan; index the field, check Query Plan.

Q. Can you roll up over a lookup?

A. No. Master-detail only, or use Apex trigger / Declarative Lookup Rollup Summaries.

Q. How do you avoid record lock contention?

A. FOR UPDATE with consistent lock ordering; keep transactions short; batch by parent.

Takeaway

Relationships, not joins - master-detail buys you roll-ups and cascade delete at the cost of the child's independence.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

7/16

Salesforce Interview Notes · Lead / Architect

★LWC CHEAT SHEET★

Lifecycle, decorators, wire vs imperative, events, LDS

🔄Lifecycle Hooks

Order: constructor ⟶ connectedCallback ⟶ render ⟶ renderedCallback
Parent constructor first; child renderedCallback fires BEFORE parent's
constructor(): first statement must be super(), no params
constructor(): no this.template (not in DOM), no document.write/open
connectedCallback(): fires EVERY insert into DOM ⟶ subscribe, init
renderedCallback(): LWC-only, fires many times ⟶ guard with hasRendered
disconnectedCallback(): cleanup, unsubscribe, releaseMessageContext
errorCallback(error, stack): error boundary, catches DESCENDANT errors
Mutating reactive props in renderedCallback ⟶ infinite render loop

🏷️Decorators

import { LightningElement, api, track, wire } from 'lwc'
@api = public + reactive; parent sets it; exposed to App Builder
@api props read-only in child ⟶ never mutate, dispatch event instead
@track: all fields reactive since Spring '20; only for deep object/array mutation
@wire(adapter, {param: '$prop'}) ⟶ leading $ = reactive, re-fetches
Wired property: @wire(getContactList) contact; ⟶ {data, error}
Wired function receives ({error, data}) on every provision
Getters compute derived values the template can reference
Scoped: @salesforce/schema, /apex, /label/c.X, /resourceUrl, /user/Id
@salesforce/client/formFactor ⟶ Large | Medium | Small

⚡Apex Rules

Method must be static + public/global + @AuraEnabled
@AuraEnabled(cacheable=true) ⟶ client cache; NO DML allowed
import m from '@salesforce/apex/ClassName.methodName'
Imperative params = ONE object, keys match Apex param names
Imperative returns a Promise ⟶ .then().catch()
refreshApex() from '@salesforce/apex' refreshes wired cached data
Imperative for: DML, multi-record CUD, on-click timing control

📡Component Comms

Parent ⟶ child: public @api properties in the markup
Parent ⟶ child method: @api method + this.template.querySelector().method()
Child ⟶ parent: dispatchEvent(new CustomEvent('nextpage', {detail:{}}))
Event name lowercase, no spaces; parent listens on<eventname>
detail is ALWAYS an object; read it via event.detail
bubbles default false, composed default false, cancelable default false
composed: true required to cross the shadow boundary
No relation: pubsub singleton - fireEvent/registerListener/unregisterListener
pubsub = Aura application event; limited to ONE Lightning page
LMS spans pages, tabs, pop-outs, utility bar, Aura + Visualforce

📨LMS Wiring

Metadata type: Lightning Message Channel (.messageChannel-meta.xml)
lightning/messageService ⟶ publish, subscribe, unsubscribe, MessageContext
import CH from '@salesforce/messageChannel/MyChannel__c'
APPLICATION_SCOPE = receive even when not on the active tab
Aura: <lightning:messageChannel>; Visualforce: $MessageChannel global
createMessageContext() ⟶ must releaseMessageContext() in disconnectedCallback

🌓Shadow DOM

this.template.querySelector() searches own shadow tree ONLY
Cannot pierce into a child component's template
Query only in renderedCallback or later - DOM must exist
querySelector returns FIRST match; prefer data-* or lwc:ref over classes
CSS does not leak in or out across the shadow boundary
Aura CAN contain LWC; LWC CANNOT contain Aura
VF ⟶ LWC chain = VF > Aura > LWC, Aura app extends ltng:outApp
Same namespace: LWC and Aura cannot share a component name

🗄️LDS + Navigation

Stack: Browser > LWC > LDS > Server > UI API > Database
LDS handles FLS, sharing, record cache ⟶ prefer over Apex when it fits
lightning-record-form: view/edit, component decides field layout
record-view-form / record-edit-form: you choose + position fields
Read: getRecord / getRecords; metadata: getObjectInfo / getObjectInfos
Single DML: createRecord, updateRecord, deleteRecord - separate transactions
Multiple records CUD ⟶ imperative Apex only
Spanning (relationship) fields: max 5 levels deep
NavigationMixin(LightningElement) + this[NavigationMixin.Navigate]({...})
To VF: type 'standard__webPage', url '/apex/MyVfPage?id='+recordId

🚨Errors + Config

Normalise with reduceErrors from c/ldsUtils (lwc-recipes)
Wired property ⟶ getter: get errors(){ reduceErrors(this.x.error) }
Wired function ⟶ handle the error member each provision
Imperative ⟶ .catch(error ⟹ this.errors = reduceErrors(error))
.js-meta.xml: isExposed, apiVersion, targets, targetConfigs, masterLabel
Every <property> in targetConfig needs a matching @api field
Targets: lightning__RecordPage|AppPage|HomePage|Tab|FlowScreen|UtilityBar
Flow output: type="@salesforce/schema/Contact[]" role="outputOnly"
fetch() allowed, but host must be a CSP Trusted Site
Jest (sfdx-lwc-jest) in __tests__/; does NOT count to Apex coverage

⇄ Quick Comparison ⇄

LWC vs Aura

Aura	Feature	LWC
Proprietary Aura framework, 2014	Base	Web standards: custom elements, Shadow DOM, ES6
Up to 8 files (cmp, controller, helper, renderer)	Bundle	html + js + js-meta.xml (+ css, svg, tests)
<lightning:input>, <c:myComp> colon	Markup	<lightning-input>, <c-my-comp> hyphen
<aura:attribute>, {!v.x} / {!c.doIt}	Binding	Direct {propertyName}, no v.
aura:if / aura:iteration	Templating	template if:true / for:each + key
Component + application events, aura:method	Events	Standard DOM CustomEvent + @api
component.get("c.m"), setParams, $A.enqueueAction	Apex	import @salesforce/apex, @wire or Promise
init, render, rerender, afterRender, unrender	Lifecycle	constructor, connected, rendered, disconnected, errorCallback
Developer Console works	Tooling	VS Code + Salesforce DX only
Can contain an LWC	Nesting	Cannot contain Aura
Locker, LDS, base components	Shared	Locker, LDS, base components

Wire vs Imperative

@wire	Feature	Imperative
Declarative, on load + param change	Trigger	You control it, e.g. button click
Requires cacheable=true	Annotation	@AuraEnabled, cacheable optional
Not allowed - cacheable forbids DML	DML	Allowed
Cached; refreshApex() to refresh	Caching	No cache, just call again
{data, error} provisioned	Result	Promise .then / .catch
'$prop' re-runs automatically	Reactivity	Manual re-invoke

Easy way to remember

Con-Con-Ren-Dis-Err = constructor, connected, rendered, disconnected, errorCallback
Constructed top-down, rendered bottom-up: child renderedCallback beats parent's
bubbles + composed are both FALSE by default

Interview tips

Q. Why does renderedCallback loop forever?

A. It mutates a reactive property, triggering another render. Guard with a boolean flag.

Q. Wire or imperative for a Save button?

A. Imperative - wire needs cacheable=true, and cacheable Apex cannot do DML.

Q. pubsub vs LMS?

A. pubsub is one Lightning page only; LMS spans pages, tabs, utility bar, Aura and VF.

Takeaway

LWC is the browser doing the work - standards, shadow boundaries, Promises; @wire reads, imperative writes.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

8/16

Salesforce Interview Notes · Lead / Architect

★AURA / LIGHTNING★

Bundles, attributes, events, interfaces, LDS, App Builder

📦Bundle Files

8 resources; only .cmp markup is mandatory
Component (.cmp) = markup; Controller = client-side event handling
Helper = shared/reusable logic, called from controller + renderer
Renderer = override render/rerender/afterRender/unrender
Style = CSS; Documentation = usage docs
SVG = icon shown in Lightning App Builder palette
Design (.design) = which attrs admins can edit in App Builder
App bundle (.app) only for standalone app or Lightning Out dependency

🏷️Attributes + Expr

<aura:attribute name type default description access/>
ONLY name + type are required
Types: String Integer Boolean Date Datetime Double Decimal Long
Also Array List Set Map, sObject, MyObj__c[], Apex class, Object
v = value provider for attributes ⟶ {!v.attr}; c = controller ⟶ {!c.handleClick}
JS: component.get("v.x") / component.set("v.x", val)
Bound {!v.x} = two-way parent<⟶child sync
Unbound {#v.x} = one-time pass, cheaper, better perf
$Label: $A.get("$Label.ns.name") or {!$Label.c.name}
Design attrs = attributes exposed to App Builder / Community Builder

🧠Controller vs Helper

Controller: handles markup events ({!c.method}), keep it THIN
Helper: common logic, no repetition; helper.method(component, event, helper)
Controller method CANNOT call another controller method
Helper CAN call another helper via this.methodName()
Helper callable from controller, renderer, other helpers
Helper inheritable: <aura:component extends="c:SharingComponent"/>
Reusability lives in the HELPER, not the controller

⚡Events

3 types: Component, Application, System (init, render, aura:doneRendering)
4 steps always: create ⟶ register ⟶ fire ⟶ handle
Component: <aura:event type="COMPONENT"> + <aura:registerEvent>
Fire: component.getEvent("name").setParams({..}).fire()
App: <aura:event type="APPLICATION">; fire via $A.get("e.c:MyEvt")
Read payload in handler: event.getParam("attrName")
Capture runs FIRST (top⟶bottom), then bubble (bottom⟶top)
Default phase = application events only, runs from root node
event.stopPropagation() halts travel; preventDefault() kills default phase
Prefer component event: localised, more secure, easier to debug
aura:method = parent calls child directly, no event needed
Args in aura:method: event.getParam('arguments')

