Release notes, new features, deprecations, and announcements from Salesforce and AWS.
68
Total
20
Salesforce
45
AWS
45 notices for AWS
Article #99 covers the Kinesis family (Data Streams, Firehose, Data Analytics), shards, partition keys, Lambda consumers with parallelization_factor, Firehose zero-code delivery, enhanced fan-out, batch ingestion, and 6 common mistakes.
Article #98 covers RDS vs Aurora architecture, instance sizing (Graviton r7g), gp3 vs io2 storage, Multi-AZ failover, Aurora Serverless v2, RDS Proxy for Lambda, Performance Insights, and 6 common production mistakes.
Article #97 covers stream view types, Lambda consumers with partial batch failures, idempotent processing, fan-out via SNS, cross-region replication, OpenSearch indexing, and DynamoDB Streams vs Kinesis comparison.
Article #96 covers Serverless vs node-based, Graviton instances, cluster mode, Lambda connection patterns, TTL jitter, distributed locking, rate limiting, and 6 common ElastiCache production mistakes.
Article #95 covers User Pools vs Identity Pools, OAuth flows, JWT tokens (ID/Access/Refresh), federated login (Google, SAML), Lambda triggers for custom claims and DB sync, groups and RBAC, API Gateway JWT authorizer integration, and 6 common mistakes.
Article #94 covers the three API types (REST $3.50/M, HTTP $1.00/M, WebSocket), Lambda integrations, authorization patterns (JWT, Lambda authorizer, IAM), throttling, caching, CORS, custom domains, and why HTTP API is the right default.
Every API needs pagination. Covers offset (simple but fragile), cursor-based (stable, O(1)), and keyset (maximum performance). Plus Relay Connection spec, DynamoDB LastEvaluatedKey, SOQL OFFSET, Elasticsearch search_after, and the fetch+1 trick.
The 90th article! Covers the inverted pyramid (conclusion first), writing for scanning, README templates, ADRs, runbooks for 3 AM, API docs, code comments, writing RFCs with options and tradeoffs, changelog patterns, and onboarding docs.
Krishna's Tech Blog reached 90 articles covering Salesforce, AWS, AI/ML, and general engineering. The Learning section now has 57 flashcard decks with 1,140 cards, 310 quiz questions, and 20 certification study guides. 70 glossary terms. 9 article series.
The four staff engineer archetypes, influence radius shift, writing RFCs, mentoring without micromanaging, navigating ambiguity, cross-team influence, picking the right battles, managing technical debt strategically, and the promotion trap.
The four DR strategies (backup-restore to multi-site active-active), Aurora Global Database, DynamoDB Global Tables, Route 53 failover, game days with AWS FIS, runbooks, and the cost tradeoffs that drive DR decisions.
The Learning section hit major milestones: 300 multiple-choice quiz questions across Salesforce, AWS, and General topics. 56 flashcard decks with 1,120 cards. 20 certification study guides. Spaced repetition, smart review, and daily deck picks.
Strangler fig pattern, characterization tests, safe refactoring steps (extract method, dependency injection), feature flags for migration, the boy scout rule, and when to actually rewrite vs incrementally improve.
set -euo pipefail, cleanup traps, lock files with stale PID detection, getopts argument parsing, retry logic, cron job setup, ShellCheck linting, and a complete production-ready template script.
The prompt -> RAG -> fine-tune decision tree, LoRA and QLoRA for parameter-efficient training, data preparation patterns, SageMaker and Bedrock custom models, evaluation metrics, real cost breakdown, and 5 common mistakes.
Event buses, content-based pattern matching, targets (Lambda, SQS, Step Functions, API Destinations), Scheduler, Pipes, archive and replay, EventBridge vs SQS vs SNS decision framework, and 6 common mistakes.
EC2 ASGs, target tracking vs step vs scheduled scaling, ECS auto scaling, Lambda concurrency, DynamoDB capacity modes, warm pools, predictive scaling, cost optimization with Spot instances, and the multi-layer scaling problem.
