Quick definitions for Salesforce, AWS, and software engineering terms. Search or browse by platform.
70 terms
Salesforce's platform for building autonomous AI agents that take actions in the CRM. Agents have topics (what they handle), actions (Apex, Flows), and guardrails. Built on the Einstein Trust Layer.
Application Load Balancer. AWS Layer 7 (HTTP) load balancer. Features: path-based routing, host-based routing, weighted target groups (canary deploys), sticky sessions, slow start, native gRPC/WebSocket support. The default choice for HTTP services. Does not provide static IPs (use NLB for that).
A distributed commit log for event streaming. Not a message queue -- data persists based on retention policy, not consumption. Key concepts: topics (named streams), partitions (unit of parallelism), consumer groups (cooperative readers), offsets (position tracking). Used for high-throughput, ordered, replayable event pipelines.
Salesforce's proprietary programming language, similar to Java. Runs on the Salesforce platform for custom business logic, triggers, and web services. Subject to governor limits.
AWS serverless query service. Runs SQL queries directly on S3 data, charges $5 per TB scanned. No infrastructure to manage. Good for ad-hoc analysis and infrequent queries. Use partitioned Parquet to minimize scan costs. For heavy concurrent use, consider Redshift instead.
AWS service for building AI applications with foundation models (Claude, Llama, Titan). Single API, built-in RAG via Knowledge Bases, content filtering via Guardrails, and IAM-based access control.
Run two identical environments (blue and green). Deploy to the idle one, switch traffic when verified. Instant rollback by switching back. Zero-downtime deployment pattern.
Run two identical environments (blue = current, green = new). Deploy to green, switch all traffic instantly. Rollback by switching back. Zero downtime. Simpler than canary (all-or-nothing switch) but riskier since 100% of traffic shifts at once.
Deploy new version alongside old, route a small percentage of traffic to it (5% -> 25% -> 50% -> 100%). Monitor metrics at each stage. Rollback by reducing percentage to 0. Safer than blue-green because you catch issues before full exposure.
A distributed system can guarantee at most two of three: Consistency (all reads see the latest write), Availability (every request gets a response), Partition tolerance (system works despite network splits). In practice, you choose between CP and AP.
Cloud Development Kit. Write infrastructure as code in TypeScript, Python, Java, or C#. Synthesizes to CloudFormation. Constructs are reusable building blocks. Higher-level abstraction than raw CloudFormation templates.
A test that captures what legacy code currently does, not what it should do. Written before refactoring to ensure you don't accidentally change behavior. Assert consistency, not correctness. If the test still passes after your change, behavior is preserved. From Michael Feathers' Working Effectively with Legacy Code.
Continuous Integration / Continuous Delivery. Automate: build, test, and deploy on every code change. CI catches bugs early. CD gets working code to production faster. Tools: GitHub Actions, Jenkins, CircleCI.
A pattern that stops calling a failing service after repeated errors. States: Closed (normal), Open (calls blocked), Half-Open (testing if service recovered). Prevents cascade failures in distributed systems.
Infrastructure as Code for AWS. Define resources in YAML or JSON templates. AWS creates, updates, and deletes resources as a stack. Alternative: AWS CDK (write IaC in TypeScript, Python, etc.).
AWS CDN service. Caches content at 600+ edge locations worldwide. Also handles TLS termination, WAF integration, and edge compute (CloudFront Functions, Lambda@Edge). Key for reducing latency, protecting origins from traffic spikes, and serving static assets cheaply.
Monitoring and observability service. Collects metrics, logs, and traces. Set alarms on thresholds. Create dashboards. Logs Insights for querying log data with a SQL-like syntax.
When removing an instance from a load balancer, letting in-flight requests finish before fully deregistering. Configured via deregistration_delay on target groups (default 300s, usually set to 30-60s). Without it, active connections drop mid-request. Deregistration delay adds directly to deployment time -- tune it based on your API latency.
Command Query Responsibility Segregation. Separate the write model (commands) from the read model (queries). Write to a normalized database, project to denormalized read stores. Common with event sourcing.
App-level configuration stored as metadata (not data). Deployable, packageable, and queryable in SOQL. Use for feature flags, mapping tables, and configuration that should move between orgs.
A batching library for GraphQL that collects individual data fetch requests within a single event loop tick, then fires one batch query. Solves the N+1 problem where resolving N items each trigger a separate database call. Create per-request, never share across requests.
A queue where messages that fail processing after multiple retries are sent. Prevents poison messages from blocking the main queue. Monitor and reprocess DLQ messages after fixing the root cause.
Fully managed NoSQL database. Key-value and document store with single-digit millisecond latency at any scale. Uses partition keys for distribution. Supports on-demand and provisioned capacity modes.