🔌Interfaces

force:appHostable ⟶ use as a custom tab (also mobile nav menu)
flexipage:availableForAllPageTypes ⟶ all Lightning pages
flexipage:availableForRecordHome ⟶ record pages only
force:hasRecordId ⟶ injects recordId attr (record page + App Builder only)
force:hasSObjectName ⟶ injects sObjectName attr
lightning:actionOverride ⟶ override standard New/Edit/View
force:lightningQuickAction ⟶ quick action w/ header+Cancel
force:lightningQuickActionWithoutHeader ⟶ no header/footer
forceCommunity:availableForAllPageTypes ⟶ Experience Builder (needs access="global")
lightning:isUrlAddressable ⟶ /lightning/cmp/c__MyCmp, state ⟶ v.pageReference

🛰️Server Calls

Apex method must be @AuraEnabled + public/global + static
1) var action = component.get("c.method")
2) action.setParams({key:val}) -- keys MUST match Apex param names
3) action.setCallback(this, fn) 4) $A.enqueueAction(action)
enqueueAction queues, does not run now ⟶ batched = "boxcarring"
response.getState() ⟶ SUCCESS | ERROR | INCOMPLETE | NEW
response.getReturnValue() = deserialized Apex return
void @AuraEnabled = useless; callback gets null ⟶ return String/Boolean
@AuraEnabled(cacheable=true) = client cache, read-only method
Storable action = server response cached on client

🗄️LDS + Data

LDS = CRUD a record with NO Apex; handles FLS + sharing
Aura: force:recordData; LWC: lightning/ui*Api wires + lightning-record-form
Shared cached copy ⟶ all components on page see same data, update together
De-duplicates identical record requests ⟶ fewer server calls
Current user w/o Apex: recordId={!$SObjectType.CurrentUser.Id}, fields="Profile.Name,Name"
FLS-safe: lightning:recordForm / recordEditForm / recordViewForm
Apex runs SYSTEM mode ⟶ WITH SECURITY_ENFORCED / WITH USER_MODE / stripInaccessible
Cross-component refresh: <aura:handler event="force:refreshView" action="{!c.doInit}"/>

🧭App Builder + Nav

Lightning App Builder page types: App Page, Home Page, Record Page
Assignment levels: Org default ⟶ App default ⟶ App+RecordType+Profile
lightning:navigation + PageReference {type, attributes, state}
Targets: standard__component / recordPage / objectPage / namedPage / webPage / knowledgeArticle / navItemPage / recordRelationshipPage
navigate(pageRef) vs generateUrl(pageRef, callback)
Legacy events: force:navigateToURL / ToSObject / ToList / ToObjectHome
force:showToast types: info, success, error, warning; mode: dismissible/pester/sticky
lightning:overlayLibrary ⟶ showCustomModal() for modals
Lightning Out: <apex:includeLightning/> + app extends ltng:outApp + $Lightning.use/createComponent
ltng:outApp applies SLDS; ltng:outAppUnstyled does not

🔁Lifecycle + Perf

init fires first: <aura:handler name="init" value="{!this}" action="{!c.doInit}"/>
doInit runs AFTER init, BEFORE render ⟶ like a constructor
Order: init ⟶ render() ⟶ afterRender(); data change ⟶ rerender(); destroy ⟶ unrender()
Overriding in an extending cmp? call superRender/superAfterRender/superRerender/superUnrender
Change handler: <aura:handler name="change" value="{!v.attr}" action="{!c.h}"/>
Perf: use LDS, unbound {#}, cacheable=true, fewer handlers, SELECT fewer cols + LIMIT
Combine actions into one composite request; no debug mode in prod
ui: tags = old; lightning: tags = SLDS-styled, preferred
Standalone app needs extends="force:slds" else raw unstyled HTML
No limit on how many components you can nest

⇄ Quick Comparison ⇄

Component vs Application Event

Component Event	Aspect	Application Event
type="COMPONENT"	Declaration	type="APPLICATION"
Self or ancestor in containment	Who handles	ANY component in the app
component.getEvent("name")	Fire syntax	$A.get("e.c:MyEvent")
Bubble + Capture	Phases	Bubble + Capture + Default
Child ⟶ parent data up	Use case	App-wide, e.g. navigate to record
More secure, easy to debug	Trade-off	Publish-subscribe, hard to trace
Default choice	Preference	Only when component event won't do

Aura vs LWC vs Visualforce

Aura	Aspect	LWC / VF
Aura cmp can contain LWC	Nesting	LWC cannot contain Aura
force:hasRecordId	Record Id	LWC: @api recordId
force:recordData	LDS entry	LWC: lightning/uiRecordApi wires
CSS cascades from parent	Styling	LWC shadow DOM: SLDS hooks / shared CSS
2015, component, client-side	vs Visualforce	VF 2008, page-centric, server round trips

Easy way to remember

Bundle = C-C-H-S-D-R-S-D (Cmp Controller Helper Style Doc Renderer SVG Design)
Events: Create-Register-Fire-Handle
Capture falls DOWN first, Bubble rises UP

Interview tips

Q. Component vs application event?

A. Component = child to ancestor only; application = any cmp, publish-subscribe, 3 phases.

Q. Why a helper, not the controller?

A. Controller methods can't call each other; helpers can via this.method() ⟶ reuse.

Q. Why Lightning Data Service?

A. CRUD without Apex, enforces FLS + sharing, shared cache, de-dupes record requests.

Takeaway

Thin controller, fat helper, component events over application events, and LDS before Apex.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

9/16

Salesforce Interview Notes · Lead / Architect

★VISUALFORCE & CLASSIC★

Controllers, view state, lifecycle, AJAX, migration

🎛️Controller Types

Standard: auto-generated per object ⟶ save edit delete cancel list
Custom: Apex class, replaces standard logic; NO-ARG constructor only
Extension: adds to standard/custom without replacing it
Standard list/set: needs recordSetVar ⟶ list views, filter, pagination
<apex:page standardController="Contact" extensions="E1, E2">
standardController + controller on same page = ILLEGAL
1 controller max + unlimited extensions
Extension constructor takes ApexPages.StandardController con
Duplicate method names ⟶ resolved left⟶right, later extension wins
~100 built-in components; custom via <apex:component>

💾View State

Holds controller instance variables between requests (postbacks)
Stored in a HIDDEN FORM FIELD ⟶ only exists inside <apex:form>
MAX 170 KB ⟶ "Maximum view state size limit exceeded"
transient keyword = excluded from view state (also excluded: static)
Fixes: transient, clear unused collections, ONE <apex:form> per page
Re-query large lists instead of parking them in view state
Inspect: Dev Mode + "Show View State in Development Mode" ⟶ View State tab
transient props do NOT round-trip ⟶ set them in the postback method

🔄Lifecycle Order

GET (first load): constructor ⟶ action="{!init}" ⟶ getters ⟶ render HTML
Constructor of standard ctrl runs BEFORE extension constructors
POSTBACK: decode view state ⟶ setters ⟶ action method ⟶ getters ⟶ re-render
Setters ALWAYS execute before any action method
action attribute on <apex:page> fires on page load only, not on rerender
Every displayed value needs a getter; {!x} maps to getX()
3 method kinds: getter, setter, action (+ navigation via PageReference)
3 binding kinds: data {!acct.Name}, action {!save}, component {!$Component.id}

⚡AJAX Actions

reRender="id" = redraw only those components, no full refresh
<apex:actionRegion> = only components INSIDE are sent + validated
actionRegion skips validation of required fields outside it
<apex:actionFunction> = JS-callable function ⟶ Apex action
<apex:actionSupport event="onclick"> = one specific component only
<apex:actionPoller interval="15"> = repeat call, seconds, min 5s
<apex:actionStatus> = spinner; facets name="start" / name="stop"
actionFunction cannot be nested inside an iteration component

📄Pagination

ApexPages.StandardSetController; DEFAULT page size 20
setCon.setPageSize(10) inside the extension constructor
next() previous() first() last(); getHasNext() / getHasPrevious()
getRecords() returns the current page; getResultSize() total
Max 10,000 records in a StandardSetController set
Extension ctor: public MyExt(ApexPages.StandardSetController c)
Override a LIST button ⟶ page MUST declare recordSetVar
setPageNumber(n) to jump; getCompleteResult() false if >10k

🏷️Key Tags

<apex:page> obligatory outer tag, only ONE per page
showHeader="false" sidebar="false" to strip Classic chrome
pageBlockTable = standard SF styling; dataTable = fully custom format
Inputs: inputField inputText inputSecret(password) inputTextarea inputFile inputHidden
outputLink = plain <a>; commandLink = <a> that fires an action
pageMessage severity: confirm|info|warning|error; strength 0-3
selectList/selectRadio/selectCheckboxes fed by selectOption(s)
<apex:inlineEditSupport>, <apex:include pageName>, <flow:interview name>
Component doing DML needs <apex:component allowDML="true">
Current record: ApexPages.currentPage().getParameters().get('id')

🔢VF Limits

View state 170 KB | PDF renderAs max 15 MB
Static resource 5 MB each, 250 MB per org
Page/HTTP response size max 15 MB
Max 10 field dependencies per VF page
Iteration components: 1,000 items (10,000 if readOnly page)
readOnly mode: 1,000,000 SOQL rows instead of 50,000
panelBar up to 1,000 items; JS remoting timeout 30,000 ms default
Remoting config: buffer=true, escape=true, timeout=30000
Excel: contentType="application/vnd.ms-excel#Contacts.xls"