Silo, pool, and bridge isolation models, database strategies, tenant context middleware with AsyncLocalStorage, automatic query scoping, PostgreSQL RLS, noisy neighbor mitigation, and onboarding automation.
Every engineer learns SOLID in interview prep and forgets it in production. This article covers all five principles with TypeScript code examples, bad/good comparisons, and honest guidance on when each principle is worth the overhead and when it creates unnecessary complexity.
Idempotency keys, retry-safe payment processing, DynamoDB conditional writes, and distributed idempotency across microservices. Includes FastAPI + Redis implementation, testing strategies, and the mistakes that cause double charges.
JSON structured logging, correlation IDs for distributed tracing, Python structlog and Node.js pino code examples, CloudWatch Logs Insights queries, log aggregation pipelines, and the mistakes that make logs useless at 3 AM.
Separate deploying code from releasing features. Covers flag types (release, ops, experiment, permission), percentage rollouts with deterministic hashing, AWS AppConfig, kill switches, flag debt, and the cleanup strategy that actually works.
Aurora Limitless Database is now GA. It automatically scales PostgreSQL-compatible Aurora beyond the limits of a single writer instance by sharding data across multiple database shards transparently. Applications connect to a single endpoint and the router distributes queries. Supports distributed transactions and cross-shard joins.
The new Step Functions Workflow Studio v2 features an improved drag-and-drop editor with inline testing, real-time ASL preview, reusable workflow snippets, and direct integration with over 200 AWS services. The editor now runs in your browser without needing the console.
Claude 4 Sonnet and Haiku are now available in Amazon Bedrock across US and EU regions. Sonnet offers improved reasoning and coding capabilities. Haiku provides the fastest response times for cost-sensitive workloads. Both support the Bedrock Converse API and Knowledge Bases.
Claude 4 Opus, Sonnet, and Haiku are now available in Amazon Bedrock. Includes improved tool use, better instruction following, and expanded context windows.
Glue 5.0 includes Spark 3.5 with improved performance for joins and aggregations, Python 3.11 runtime, native Iceberg table support, and faster job startup times (reduced by up to 40%). Existing Glue 4.0 jobs continue to work but should be migrated for performance gains.
Python 3.13 is now available as a managed runtime in AWS Lambda. Includes the new JIT compiler for improved performance and free-threaded mode (experimental).
R8g instances powered by Graviton4 processors deliver up to 30% better compute performance over Graviton3. Available in US East, US West, and EU West regions.
S3 Express One Zone, the single-digit millisecond latency storage class, is now available in 10 additional regions. Up to 10x faster than S3 Standard for frequently accessed data.
Cognito User Pools now support passkeys and WebAuthn as a primary authentication factor. Users can sign in with biometrics (Face ID, fingerprint) or hardware security keys instead of passwords. Available in all commercial regions.
HTTP APIs now support mutual TLS (mTLS) authentication, previously only available on REST APIs. Clients present certificates validated against a truststore in S3. Useful for B2B integrations and zero-trust architectures.
EventBridge now supports detailed execution logging at error, info, and trace levels for both event buses and Pipes, with output to CloudWatch Logs, S3, or Kinesis Data Firehose. Logs capture matched rules, target invocation results, and optionally full event payloads, enabling faster root-cause analysis with no additional service charge.
Lambda SnapStart is now GA for Python 3.12+ and .NET 8+ managed runtimes, reducing cold start latency from several seconds to sub-second with minimal or no code changes. The feature was later expanded to 23 additional regions in June 2025.
Aurora Serverless v2 can now scale down to 0 ACUs, automatically pausing the database after a configurable idle period with no compute charges while paused. This resolves a long-standing gap between v1 and v2, making Aurora Serverless v2 a true pay-per-use option for dev and test workloads.