Elastic Compute Cloud. Virtual servers in AWS. Choose instance type (CPU, memory, GPU), AMI (OS image), and VPC placement. Pricing: On-Demand, Reserved, Spot, Savings Plans.
Elastic Container Service runs Docker containers. Fargate is the serverless launch type -- no EC2 instances to manage. You define task definitions (CPU, memory, image) and ECS handles placement and scaling.
Extract, Transform, Load. A data pipeline pattern: pull data from sources, clean/reshape it, then put it in a destination (warehouse, data lake). ELT is the reverse: load raw data first, then transform inside the destination using its compute power.
Systems communicate by producing and consuming events (facts about something that happened). Decouples services in time and space. Components: event producers, event brokers (Kafka, SQS), event consumers.
A consistency model where all replicas will eventually converge to the same state, but reads might return stale data temporarily. Common in distributed databases and caches. Trade-off for higher availability.
A conditional in code that checks external configuration to decide which code path to execute. Separates deploying code from releasing features. Types: release (temporary), ops/kill switches (safety), experiment (A/B tests), and permission (entitlements). Clean up after full rollout to avoid flag debt.
Salesforce's declarative automation tool. Build screen flows (user-facing), record-triggered flows (on data change), and scheduled flows without writing code. Replaced Workflow Rules and Process Builder.
AWS managed Apache Spark service for ETL. Components: Data Catalog (Hive-compatible metastore for schemas), Crawlers (scan data sources, infer schemas), ETL Jobs (PySpark/Scala on managed clusters), Data Quality (validation rules). Cold start is 5-10 minutes -- use Lambda for small datasets.
Hard runtime limits enforced by Salesforce to prevent any single tenant from monopolizing shared resources. Examples: 100 SOQL queries per transaction, 150 DML operations, 6MB heap size.
A query language for APIs where the client specifies exactly which fields it wants. One endpoint serves all clients. Solves overfetching (REST returns too much data) and underfetching (need multiple REST calls). Trade-offs: caching is harder (POST to one URL), N+1 query problem requires DataLoader.
Identity and Access Management. Controls who can do what in AWS. Users, groups, roles, and policies. Policies are JSON documents with Allow/Deny statements. Use roles for services, not access keys.
An operation is idempotent if running it multiple times produces the same result as running it once. Upserts are idempotent, inserts are not. Required for reliable retry logic in distributed systems.
A client-generated unique identifier (UUID) attached to API requests to prevent duplicate processing. The server checks if it has seen this key before. If new: process and store result. If duplicate: return stored response. Prevents double charges, duplicate records. Must fingerprint the request body to prevent key reuse across different operations.
An operational feature flag that lets you instantly disable non-critical functionality during incidents. Must be fast to flip (admin UI, not code deploy) and default to safe when the flag service is unreachable. Examples: disable recommendations, queue notifications instead of sending, enable read-only mode.
Serverless compute. Upload code, AWS runs it on demand. No servers to manage. Pay per invocation and duration. Max 15 minutes per execution. Supports Node.js, Python, Java, Go, .NET, Ruby.
Salesforce's modern UI framework based on Web Components standards. Uses HTML templates, JavaScript controllers, and CSS. Replaced Aura Components as the recommended front-end approach.
Low-Rank Adaptation. A parameter-efficient fine-tuning method that freezes base model weights and trains small adapter matrices (0.5% of total parameters). Results are close to full fine-tuning with much less GPU memory. The saved adapter is 50-200 MB instead of 16+ GB. QLoRA adds 4-bit quantization to reduce memory further.
A distributable bundle of Salesforce components (Apex, LWC, objects, flows) with a namespace. Published on AppExchange. Code is IP-protected and upgradable.
Managed Streaming for Apache Kafka. AWS runs the Kafka brokers and ZooKeeper/KRaft. You manage topics, partitions, and consumers. MSK Serverless removes even more operational burden.
Salesforce configuration that stores the endpoint URL and authentication for external services. Keeps credentials out of code. Supports OAuth, JWT, and basic auth. The right way to call external APIs from Apex.
Network Load Balancer. AWS Layer 4 (TCP/UDP) load balancer. Key advantages over ALB: static IPs per AZ, ultra-low latency (microseconds), TLS passthrough, required for VPC PrivateLink. Use when you need static IPs, raw TCP/UDP, or lowest latency. Chain NLB -> ALB for static IPs + HTTP routing.
The ability to understand a system's internal state from its external outputs. Three pillars: logs (events), metrics (measurements), traces (request flows). Not just monitoring -- it's about asking new questions.
The baseline level of record access for your org. Set per object: Private (owner only), Public Read Only, Public Read/Write. The starting point for all sharing rules.
A columnar file format for analytics. Queries can skip columns they don't need (column pruning). Built-in compression (Snappy gives 5-10x reduction over CSV). Schema embedded in file. The default format for data lake processed zones. Supported by Athena, Spark, Glue, Redshift Spectrum.