🚀To Lightning

Tick "Available for Lightning Experience / mobile" on the page
window.location + sforce.one differ ⟶ use lightning/navigation or sforce.one
Style with <apex:slds/> instead of pageBlock Classic markup
Lightning Out: includeScript /lightning/lightning.out.js + $Lightning.use
Out app must be <aura:application extends="ltng:outApp"> + aura:dependency
VF inside LEX runs in an IFRAME (visual.force.com) ⟶ cross-domain limits
Migrate: VF page ⟶ LWC; controller ⟶ @AuraEnabled(cacheable=true)
No view state in LWC ⟶ state lives client-side, so re-think chunky controllers
S-Controls (client-side JS widgets) were superseded by Visualforce

⇄ Quick Comparison ⇄

actionFunction vs Remoting

apex:actionFunction	Aspect	JS Remoting @RemoteAction
Full postback, view state sent	Payload	Lightweight JSON, NO view state
Must sit inside <apex:form>	Form	No form required
Synchronous-feeling, slower	Speed	Faster, truly async
Instance methods	Method type	static + @RemoteAction required
reRender rebuilds markup	UI update	Callback JS updates the DOM
Controller state preserved	State	Stateless, no view state access

Controller Choice

Standard	Need	Custom / Extension
Auto CRUD, sharing enforced	Behaviour	Custom logic; extension keeps standard actions
No Apex, no tests needed	Effort	Apex class + 75% coverage
Runs in user mode	Security	Custom runs SYSTEM mode unless "with sharing"
recordSetVar for lists	Lists	StandardSetController in extension

Easy way to remember

View State = 170 KB in a hidden field, so it needs a FORM
Load: constructor ⟶ action ⟶ getters; Postback: setters ⟶ action ⟶ getters
setRedirect(true) = new request, state LOST; false = same request, state KEPT

Interview tips

Q. Why is my page throwing view state errors?

A. Instance vars over 170 KB - mark transient, drop collections, keep one apex:form.

Q. actionFunction or JS Remoting?

A. Remoting for speed/stateless JSON; actionFunction when you need view state and reRender.

Q. Standard controller vs extension?

A. Extension keeps standard save/edit/delete and layers custom logic; custom controller replaces it entirely.

Takeaway

Visualforce is stateful and server-rendered - the 170 KB view state and its postback lifecycle explain nearly every VF bug and every reason to migrate to LWC.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

10/16

Salesforce Interview Notes · Lead / Architect

★ADMIN, CONFIG & AUTOMATION★

Flows, declarative tools, layouts, reports, data tools

🌊Flow Types & Anatomy

Screen Flow ⟶ UI screens; Lightning page, quick action, button, Experience Cloud, URL
Record-Triggered ⟶ create / update / delete; before-save or after-save
Schedule-Triggered ⟶ start date+time; once, daily or weekly + entry conditions
Platform Event-Triggered ⟶ fires when an event message is received
Autolaunched (no trigger) ⟶ called from Apex, REST API, process, or as subflow
Built in Flow Builder (ex Visual Workflow / Cloud Flow Designer)
3 parts: Elements + Resources + Connectors (incl. fault connectors)
Subflow: only Screen + Autolaunched; screen flow may call either, autolaunched cannot hold screens
Apex Action = @InvocableMethod; Local Action = client-side Lightning component
Pass record: Text var named recordId, "Available for input"; or Record-type var for whole sObject
Fast Lookup / Fast Create = bulk sObject-collection elements
Distribute + grant "Run Flows" perm or flow access on profile/permission set

⚡Before vs After Save

BEFORE save ⟶ set fields on the TRIGGERING record, no extra DML, ~10x faster than Process Builder
BEFORE save runs before before-triggers; AFTER save runs after after-triggers, with processes
AFTER save ⟶ related records, email alerts, Chatter, custom notification, Apex, submit for approval
Entry condition "Only when a record is updated to meet the condition requirements" ⟶ anti-recursion
$Record vs $Record__Prior to detect a genuine field change
One flow per object per timing; set Trigger Order for predictable sequence
Scheduled Paths (after-save only) for time-delayed work ⟶ never loop or poll
Async path for callouts / non-transactional work
Flows share the transaction with triggers ⟶ same SOQL, DML, CPU limits

🛠Flow Design & Gotchas

200 records = ONE bulk interview batch; only elements OUTSIDE loops get batched
NEVER Get / Create / Update / Delete / Apex / email inside a Loop
Inside loop use Assignment only ⟶ build collection ⟶ single DML after the loop
Fault path on every Get, DML, Apex, callout, email, approval element
Screen fault ⟶ error screen with {!$Flow.FaultMessage}; background ⟶ log to custom object
Monitor Setup > Paused and Failed Flow Interviews
Get Records: only the fields you use, tight filter, "Only the first record" when one is enough
Loop cap ~2,000 iterations per interview; shares 100 SOQL / 150 DML / 10,000 ms CPU
Debug with a real record + "Debug in rollback mode"; bulk-test 200 rows via Data Loader
Every edit creates a new VERSION; only one active version; revert supported
Record Choice Set (queries at runtime) vs Collection Choice Set (reuse a Get) = more efficient
Screen "Validate Input": formula must be TRUE for VALID ⟶ inverse of validation rules
No true unit-test framework; XML merge pain in version control

🧭Tool Choice & Migration

Workflow: only 4 actions - field update, email alert, task, outbound message
WF field update ⟶ same record or MASTER-DETAIL parent only; never through a lookup
WF limits: 500 rules per object, 50 active; fires on insert + update only
Eval criteria: created / created+every edit / created+edited to subsequently meet
Time-dependent action NOT allowed with "created, and every time it's edited"
Pending time actions: Setup > Monitoring > Time-Based Workflow
All matching WF rules fire in one transaction, ORDER NOT GUARANTEED ⟶ updates overwrite
"Re-evaluate Workflow Rules After Field Change" ⟶ re-runs all rules AND re-fires triggers
Process Builder: multi criteria nodes, related records, Apex, approvals; no delete, no UI, after-save only
WF + Process Builder are RETIRED ⟶ all new automation in Flow; use Migrate to Flow tool
Choose Apex when: complex logic, dynamic SOQL, callouts, huge volume, real error handling
Trigger beats PB: before-context edits with no extra DML, delete events, bulk-safe

🧮Formulas, Roll-ups, Rules

Formula field: read-only, calculated on view, NOT stored in the database
Return types: Checkbox, Currency, Date, Date/Time, Number, Percent, Text
Cross-object formula: up to 10 relationships spanning 5 unique objects, child ⟶ parent
Cross-object formulas work on lookup AND master-detail
Roll-up summary: master-detail only, read-only, CANNOT aggregate a cross-object formula
Workaround: automation copies value into a plain Number field ⟶ roll THAT up
Roll-up on a lookup ⟶ Apex trigger, Flow, or DLRS / Rollup Helper from AppExchange
ISBLANK() any type + preferred; ISNULL() legacy, always false on text (text is never null)
ISPICKVAL(field,'value') to compare picklists; CASESAFEID(Id) turns 15 ⟶ 18 chars
IMAGE(url, alt) shows an image as a field on the detail page
Validation rule: formula TRUE = error + save blocked; 100 active per object
500 custom fields per object, extendable to 800 via a support case
3 ways to make a field required: field definition, page layout, validation rule
Custom labels: 5,000 per org, 1,000 chars each, $Label / System.Label, translatable

🧱Layouts & Record Types

Page layout = fields, buttons, custom links, related lists, sections; assigned by profile + record type
Record type = different picklist value sets + layout + business process per profile
Different picklist values per layout ⟶ use RECORD TYPES, not layouts
Dynamic Forms = fields/sections placed on the Lightning record page with component visibility rules
Mini page layout ⟶ hover detail + console tab fields (no related lists in hover)
Field dependency: controlling field must be a picklist or CHECKBOX; dependent is a picklist
Global action = create only, no parent link, sits on global publisher layout
Object-specific action = create + update + log a call + email, auto-related to the record
3 custom tab types: custom object, Visualforce, Web tab
App Launcher visibility = profile tab settings (Default On/Off/Hidden) + assigned apps
Record Name field data type: Auto Number or Text only
Skip record type selection page ⟶ personal setting > Record Type Selection
Email templates: Text, HTML (letterhead), Custom (no letterhead), Visualforce

✅Approvals & Assignment

Approval process = entry criteria + steps + approvers + submit/approve/reject/recall actions
Records can be locked while pending approval
Records still PENDING ⟶ process can only be deactivated, never deleted; approve/reject/recall first
Email Approval Response: Setup > Process Automation Settings; reply line 1 keyword, line 2 comment
Keywords: Approve, Approved, Yes // Reject, Rejected, No
"Dynamic approval" = approver resolved at runtime: Related User lookup or submitter chooses
Pattern: Flow stamps a custom User lookup pre-submission ⟶ step = Related User > that field
Queues: Lead, Case, all custom objects (+ Order, Service Contract, Knowledge Article Version)
Assignment rules: Leads and Cases ONLY - never custom objects
Many rules per object but only ONE ACTIVE; ordered entries, first match wins
From Apex set Database.DMLOptions.assignmentRuleHeader explicitly
Auto-response rules: Web-to-Case / Web-to-Lead, one active per object, first match wins
Escalation rules: age-based case reassign + notify
Lead convert ⟶ Account + Contact + optional Opportunity; map custom fields or data is LOST
Database.LeadConvert + setConvertedStatus + Database.convertLead; merge max 3 leads