CloudWatch Application Signals is now GA in 28 commercial regions, providing an OpenTelemetry-compatible APM dashboard that surfaces volume, availability, latency, faults, and errors per service. Runtime metrics support for garbage collection, memory, and CPU was added in November 2024, and GovCloud availability followed in November 2025.
Amazon Q Developer, AWS's generative AI-powered coding assistant, reached general availability with support for inline code suggestions, conversational coding in the IDE, and an autonomous agent for software development that can implement features and fix bugs across entire codebases. The assistant covers the full software development lifecycle from design through testing and maintenance.
Guardrails for Amazon Bedrock reached general availability, giving teams a centralized policy layer that sits between applications and any foundation model on Bedrock. Controls include content filtering for hate, violence, and misconduct at configurable thresholds; topic denial to block off-topic prompts; PII redaction and anonymization; grounding checks that flag hallucinated responses against a reference source; and word and phrase blocklists. Guardrails apply to both user inputs and model outputs and work across Claude, Titan, Llama, Cohere, and other models available in Bedrock, requiring no changes to model code. In December 2024 Bedrock reduced Guardrails pricing by up to 85 percent.
At AWS re:Invent 2023, Amazon Bedrock Agents and Knowledge Bases reached general availability, enabling fully managed RAG pipelines and multi-step autonomous agents without custom orchestration code. Knowledge Bases automatically syncs documents from S3, chunks and embeds them into a vector store, and retrieves relevant context at inference time, while Agents can invoke Lambda actions and API schemas to complete multi-turn tasks.
Announced at AWS re:Invent 2023, ElastiCache Serverless lets you create a fully managed, highly available cache for Redis 7.1 or Memcached 1.6 in under a minute with no cluster sizing or capacity planning required. The service scales instantly to meet demand, stores data redundantly across multiple Availability Zones with a 99.99% SLA, and charges only for data stored and ECPUs consumed.
Cost Optimization Hub is a new AWS Billing and Cost Management feature that aggregates over 15 types of cost optimization recommendations, including EC2 rightsizing, Graviton migration, idle resource detection, and Savings Plans suggestions, into a single dashboard across all AWS Organizations member accounts and regions. The feature is available at no additional charge in all commercial regions outside China and GovCloud.
AWS launched EC2 Instance Connect Endpoint (EIC Endpoint), a free feature that lets you open SSH or RDP sessions to EC2 instances in private subnets directly from the AWS console, CLI, or any SSH client -- with no bastion host, no public IP on the instance, no VPN, and no agent required. An EIC Endpoint acts as a VPC resource that proxies the connection from your workstation through AWS private networking to the target instance, with all sessions logged in CloudTrail and access controlled via IAM. The feature is available at no additional charge in all commercial and GovCloud regions and supports both Linux and Windows instances.
AWS Lambda now supports response payload streaming, a new invocation pattern that lets functions progressively send partial responses back to clients as data becomes available rather than waiting for the entire payload to be ready. Functions can return payloads up to 20 MB via function URLs using RESPONSE_STREAM invoke mode or through the new InvokeWithResponseStream SDK API, compared to the 6 MB limit for buffered responses. The feature dramatically improves time-to-first-byte for web and mobile workloads such as AI-generated content and large file downloads. Node.js 14.x and later runtimes support native streaming helpers, and custom runtimes can use the runtime API directly.
AWS Step Functions launched Distributed Map, a new Map state mode that can iterate over millions of objects in Amazon S3 and launch up to 10,000 parallel child workflow executions concurrently -- far exceeding the previous 40-iteration inline Map limit. The feature enables serverless ETL, batch inference, and large-scale data processing without managing infrastructure.
Amazon RDS introduced a new Multi-AZ deployment option for MySQL 8.0.28+ and PostgreSQL 13.4+ that places one primary and two readable standby instances across three Availability Zones. The configuration delivers up to 2x faster transaction commit latency compared to single-standby Multi-AZ and supports automated failovers typically completing in under 35 seconds.