A collection of settings and permissions that extend a user's access without changing their profile. Can be grouped into Permission Set Groups. The future of Salesforce access control.
An evaluation metric for language models. Measures how "surprised" the model is by correct outputs. Computed as exp(average cross-entropy loss). Lower is better. Perplexity of 1.0 means perfect prediction. Compare train vs test perplexity to detect overfitting. Not a complete metric -- always pair with task-specific evaluation and human review.
Salesforce's pub/sub messaging system. Publish events from Apex, Flows, or external systems. Subscribers receive them in near real-time. Used for event-driven architecture within and beyond Salesforce.
Retrieval-Augmented Generation. Before sending a prompt to an LLM, retrieve relevant documents from a knowledge base and include them as context. Reduces hallucination by grounding answers in real data.
Controlling how many requests a client can make in a time window. Algorithms: token bucket, sliding window, fixed window, leaky bucket. Protects services from abuse and overload.
Relational Database Service. Managed databases: PostgreSQL, MySQL, MariaDB, Oracle, SQL Server. Handles backups, patching, and failover. Multi-AZ for high availability. Read Replicas for read scaling.
Simple Storage Service. Object storage with 11 nines of durability. Store files (objects) in buckets. Storage classes: Standard, Intelligent-Tiering, Glacier (archive). Backbone of most AWS architectures.
AWS's full ML platform for training, tuning, and deploying models. Supports built-in algorithms (XGBoost, Linear Learner), custom training scripts, Hugging Face integration, and managed endpoints. JumpStart provides pre-configured fine-tuning for popular models. Spot training reduces costs up to 90%. Not the same as Bedrock (which is for inference only).
A temporary, configurable Salesforce org used for development and testing. Created via Salesforce CLI, lasts up to 30 days. The foundation of Salesforce DX development workflow.
Three bash safety flags that should start every production script. -e: exit on any non-zero exit code. -u: treat unset variables as errors (prevents rm -rf /$EMPTY_VAR). -o pipefail: pipeline returns the exit code of the last failed command, not just the last command. Without these, scripts continue silently after errors.
Splitting a database across multiple servers by a shard key. Each shard holds a subset of data. Enables horizontal scaling. Trade-off: cross-shard queries are expensive. Choose your shard key carefully.
Rules that extend record access beyond OWD. Criteria-based (match field values) or owner-based (share records owned by specific users/roles). Applied on top of OWD, never restrict further.
SLI: the metric (p99 latency, error rate). SLO: internal target for the SLI (p99 < 200ms). SLA: external contract with penalties (99.95% uptime or refund). SLOs should be stricter than SLAs.
Salesforce Object Query Language. SQL-like syntax for querying Salesforce records. Does not support JOIN -- you traverse relationships instead (Account.Contacts, Contact.Account.Name).
Simple Queue Service. Fully managed message queuing. Standard queues (at-least-once, best-effort ordering) and FIFO queues (exactly-once, strict ordering). Decouples producers from consumers.
Serverless workflow orchestration. Define state machines in JSON (ASL). Coordinates Lambda, ECS, SQS, and other services. Standard (long-running) and Express (high-volume, short) workflow types.
An incremental migration strategy. Put a proxy in front of the legacy system, route some traffic to the new system and the rest to the old. Migrate one endpoint at a time. Each migration is small and reversible. Named after a tree that grows around its host. Almost always better than a big-bang rewrite.
Apex code that executes before or after DML events (insert, update, delete, undelete) on a Salesforce object. Best practice: keep triggers thin, delegate logic to handler classes.
Time To Live. How long a cached item stays valid before being evicted. Set per key or per cache policy. Short TTL (seconds-minutes) for frequently changing data. Long TTL (hours-days) for stable data. Add jitter (random variation) to prevent thundering herd when many keys expire simultaneously.
A methodology for building SaaS applications. Key principles: config in environment variables, stateless processes, port binding, disposability, dev/prod parity, logs as event streams.
Virtual Private Cloud. Your isolated network in AWS. Contains subnets (public/private), route tables, internet gateways, NAT gateways, and security groups. The networking foundation for everything.
In AWS Auto Scaling, pre-provisioned EC2 instances in Stopped state ready to start quickly. Cuts scale-out time from 5 minutes (cold launch) to under 1 minute. You pay for EBS volumes while instances are stopped. Configure reuse_on_scale_in to return instances to the pool instead of terminating.
An HTTP callback -- your server POSTs data to another server's endpoint when an event occurs. Provides at-least-once delivery (consumer will see duplicates). Must be signed (HMAC-SHA256), consumer must respond fast (200 immediately, process async), and handling must be idempotent. Common in Stripe, GitHub, Salesforce integrations.