📊Reports & Dashboards

4 formats: Tabular, Summary, Matrix, Joined
Tabular = fastest, no grouping; usable on a dashboard ONLY with a row limit set
Summary = row groupings + subtotals; grouping mandatory before adding a chart
Matrix = grouped by rows AND columns for two-dimension comparison
Joined = up to 5 blocks, each block its own report type
Report type = the template giving a report its structure; standard or custom
Bucket field exists only inside the report; supports Picklist, Number, Text
Conditional highlighting: max 3 colours / 2 breakpoints; needs a summary or matrix report
Dashboard: max 20 components, each backed by ONE source report; folders control access
Components: bar, line, pie, donut, funnel, scatter, Gauge, Metric, Table, Visualforce page
Standard dashboard shows the RUNNING USER's data; dynamic shows the logged-in viewer's
Report exports to Excel/CSV only ⟶ PDF via Printable View or dashboard "Save as PDF"
Analytic snapshot FAILS if the target object has a trigger or the running user is inactive

💾Data Ops Gotchas

Mass Delete Records: standard objects, 250 records at a time
Hard delete needs "Bulk API Hard Delete" perm + Bulk API option ⟶ bypasses Recycle Bin
Export excludes Recycle Bin; Export All includes it; SELECT * is not supported
Recycle Bin records AND deleted custom objects restorable for 15 days
Data Export (Setup > Data Export): CSV backup, weekly/monthly, min 48-hour interval
Relationships: load parents first; child column header Account:MyExtId__c links by External Id
Upsert needs an External Id or record Id; plain update matches on SF Id only
NO rollback in Data Loader ⟶ export a backup first; revert using success-file Ids
CSV only, UTF-8 encoding; consumes the org's API limits; Bulk API serial mode avoids row locks
Data loss causes: changing a field's data type, bad imports, integration errors, governor limits
Setup Audit Trail: last 20 changes on screen, 6 months downloadable as CSV

🗄Settings vs Metadata

Custom Setting types: List (static org-wide) vs Hierarchy (org > profile > user, most specific wins)
Cached in app cache ⟶ reads cost NO SOQL query; 10 MB org cap
Usable in formulas, validation rules, workflow, Apex, VF, flows, API ($Setup)
Records are DATA ⟶ NOT deployable; only the definition deploys
No page layouts, lookups, long text areas, validation rules or triggers on custom settings
Classic use: production kill-switch to disable a trigger
Custom Metadata (__mdt): records are METADATA ⟶ deploy via change sets / packages
Auto-present in every new or refreshed sandbox; read-only in Apex at runtime
Supports fields, page layouts, validation rules, metadata relationships, long text areas
SOQL-queryable but does NOT count against SOQL governor limits
getValues(id) ⟶ the raw single row exactly as stored ⟶ use when WRITING
getInstance(id) / getOrgDefaults() ⟶ merged up the hierarchy ⟶ use when READING
getInstance() no-arg = running user's merged view; always returns a row even if none stored

⇄ Quick Comparison ⇄

Flow Types At A Glance

Flow Type	Runs When	Use It For
Screen Flow	User launches it, interactive	Guided input, surveys, wizards
Record-Trig (Before Save)	Before record written to DB	Same-record field updates, no extra DML
Record-Trig (After Save)	After record saved, not committed	Related records, email, Chatter, Apex
Record-Trig (Delete)	Before the delete	Blocking / cleanup on delete
Schedule-Triggered	Set time; once / daily / weekly	Batch cleanup, nightly recalcs
Platform Event-Trig	On event message received	Integration, decoupled async work
Autolaunched (no trigger)	Called by Apex, API, process, parent flow	Reusable background logic, subflows

Import Wizard vs Data Loader

Data Import Wizard	Feature	Data Loader
Under 50,000 records	Volume	50k up to 5 million
Browser-based, inside Setup	Install	Client app, needs Java/JRE + API Enabled
Lead, Contact, Account, Campaign Member, Solution + all custom	Objects	All standard + custom (no User in Wizard)
Import only	Direction	Import + Export + Export All
Cannot save mappings	Mapping	Saves reusable .sdl mapping file
Insert, update, upsert	Operations	+ delete and hard delete
Name, email or SF Id	Match on	SF Id, or External Id for upsert
No	CLI / scheduling	Yes - process-conf.xml + cron/Task Scheduler

Easy way to remember

Before-save = same record, no DML. After-save = everything else
Workflow 4 actions, PB no delete, Flow does all ⟶ both retired
getValues to WRITE, getInstance to READ

Interview tips

Q. When do you use a before-save record-triggered flow?

A. Same-record field updates only - no extra DML, roughly 10x faster than Process Builder.

Q. Roll-up summary on a lookup relationship?

A. Not supported - use Apex, Flow, or DLRS/Rollup Helper; master-detail is required for native roll-ups.

Q. Custom Setting or Custom Metadata Type?

A. CMDT when the config must deploy between orgs; Custom Setting when it varies per profile/user.

Takeaway

Build everything in Flow - before-save for the record itself, after-save for the rest - keep queries and DML outside loops, and put fault paths on every element.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

11/16

Salesforce Interview Notes · Lead / Architect

★TESTING, DEPLOY & DEVOPS★

Coverage, mocks, sandboxes, change sets, CI/CD

🧪Test Class Rules

@isTest on class = test class; @isTest on method = test method
testMethod keyword = legacy; use @isTest per method
Test code does NOT count vs org Apex code size limit
Asserts both positive AND negative: System.assertEquals(expected, actual)
NEVER hardcode Ids in test or Apex classes
Bulk test with up to 200 records + real-world scenarios
@TestVisible ⟶ test can reach private members
System.runAs(user) ⟶ profile / permission / sharing context
Test.isRunningTest() ⟶ bypass code that must not run in test
Test.loadData() ⟶ build records from static resource CSV
Test.setCurrentPage(pageRef) ⟶ else ApexPages.currentPage() is null

📊Coverage Numbers

75% ORG-WIDE Apex coverage required to deploy to production
Every TRIGGER must have SOME coverage (> 0%)
"1% per trigger" is a myth; no real per-trigger percentage
NO per-class minimum ⟶ a class can sit at 0%
All tests must PASS, not just cover
Braces, comments, System.debug lines = white, not counted
Apex cannot be edited in production; Visualforce CAN be
Apex Hammer = SF runs your tests on current + next release, diffs results

⏱️startTest / stopTest

Test.startTest() ⟶ fresh set of governor limits for code under test
Test.stopTest() ⟶ forces queued async work to run synchronously
ONE startTest/stopTest pair per test method
Tests future, Queueable, Batch, Scheduled ⟶ assert AFTER stopTest()
Batch: execute() runs ONCE in test ⟶ insert <= 200 records (batch size)
Create batch data BEFORE Database.executeBatch(), matching start() query
start() query returns nothing ⟶ execute() never runs ⟶ 0% coverage
Chained batches: separate test method per batch class

🗄️@TestSetup + Data

@testSetup method creates data once, reused by all test methods
ONLY ONE setup method per test class
Rollback to setup state between test methods
Custom METADATA records visible in tests; custom SETTINGS must be created
@isTest(SeeAllData=true) ⟶ sees real org data; AVOID it
Default = tests see NO org data (isolation)
SeeAllData only for exceptional objects (e.g. old Pricebook)
Set read-only CreatedDate via JSON.deserialize(json, Case.class) - in memory only
Cannot commit an arbitrary CreatedDate to the DB

🔌Mocking Callouts

Real callouts in tests THROW ⟶ must mock
implements HttpCalloutMock ⟶ HTTPResponse respond(HTTPRequest req)
Register BEFORE code under test: Test.setMock(HttpCalloutMock.class, new MyMock())
SOAP stubs ⟶ WebServiceMock + Test.setMock(WebServiceMock.class, ...)
StaticResourceCalloutMock / MultiStaticResourceCalloutMock = body in static resource
Many endpoints ⟶ one mock, if/else branch on req endpoint/method/body
MultiStaticResourceCalloutMock maps endpoint ⟶ static resource
Callouts block the thread ⟶ use @future(callout=true) or Queueable
Cover catch block: runAs a user without access, or violate validation/required field

🏖️Sandboxes

4 types: Developer, Developer Pro, Partial Copy, Full
Copy of production metadata; data depends on type
Sandbox NAME max 10 characters
Org ID CHANGES on every sandbox refresh
Dev/Dev Pro = metadata only, no data
Partial Copy = sample data, up to 10,000 records per object
Full = complete copy incl. all data ⟶ use for UAT / perf / training

🚀Deployment Paths

Change Sets: outbound (source) + inbound (target), needs deployment connection
Related orgs only; METADATA only - records never move (use Data Loader)
Steps: add components + dependencies ⟶ upload ⟶ validate ⟶ deploy
Validate with named test classes ⟶ Quick Deploy inside validation window
Ant / Force.com Migration Tool: Java+ANT, Metadata API, can DELETE metadata
destructiveChanges.xml = deletions via Metadata API
SFDX/CLI: sf project deploy start, source-tracked, scratch orgs, CI-friendly
Packages: managed (locked, upgradable, namespaced) vs unmanaged (editable, no upgrade)
Unlocked / 2GP packages = modern modular org deployment
Data Loader CANNOT move metadata; Communities need Network + CustomSite + ExperienceBundle

🔁CI/CD + Git

Salesforce stores no versions ⟶ Git repo lives OUTSIDE the org
3 dev models: change set, org development, package development
Branch per user story ⟶ dev sandbox/scratch org ⟶ integration ⟶ UAT ⟶ prod
Commands: init, config, clone, checkout, status, add, commit, pull, push, merge
Tools: Azure DevOps, Copado, Gearset, Jenkins, AutoRABIT, Flosum
Copado = SF-native DevOps; PMD / SonarQube / CodeScan static analysis
Copado nouns: Credential, Environment, User Story, Commit, Promotion, Release
Back promotion = push prod/UAT changes DOWN to lower orgs after hotfix
Flows: never edit active version ⟶ Save As New Version ⟶ activate; rollback = reactivate
Flow XML merges badly ⟶ one owner per flow per sprint

⇄ Quick Comparison ⇄

Sandbox Types & Refresh

Sandbox	Data + Storage	Refresh
Developer	Metadata only, 200 MB	Daily (1 day)
Developer Pro	Metadata only, 1 GB	Daily (1 day)
Partial Copy	Sample data, 5 GB, 10k rec/object	Every 5 days
Full	Full copy of production data	Every 29 days

Change Sets vs SFDX

Change Sets	Feature	SFDX / CLI + Ant
Point-and-click in org	Interface	CLI + XML manifest
Related orgs only	Scope	Any org, any repo
No	Version control	Yes, Git-native
No (manual re-add)	Deletions	destructiveChanges.xml
Manual, not repeatable	CI/CD	Fully automatable

Easy way to remember

75 / 200 / 1
Seventy-five percent org, 200 records per batch test, ONE @testSetup
Dev daily, Partial 5, Full 29

Interview tips

Q. Minimum coverage to deploy?

A. 75% org-wide, all tests pass, every trigger has some coverage; no per-class minimum.

Q. Why Test.startTest()?

A. Fresh governor limits for code under test + forces async work to finish at stopTest().

Q. How do you test a callout?

A. Implement HttpCalloutMock.respond(), register with Test.setMock() before invoking the code.

Takeaway

75% is a deployment gate, not a quality bar - assert behaviour, mock callouts, and let Git, not the org, hold your versions.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

12/16

Salesforce Interview Notes · Lead / Architect

★SCENARIO PLAYBOOK★

Situational design answers for lead-level interviews

📦Batch Volume

150 batch jobs at once? ⟶ No. Flex queue caps at 100
Beyond 100 ⟶ Database.executeBatch throws LimitException, job not queued
Flex queue jobs sit in Holding status
Flex queue NOT enabled ⟶ status Queued, only 5 concurrent ⟶ max 5 in one go
Ordering across batches? ⟶ can't control order
Fix ⟶ Database.Stateful + class var flag; skip B until A processed
Then re-run batch to pick up skipped records
Error on 395th of 1000, size 200 ⟶ batch 1 (1-200) already committed
Batch 2 (201-400) rolls back whole chunk; earlier chunks NOT rolled back

⚡Trigger Patterns

Need parent Id to create child ⟶ after insert / after update only
before insert ⟶ no Id yet, relationship fails
Build List in loop, single DML outside loop ⟶ works for 1 or 200
One trigger per object; logic in handler class
Account created ⟶ auto Opportunity: after insert + handler.createOpportunities()
Closed Won ⟶ Contract: compare Trigger.oldMap to catch the transition only
Guard: skip if opp.AccountId == null (Contract needs AccountId + Status)
Without oldMap check ⟶ new Contract on every later edit of won opp

🔄Round Robin

Create queue + add users first, then after insert trigger
Read members: GroupMember WHERE GroupId IN (Group WHERE Type='Queue')
Filter Users on IsActive = true
Simple version ⟶ index = Math.mod(ticketNumber, agentSize)
OwnerId set after insert ⟶ explicit update ticketList
Remember pointer across transactions ⟶ hierarchy custom setting
getOrgDefaults() ⟶ User_Index__c = last assigned index
Advance index per record, write setting back once ⟶ 1 DML regardless of volume

🔧Config Gotchas

Object missing in report builder ⟶ Allow Reports unchecked on custom object
No List custom setting option ⟶ Setup > Schema Settings > Manage List Custom Settings Type
User can't create campaigns despite profile ⟶ Marketing User checkbox on user record
@wire returns nothing ⟶ Apex needs @AuraEnabled(cacheable=true), static, public/global
Change Field Type button hidden on M-D ⟶ roll-up summary exists on parent
Also: deleted roll-up still in Recycle Bin ⟶ must erase permanently
Other blockers ⟶ too many M-D on child, detail records needing optional lookup

🗂️Data Model

a2 parent of a1, then a1 parent of a2 ⟶ blocked, circular reference error
A record can never be its own parent in the hierarchy
M-D on object that already has records ⟶ not directly
Steps: create Lookup ⟶ populate parent on EVERY record ⟶ convert Lookup to M-D
Clone from UI, field has default but not on layout ⟶ new record gets FIELD default
Source record's value cannot carry forward if field is off the layout
Time-based WF queued, record edited out of criteria ⟶ action removed, never fires

🧭Declarative First

Days in current stage, no code ⟶ standard Last Stage Change Date + Stage Duration
Enable via Opportunity Update Reminders / Stage Change Fields
Alt 1 ⟶ field history tracking + Opportunity Field History report type
Alt 2 ⟶ Stage_Changed_Date__c set by Flow, formula TODAY() - Stage_Changed_Date__c
Surface with conditional highlighting or Path key fields ⟶ spot stale deals
Rule of thumb: config > Flow > Apex; declarative lets ops tune thresholds

🌊Flow STAR Story

Situation ⟶ manual renewal opps for entitlements expiring in 60 days
Task ⟶ automate chain, no Apex, business owns the rules
Action ⟶ schedule-triggered flow nightly + Get Records (Account, Contract)
Decision on contract value ⟶ named AM vs queue
Assignment builds collections ⟶ all DML OUTSIDE the loop
Fault paths ⟶ log to Integration_Log__c + email admin
Screen flow on Opportunity ⟶ rep confirms/declines, posts to Chatter
Result ⟶ ~10 hrs/week saved, no missed renewals, threshold editable by ops

🔌Callouts In Batch

Callouts from batch ⟶ must implement Database.AllowsCallouts
Up to 100 callouts each in start, execute, finish
Two calls per record ⟶ wrap both in one service class, call in sequence from execute
Keep scope small: Database.executeBatch(new TwoCalloutBatch(), 10)
Gate 2nd call on first.getStatusCode() == 200
Test ⟶ Test.setMock(HttpCalloutMock.class, ...) canned response per endpoint
2nd call depends on async completion ⟶ chain a Queueable from finish()

⇄ Quick Comparison ⇄

Batch Queue Limits

Flex Queue ON	Aspect	Flex Queue OFF
100 jobs	Max submitted at once	5 jobs
Holding	Status after executeBatch	Queued
LimitException at 101	Overflow behaviour	LimitException past 5
Yes, reorder in queue	Can change order	No

Trigger Timing Choice

Before Trigger	Need	After Trigger
No Id yet	Parent record Id	Id available
Set fields on same record	Best for	Create/relate child records
Not possible	Change OwnerId post-insert	Yes + explicit update
N/A	Detect field transition	Trigger.oldMap compare

Easy way to remember

Flex 100 Holding, no Flex 5 Queued
Child records need a parent Id ⟶ AFTER trigger, DML outside loop
Blocked button or missing option? Look for a checkbox: Allow Reports, Marketing User, cacheable=true

Interview tips

Q. 150 batches in one go?

A. No - Flex queue caps at 100; without Flex only 5 concurrent.

Q. Error on record 395 of 1000, size 200?

A. First chunk stays committed; only chunk 201-400 rolls back.

Q. Why can't I convert master-detail to lookup?

A. A roll-up summary exists on the parent - including one still in the Recycle Bin.

Takeaway

Scenario answers win on the constraint you name - the exact limit, the checkbox, or the after-trigger reason.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

13/16

Salesforce Interview Notes · Lead / Architect

★PLATFORM FUNDAMENTALS★

Multitenancy, clouds, objects, licences, releases, integration architecture

🏢Multitenancy

One shared infrastructure + ONE code base serves every customer org
Your org = your tenant; OrgId partitions every row of shared tables
Metadata-driven runtime: apps are metadata rows, not per-tenant schema
No servers, no version to install, no upgrade project ⟶ SaaS
Governor limits exist so one greedy tenant cannot starve the others
Limits are per-transaction and reset at each new transaction
Sync: 100 SOQL, 150 DML, 50,000 rows queried, 10,000 rows DML'd
CPU 10,000 ms sync vs 60,000 ms async; heap 6 MB vs 12 MB
100 callouts / txn, 120 s max callout time, 10 s per-callout default
Async (Batch/Queueable/@future) doubles SOQL to 200 + heap to 12 MB
Certified managed package gets its own limit pool; cumulative cap ~11x
Trade-off framing: "limits are the rent you pay for shared infrastructure"

☁️The Clouds

Sales Cloud - the selling process: Lead ⟶ Opportunity ⟶ Closed Won
Service Cloud - post-sale support: Cases, Knowledge, Console, Omni-Channel
Marketing Cloud - Email Studio, Journey Builder, Automation Studio, Mobile Studio
Account Engagement (Pardot) = B2B marketing automation, not Marketing Cloud
Commerce Cloud - B2C/B2B storefront: catalogue, cart, checkout, OMS
Experience Cloud (ex-Community Cloud) - portals, help centres, partner sites
CRM Analytics (ex-Einstein/Analytics Cloud) - datasets, lenses, dashboards
IoT Cloud - device event streams ⟶ orchestration rules ⟶ SF actions
Heroku - PaaS in Node/Java/Ruby/Python, wired back via Heroku Connect
Quip - docs, sheets and chat embedded on records
Industry clouds e.g. Automotive: Dealer Mgmt, digital retailing, connected car
Einstein = AI layer over CRM data; now Einstein GenAI + Agentforce

🔄Sales & Service Flow

Sales: Campaign ⟶ Lead ⟶ qualify ⟶ Convert ⟶ Account+Contact+Opportunity
Convert creates 3 records; Opportunity is optional at conversion
Lead = unqualified, holds person AND company on one unverified record
Opportunity = the money: Stage, Amount, Close Date, Forecast Category
Converted Lead detail page is NOT shown ⟶ links to Acct/Con/Opp instead
Sales objects: Lead, Account, Contact, Opportunity, Quote, Product, Price Book
Service capture: Web-to-Case, Email-to-Case, phone/CTI, Chat, community
Service routing: assignment rules ⟶ queues ⟶ Omni-Channel ⟶ escalation
SLAs via Entitlements + Milestones; auto-response rules, macros, Case Feed
Resolution aids: Knowledge articles, Case Teams, Case Comments

🆔Objects & IDs

5 system/audit fields on every object: Id, Name, CreatedBy, LastModifiedBy, Owner
Also CreatedDate + LastModifiedDate + SystemModstamp (delta-sync key)
Detail side of master-detail has NO OwnerId - inherits parent's owner
Id 15-char = case-SENSITIVE (UI/URL); 18-char = checksum, case-insensitive
Use 18-char in Excel/exports/integrations to avoid case-collision bugs
First 3 chars = key prefix naming the object: Account = 001, Contact = 003
Suffixes: __c custom object/field, __e event, __b big object, __mdt custom metadata
Field types: Text, Number, Picklist, Formula, Roll-Up, Lookup, Master-Detail
External Id field = upsert match key from an outside system
App = group of tabs working as a unit; tabs = the menu items of an app

🔗Relationships

Lookup: loose, optional, no cascade delete, child keeps own owner + sharing
Master-Detail: tight; child inherits owner + sharing, cascade delete, required
Roll-Up Summary fields only on the MASTER side of master-detail
Max 2 master-detail + up to 40 relationship fields per object
Many-to-many = junction object with 2 master-details
FIRST master-detail on a junction = primary: controls look/feel + detail page
Reparenting off by default on master-detail; can be enabled per field
Hierarchical relationship = User object only (e.g. Manager)
External lookup / indirect lookup join external objects via Salesforce Connect
Relationship names: child⟶parent Account.Name; parent⟶child Contacts (r for custom)

🏛️MVC & Anatomy

Model = objects, fields, relationships (data layer, held as metadata)
View = LWC, Aura, Visualforce, Lightning App Builder pages
Controller = Apex classes/triggers holding the business logic
Salesforce.com = SaaS CRM apps (Sales/Service) sold prepackaged
Force.com = the PaaS beneath it; SF's own apps are built on it
Now branded Salesforce Platform / Lightning Platform
User = unique username, email, Profile, optional Role, user licence
Licence sets what you CAN access; Profile + Perm Sets set what you DO
Chatter = real-time collaboration: feeds, groups, follows, @mentions, files
Mobile: Salesforce Mobile SDK, hybrid, or React Native

🎫Licences & Experience

5 Experience Cloud licence types for external users
External Apps (B2C), Customer Community (B2C), Customer Community Plus (B2C+B2B)
Partner Community (B2B), Channel Accounts (per-account, not per-user)
Customer Community: no roles, no sharing rules, high-volume portal user
Customer Community Plus + Partner: roles + sharing rules + reports
Partner Community = PRM site: leads, deal registration, shared content
Person Accounts: Customer Community + Customer Community Plus only
Business Accounts: Customer, Customer Plus and Partner Community
Build in Setup > Digital Experiences > All Sites > Experience Builder
Add partner users: Manage Users tab ⟶ invite by email ⟶ grant permissions
Manage Partners permission = view Consulting Partner status + certifications

🗓️Releases & AppExchange

THREE releases a year: Spring, Summer, Winter - auto-applied, no opt-out
Named for the upcoming year (Winter '26 ships in 2025)
Preview/Release-Update sandboxes get the new version early for regression testing
Release Updates page = breaking changes with an enforcement deadline
API versions are honoured, so old code keeps running post-upgrade
AppExchange = marketplace of apps, components, flows, consultants
Managed package: namespaced, upgradeable, code hidden, ISV security-reviewed
Unmanaged package: one-off template, editable, NOT upgradeable
Installed managed code is exempt from the org's Apex character limit
Install for admins only / all users / specific profiles

🔌Integration Architecture

Architect answers: who talks to whom, sync or async, what if SAP is down
APIs: REST (light/JSON), SOAP (WSDL contract), Bulk (millions, async)
Metadata API = config deploy; Tooling API = classes, coverage, logs
Composite API: many REST calls in ONE transaction via @{refAccount.id}
GraphQL API: request exactly the fields needed in one query (UI API)
Patterns: Request-Reply (user waits) vs Fire-and-Forget (no response)
Event-driven: Platform Events, CDC, Event Bus, Pub/Sub API (gRPC)
Middleware (MuleSoft/CPI/Boomi) turns point-to-point into one hub
API-led connectivity: System API ⟶ Process API ⟶ Experience API
Auth: OAuth 2.0 - JWT Bearer (server-to-server), Client Credentials, Web Server
Named + External Credentials store endpoint/secrets; mTLS = both sides verify
SAP down: queue message, retry 5/15/30 min, Dead Letter Queue, alerting
Delta records only, keyed on LastModifiedDate / SystemModstamp / CDC
Cut API usage: Composite, Bulk 2.0, events over polling, middleware cache

⇄ Quick Comparison ⇄

Editions Compared

Professional	Feature	Enterprise / Unlimited
50	Custom objects	EE 200 / UE 2,000
100	Custom fields per object	EE 500 / UE 800
Paid add-on	API access	EE 1,000 calls per licence/day
Managed packages only	Apex code	Full Apex, triggers, dev sandboxes
Limited automation	Record types + workflow	Full record types + Flow suite

Platform Events vs CDC

Platform Event	Criterion	Change Data Capture
Custom, business event	What it models	Record data change
Developer publishes (EventBus.publish)	Publishing	Automatic, zero Apex
Order Created, Warranty Registered	Examples	Account/Product/Price Updated
__e object, custom fields	Definition	Enable per object in Setup
Create only	Operations	Create, Update, Delete, Undelete

Easy way to remember

Clouds = Sales, Service, Marketing, Commerce, Experience
One code base + 3 releases a year ⟶ limits are the rent
001 = Account: first 3 chars of any Id name the object

Interview tips

Q. Why do governor limits exist?

A. Multitenancy - shared code and infrastructure, so no single tenant can starve the others.

Q. Salesforce.com vs Force.com?

A. Salesforce.com = SaaS CRM apps; Force.com = the PaaS platform they are built on.

Q. 15 vs 18 character record Id?

A. 15 is case-sensitive from the URL; 18 adds a checksum, safe in Excel and integrations.

Takeaway

Everything on this platform is metadata on shared infrastructure - clouds, licences, limits and releases all follow from that one fact.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

14/16

Salesforce Interview Notes · Lead / Architect

★HR + BEHAVIOURAL★

Lead/architect stories, STAR frames, closing questions

🎤TELL ME ABOUT YOURSELF

Structure = Present ⟶ Past ⟶ Future. ~90 seconds. Then stop.
Present: current role + scope. "SF Tech Lead/Architect, X yrs, N devs."
Name the clouds you own: Sales, Service, Experience Cloud.
Past: 2-3 proof points ⟶ big integration, migration led, perf/data-volume fix.
Every proof point carries a number: time saved, users onboarded, defects down.
Future: why THIS role is the logical next step ⟶ "own solution design end to end."
Mirror JD keywords: LWC, Apex, integration patterns, DevOps tooling.
Professional story only, never personal history.
Close by inviting them to dig into any project named. (Asked: Accenture)

⭐STAR FRAME

S = situation (org, scale, constraint). 1 line only.
T = task (what YOU owned, not the team).
A = action = 60% of airtime ⟶ design choices + trade-offs.
R = result with a metric + what you'd do differently.
Architect twist: always state the alternative you rejected and why.
Keep each story to 2 minutes; park detail, let them pull.
Use "I" for your decisions, "we" for delivery. Never only "we".
Prep 6 stories, reuse them across 20 questions.

🩹WEAKNESS

Real but non-critical. Never "I have no weaknesses". Never core to the job.
4 steps: Name ⟶ Impact you noticed ⟶ Fix you built ⟶ Result.
Sample: "Took too much on myself instead of delegating - I was faster."
Impact: "Became a bottleneck as lead; team waited on my reviews."
Fix: code-review rotation + design-review checklist + pairing juniors.
Result: review turnaround 2 days ⟶ a few hours; team designs without me.
Other safe picks: public speaking, over-polishing docs, impatience with slow decisions.
Always pair the weakness with the corrective action in flight.

💪STRENGTHS

Pick 2-3 that match the JD; back each with a STAR mini-story + numbers.
Solution design under constraints: SF⟶SAP order flow, platform events + middleware queue.
> 200k orders/month inside governor limits, sync failures near zero.
Leading + mentoring: run design reviews + onboarding; 3 juniors own modules solo.
Bridging business + technical: requirements ⟶ design the team builds without rework.
2-3 evidenced strengths > a long list of adjectives.

⚔️CONFLICT + FAILURE

Conflict = disagree on design, not personality clash. Data ends the argument.
Frame: position ⟶ shared goal ⟶ evidence (limits, POC, cost) ⟶ decision ⟶ outcome.
Classic SF conflict: Flow vs Apex, or "just add another trigger" vs handler pattern.
Never trash a colleague, client or ex-employer. Escalation is a last step, not step one.
Failure story: own it, no scapegoat. Missed limit, bad estimate, skipped regression.
Failure must end in a control you added: checklist, test, gate, monitoring.
Say what it cost, then what has never happened since.
Red flag answers: "no failures", or a humblebrag failure.

🧭LEADERSHIP

Lead stories they probe: mentoring, estimation, pushing back on scope, tech debt.
Show influence without authority ⟶ convinced stakeholders with a POC, not a title.
Quantify the team: N devs, M sprints, X stories, Y orgs.
Deadline pressure: cut scope, not quality gates; phase the release.
Say no with an option: "not this sprint, but here is the phased path."
Prep an answer for "biggest architecture decision you regret".

📅DAILY + CLOUDS

Daily: stand-up ⟶ pick sprint stories ⟶ build + unit test config/Apex/LWC.
Then defects/change requests (CDEX/bugs) on current release.
Peer code reviews + commits to the version-control branch.
Coordinate with QA and the BA on acceptance criteria. (Asked: GenPact)
Cloud answer: Sales Cloud ⟶ Leads, Accounts, Contacts, Opportunities.
+ Products, Quotes, Campaigns, forecasting, reports and dashboards. (Cognizant)
Dev skills: declarative (objects, fields, relationships, security, automation).
+ Apex, SOQL/SOSL, triggers, Visualforce, Aura, LWC, integration/APIs, testing, deployment.
+ basic OOP (class, object, attributes) + layered UI / business logic / data model.

🏗️TOUGHEST MODULE

Go-to story: legacy ⟶ Salesforce data migration. (Asked: Cognizant)
Challenges: inconsistent source data, match keys, volume, downtime window.
Match keys ⟶ External ID fields + upsert. De-dup and cleanse at source.
Volume ⟶ Bulk API + Batch Apex, scope of 200, millions of rows.
Load order: parent before child; preserve record ownership and relationships.
Disable automation during load; stay inside governor + API limits.
Cut-over ⟶ phased, with reconciliation reports on counts afterwards.
Close with numbers: records migrated, error rate, how issues were logged and fixed.

❓ASK THEM

Never say "no questions" ⟶ reads as no interest. Ask about work, team, tech.
What does the delivery team look like and where do I fit?
Biggest technical challenges now: data volume, technical debt, integrations?
Release process + DevOps tooling: change sets, Copado, Azure DevOps, SFDX pipelines?
Org maturity: greenfield build or long-lived org needing modernisation?
How is success measured at 90 days and at one year?
Growth path and certification support?
Close: restate interest + ask about next steps in the process.

💰MONEY + NOTICE

Let them name a number first: "what's the band for this role?"
If pushed, give a researched range, not a point, and anchor on total CTC.
Justify with scope: architecture ownership, team size, certs, domain.
Never inflate current salary; offers get verified against payslips.
Notice period: state it exactly, then the buyout/early-release option.
Counter-offer question ⟶ say you have decided to move, and why.
Have a start date ready; vagueness reads as a competing offer.

⇄ Quick Comparison ⇄

WEAK VS STRONG

Weak answer	Question	Strong answer
Life story from college	Tell me about yourself	Present-Past-Future, 90s, JD keywords
"I'm a perfectionist"	Weakness	Real gap + fix + measured result
"We fixed it somehow"	Toughest module	STAR + External IDs, Bulk API, reconciliation
"No, all clear"	Any questions?	Tech debt, DevOps tooling, 90-day success
"Whatever you offer"	Salary	Researched range + scope justification

STORY BANK

Story	Tests	Best used for
SF⟶SAP platform events	Design under constraints	Strengths, architecture decision
Legacy data migration	Scale + governor limits	Toughest module, pressure
Review rotation as lead	Delegation	Weakness, leadership
Flow vs Apex debate	Influence + evidence	Conflict, disagreement
Missed limit in prod	Ownership	Failure, what you changed after

Easy way to remember

Present - Past - Future in 90 seconds
STAR: Action is 60%, Result has a number
Every weakness ships with its fix

Interview tips

Q. Why should we hire you?

A. Match 3 JD needs to 3 evidenced results, then state the gap you close.

Q. Where do you see yourself in 5 years?

A. Owning end-to-end solution design and growing architects - the path this role starts.

Q. Why are you leaving?

A. Forward-looking and neutral: scope and ownership, never money or a bad manager.

Takeaway

Six rehearsed STAR stories with numbers beat any amount of clever improvisation.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

15/16

Salesforce Interview Notes · Lead / Architect

★INTEGRATION ARCHITECT CRIB★

Pattern picking, API limits, auth, middleware, idempotency

🧭Pattern Picker

If SF calls out + needs the answer on screen now ⟶ Request & Reply (sync)
If SF fires the update and does not wait ⟶ Fire & Forget (Outbound Messaging)
If the external system creates/updates SF records ⟶ Remote Call-In
If nightly/hourly sync of changed rows ⟶ Batch Data Synchronization (ETL)
If screen must refresh with no user action ⟶ UI Update Based on Data Changes
If request is queued to devices checking in every 30s ⟶ Fire & Forget + UI Update
If rep must move on and ERP writes Shipping Number back later ⟶ Fire & Forget
Metadata API deploy tool waits for the result ⟶ Request & Reply
If source query takes 7-12 seconds ⟶ never block the UI; async + callback
Sync only when the user genuinely cannot proceed without the reply

📤Push Out Of SF

If real-time push, no Apex, guaranteed delivery ⟶ Workflow/Flow Outbound Message
OM retries up to 24 hours until ACK; carries a session ID for callback
OM is SOAP-only ⟶ REST/JSON target? send OM to middleware to translate
If target is offline 3-4 hours at month-end ⟶ OM queue or ESB queue, no lost txns
If JSON endpoint + real logic ⟶ Apex REST callout, async (@future / Queueable)
If many/unknown subscribers + custom payload ⟶ Platform Events (pub-sub, 72h replay)
If "every field change on this object, no code" ⟶ Change Data Capture
If users must see live record changes ⟶ Streaming API PushTopic (SOQL-based)
If the event starts outside Salesforce ⟶ Generic Streaming
If an outside app polls SF every 2 minutes ⟶ replace polling with Streaming subscribe
Trigger + @future at 300k orders/day ⟶ hits @future daily limit; move to OM/middleware

🪟Bring Data In

If millions of rows + PII must stay in source ⟶ Salesforce Connect external objects
Salesforce Connect: SOQL, global search, reports, related lists; OData 4.0 is writable
External objects: lookup / external lookup / indirect lookup - NEVER master-detail
If only the 5 latest orders show on Account ⟶ external object, do not replicate
If sibling Salesforce orgs need the data ⟶ Cross-Org adapter for Salesforce Connect
If external UI inside SF that must know the user ⟶ Canvas (signed request)
Canvas: SDK matches SLDS; Lifecycle Handler changes endpoint URL dynamically
If a simple unauthenticated web app ⟶ Custom Web Tab, zero code
If org-to-org sharing after a merger ⟶ Salesforce-to-Salesforce
S2S gotcha: no built-in support for parent-child related objects
If Postgres/Heroku sync ⟶ Heroku Connect
If legacy sends SMTP mail ⟶ Email Services + custom InboundEmailHandler

📦Volume + Bulk

If 10M+ records to migrate ⟶ Data Loader / Bulk API (Import Wizard caps at 50k)
If 80M record load ⟶ pre-process, deactivate triggers+workflows, test on Full Sandbox
Bulk API: async, CSV, 10,000 records per batch, monitored via API and Setup UI
Bulk API = fewer round trips + far fewer API calls than row-by-row SOAP
Batch too LARGE ⟶ "Max CPU time exceeded"; too SMALL ⟶ long jobs + record locks
If lock errors on master-detail parents ⟶ Bulk API serial mode, or sort by parent Id
Data skew (>10,000 children per parent) ⟶ random-looking parent lock failures
If nightly delta of 500-1500 Accounts ⟶ getUpdated()/getDeleted() or ETL on SystemModstamp
Never re-extract all 20M opportunities nightly ⟶ pull modified records only
Migration + ongoing feed ⟶ Bulk API once, then API call-in for incremental
Bulk API can load multiple attachments from a single ZIP

🚦Limits + Errors

"Login Rate Exceeded" ⟶ cache the session ID; one integration user per system
Log in only when the session/token expires; not per call, not stored forever
"Concurrent Request Limit Exceeded" (sync callout >5s) ⟶ Continuation from the controller
Continuation = async, up to 3 callouts, frees the long-running concurrent slots
"Max CPU time exceeded" in a trigger ⟶ smaller batch or move logic to Queueable
API call limit hit ⟶ Bulk API, REST composite/batch, or declarative Outbound Messaging
Callout after DML ⟶ "uncommitted work pending"; use @future(callout=true) / Queueable
5-minute batch job that takes >5 minutes ⟶ silent overlap, records quietly missed
Apex callouts burn callout limits, NOT the org's API request limit
WSDL2Apex generates stub code; document-literal only, no RPC-style SOAP
Apex REST services cannot be built or maintained declaratively; need unit + functional tests

🔐Auth + Certs

If Apex callout needs stored creds ⟶ Named Credential, never custom settings/hardcode
If backend rights differ per user ⟶ Named Credential with Per-User identity type
Per-user NC works for Apex callouts + Salesforce Connect OData; NOT outbound messaging
Managed package secrets ⟶ Named Credentials + Protected Custom Settings
If native mobile app ⟶ OAuth User-Agent flow (access + refresh token); REST supports OAuth
If a portal acts for the logged-in employee ⟶ OAuth on behalf of that user, no shared login
OM callback auth: session ID in the message, Enterprise WSDL login(), or REST login
If the remote system must trust Salesforce ⟶ CA-signed cert + two-way (mutual) SSL
On-prem reachability ⟶ DMZ / reverse proxy, whitelist Salesforce IP ranges on firewall
Prove payload origin ⟶ digitally sign with a private key; Base64 is encoding, not security
Platform Encryption protects data at rest, never the network hop - TLS does that
Least privilege: one credential + profile per integration; never a shared Modify All Data admin
Client-side callouts from a page ⟶ Remote Site Settings + CORS whitelist

🔀Middleware / ESB

"We write custom code anyway, why middleware?" ⟶ error handling, orchestration, logging
If complex transformation + process automation ⟶ middleware, minimize Apex
If a system is retired in 3 months ⟶ ESB not point-to-point; least throwaway code
If serial acquisitions with different ERPs ⟶ ESB abstracts Salesforce from each system
Nightly batch middleware must offer ETL + message queuing, not just sync calls
If 3 systems must update as one unit of work with rollback ⟶ message-oriented middleware
If credentials must not live in Salesforce ⟶ ESB holds them; SF just fires the message
If analytics needs SF + other sources ⟶ ETL joins them into one dataset first
If consolidating orgs ⟶ ETL staging tables for cleansing and standardization
Middleware also de-dupes repeat calls, retries, replays, and logs the payloads

🔁Idempotency + Quality

If inbound creates may repeat ⟶ upsert() on an External ID, never create()
Second guard: unique message ID, de-duplicated by the middleware
External ID: indexed + searchable, upsertable, but NOT unique unless you tick Unique
External ID can be Text, Number, Email or Auto Number - a formula field cannot
Cannot flag another External ID ⟶ per-object External ID cap or index cap reached
No natural key? ⟶ hash first+last+street into a text field and mark it External ID
If loading a large purchased list ⟶ de-duplicate off-platform BEFORE the load
Ongoing in-org prevention ⟶ Duplicate Management rules + AppExchange data tool
Integration write refires the rule ⟶ exclude the Integration User in entry criteria
Duplicate orders from OM ⟶ compare OM delivery status log against ESB logs
OM retry after an outage ⟶ expect duplicates + out-of-order; make the consumer idempotent

🧪Test + Environments

Callout tests: DML OUTSIDE Test.startTest, the callout INSIDE start/stop
Mock with HttpCalloutMock / WebServiceMock - live callouts are banned in tests
Customer CSV test data ⟶ load as a Static Resource + Test.loadData()
If perf + UAT must mirror prod ⟶ Full Sandbox; Partial Copy for integration testing
UAT in a Developer sandbox ⟶ false confidence on performance and volume
Prod data into a sandbox ⟶ Full/Partial refresh or Data Loader
"New release broke old features" ⟶ Regression testing + continuous integration
"Meets the spec but feels slow" ⟶ Performance testing + UAT
Data model churn mid-project ⟶ requirements traceability matrix + CI
Diagnosing an inbound REST call ⟶ Workbench

⇄ Quick Comparison ⇄

Scenario To Mechanism

If the scenario says	Choose	Because
Real-time push, zero code, must not lose messages	Outbound Messaging	24h retry queue + session ID callback
Target speaks REST/JSON only	Apex callout, or OM to middleware	OM emits SOAP only
Many or unknown subscribers, custom payload	Platform Events	Pub-sub with 72h replay window
Every field change on an object, no code	Change Data Capture	Automatic record-change events
Screen updates with no user action	Streaming API PushTopic	UI Update Based on Data Changes
Event originates outside Salesforce	Generic Streaming	Non-record notifications
Millions of rows, PII must stay in source	Salesforce Connect OData	External objects, nothing copied
Nightly sync of changed rows only	ETL + Bulk API + getUpdated()	Batch Data Synchronization
Page callout takes 7-12 seconds	Continuation	Dodges concurrent long-running limit
Three systems, one transaction, rollback	Message-oriented middleware	Orchestration + compensating rollback
External UI in SF that knows the user	Canvas signed request	Passes user and org context

Enterprise vs Partner WSDL

Enterprise WSDL	Aspect	Partner WSDL
Strongly typed, org-specific	Shape	Generic sObject, loosely typed
Regenerate after every schema change	New object	One stub covers current + future objects
No	Runtime field discovery	Yes - inspect field names at runtime
Supports WS-Security	Security	REST API is the OAuth-friendly option
Single org, stable data model	Best for	ISV / multi-org, dynamic model

Easy way to remember

Six patterns: Request&Reply, Fire&Forget, Batch Sync, Remote Call-In, UI Update, Data Virtualization
OM = SOAP + 24h retry + session ID
Upsert on External ID = idempotent

Interview tips

Q. Outbound Messaging vs Apex callout?

A. OM is declarative SOAP with a 24h retry queue; Apex gives REST plus logic, but you own retries.

Q. Target system goes offline for hours - now what?

A. Queue it - OM retry or an ESB message queue; never a synchronous trigger callout.

Q. How do you stop an integration creating duplicates?

A. Upsert on External ID, unique message ID in middleware, and de-dupe off-platform before load.

Takeaway

Read the cue - who initiates, how fast, how much - then make every integration queued, idempotent, and least-privileged.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

16/16

Salesforce Interview Notes · Lead / Architect

★COMPANY-WISE FOCUS★

What each interviewer actually asked - revise by company

🅰️Accenture

Callout from Salesforce ⟶ where do credentials live? (Named Credential)
Future method limits: 50 per transaction, 250,000 / 24h.
Batch Apex from a trigger - allowed? Callout from a trigger?
Remove duplicates from a List in Apex (Set / Map).
Governor limits - name three. SOQL rows per transaction = 50,000.
Delete lookup child records when parent is deleted.
External ID vs Unique ID.
SOQL for unique designations of employees (GROUP BY).
Aura framework = MVCC / component-based.
Dynamic approval process; do workflow rules chain each other?
Lead conversion + assignment rules.
Test.setCurrentPage() / Test.setPage() in Apex tests.
Case management. Plus HR: "Tell me about yourself".

☁️Cloud 360

Platform events - when to use them.
Permission sets: grant perms to one user, one object.
Record access restriction levels; OWD default levels.
OWD internal Private ⟶ can external user be Read/Write?
Two profiles, R/W vs W/D on same object ⟶ most permissive wins.
Database.AllowsCallouts - where used.
Schedulable = 1 method (execute). Queueable = 1 method (execute).
global access modifier; break vs continue in a for loop.
Find duplicates using an index.
Component event propagation phases (capture ⟶ bubble).
Aura navigate-to-URL; redirect from JS controller.
Child fires event with X ⟶ parent or super parent?
Expose component to community via interfaces.
setRedirect(true) vs (false); duplicate method names across extensions.
Data Loader Export vs Export All. Custom Settings vs Custom Metadata.
Community deploy needs Site.com / Network + Experience Bundle.

🧩Cognizant

Do you HAVE an exception handling framework? Explain yours.
One-time migration of lakhs of records - approach.
MuleSoft <⟶ Salesforce integration; MuleSoft vs Heroku.
High-volume data in an integration - how to handle.
Trigger best practices + bulkification example + its aim.
Order of execution on record save - know it cold.
Aura event life cycle; framework; design attributes; interfaces.
CI/CD tool used + Git commands used.
Sales Cloud life cycle + its objects.
Lead vs Opportunity - why Opportunity?
HR: most challenging module; which cloud.

🔌GenPact

Platform Events - worked on them? What are they?
REST APIs; call external web service from Salesforce.
Authentication handling (OAuth flows, Named Credentials).
Future from Batch? Future from Future? Both NO.
How many future methods per class ⟶ 50 per transaction.
Accounts + related Contacts in Aura via custom code.
JS controller vs helper in Aura.
Slow component ⟶ how to improve performance.
Scenario: notify managers on an Opportunity.
Record types driving picklist values.
HR: what do you do on a daily basis?

📄Appcino

Only 4 Qs - all Aura + Visualforce.
Component event vs application event.
View state in Visualforce - what it is.
Max view state size = 170 KB (else error; use transient).
Action region - what it does.
Verdict: VF-heavy shop. Revise view state + rerender.

🚙Mahindra & Mahindra

Give a user permission to access an Aura component.
Batch vs Future vs Queueable - the classic contrast.
Queueable example (implements Queueable, execute).
Governor limit hit ⟶ how do you rectify it?
SOQL returns more than 50,000 records ⟶ use Batch / query locator.
Trace where a quick action / section on a Lightning page comes from.
doInit in Aura (init handler).
Partner Community - what it is.
Verdict: short, async + Aura + community basics.

♻️Repeat Offenders

Q36 callout to external web service ⟶ Accenture AND GenPact.
Q480 Aura framework ⟶ Accenture AND Cognizant.
Async limits show up at Accenture, GenPact, Cloud 360, M&M.
Aura events appear at Appcino, Cloud 360, Cognizant, GenPact.
Governor limits + 50,000 SOQL rows: Accenture and M&M.
Every company slipped in 1-3 HR / experience questions.

⇄ Quick Comparison ⇄

Future vs Queueable

@future	Feature	Queueable
static void only	Signature	execute(QueueableContext)
primitives only	Params	sObjects + Apex types
no job Id	Monitoring	returns AsyncApexJob Id
cannot chain	Chaining	1 child per parent
50 per transaction	Limit	50 sync / 1 async

Settings vs Metadata

Custom Settings	Feature	Custom Metadata
no	Deployable records	yes
yes (List/Hierarchy)	Per user/profile	no
no limit consumed	SOQL limits	queries do not count
runtime config data	Best for	app config shipped with code

Easy way to remember

Accenture = limits + admin
Cloud 360 = security + Aura
Cognizant = integration + order of execution

Interview tips

Q. Which company is the async-heavy one?

A. Mahindra and GenPact - batch vs future vs queueable, chaining rules.

Q. Cloud 360's favourite area?

A. Sharing/OWD plus Aura events - know propagation phases and OWD guest access.

Q. One answer that covers three companies?

A. Named Credentials for callouts - Accenture, GenPact and Cognizant all ask it.

Takeaway

Same 20 topics rotate - master async limits, OWD, order of execution and Aura events.

💡 Read → Recall out loud → Explain to someone → Answer in the room

↑ back to index

Tag: experience architect intervew and answers

Short Notes Salesforce interview Questions and Answers Summary