Lab
Les dernieres editions de ma newsletter technique : architecture logicielle, cloud et engineering.
Synchronise en direct avec softwareletters.comSL#134 - AWS Services (24/120) - Amazon MemoryDB: The Transaction Log That Makes Redis a Database, and a Failover Drill That Counts Your Acknowledged Writes
Last time we looked at ElastiCache, the cache you put in front of a database. MemoryDB is the same wire protocol with a different promise: the data in memory is the database, and an acknowledged write survives a failover.
Editions récentes
SL#134 - Factory & Abstract Factory: The Pattern You Already Use Every Time You Call open()
Most factory code is an if-statement with a class name on it. Here is the one problem the pattern really solves, the two GoF versions, and the registry you will actually write.
- design patterns
SL#134 - Your Timeout Is Destroying the Connection Pool It Was Supposed to Protect
Inngest's October 10 outage was not a database failure. It was a deadline that cancelled a queued query, a pooler that closed the connection in response, and a loop that fed itself for three hours.
- DevOps
SL#133 - AWS Services (23/120) - Amazon ElastiCache: The Cache-Invalidation Problem You Are Actually Buying, and a Hit Rate You Measure
ElastiCache is a managed Valkey, Redis OSS, or Memcached endpoint, and the managed part is the easy half. We create a serverless Valkey cache with boto3, put a cache-aside layer in front of a slow fake database, and read the hit rate from CloudWatch. Plan for 60 minutes and well under a dollar.
- ElastiCache
- Tutorial
- AWS Services
SL#132 - AWS Services (22/120) - Amazon Aurora DSQL: Optimistic Concurrency, the Retry Loop You Now Own, and a Conflict You Trigger on Purpose
Aurora DSQL is serverless, PostgreSQL-compatible, and active-active, and it has no locks at all. We create a cluster with boto3, connect with an IAM token, and cause a write conflict on purpose to measure what optimistic concurrency costs. Plan for 60 to 90 minutes and under a dollar.
- Aurora DSQL
- Tutorial
- AWS Services
SL#131 - AWS Services (21/120) - Amazon Aurora: The Shared Storage Volume That Makes It Not Just RDS, and a Failover You Measure to the Second
Aurora keeps the database engine you know and replaces the storage underneath it. We build a Serverless v2 PostgreSQL cluster with a writer and a reader, point a connection prober at it, force a failover, and read the outage off the clock. Plan 60 to 90 minutes and a small bill.
- Aurora
- Tutorial
- AWS Services
SL#129 - AWS Services (20/120) - Amazon RDS: What Managed Actually Manages, a Failover You Trigger on Purpose, and the Downtime You Can Measure
RDS runs the database engine you already know and takes over everything underneath it. We build a Multi-AZ PostgreSQL instance, hammer it with a connection prober, force a failover, and read the outage off the clock. Plan 60 to 90 minutes and a small bill.
- RDS
- Tutorial
- AWS Services
SL#128 - AWS Services (19/120) - Amazon DynamoDB: Single-Table Design, the Hot Partition You Will Create, and a Table You Can Break on Purpose
DynamoDB is a key-value store that punishes you for modeling it like a database. We build an orders table with a GSI, a transaction, and a deliberate hot key, then read the throttling like a postmortem. Plan 60 to 90 minutes and well under a dollar.
- Tutorial
- AWS Services
- DynamoDB
SL#127 - AWS Services (18/120) - AWS Backup: The Restore You Never Tested, a Locked Vault, and the Drill That Proves It
We will put a DynamoDB table under a tag-driven backup plan, lock the vault so nobody can delete the backups, restore into a fresh table, and compare the data. Plan for 45 to 60 minutes and well under a dollar.
- Tutorial
- AWS Services
- AWS Backup
SL#126 - Your Code Review Queue Has a Cliff, and AI Adoption Just Pushed You Toward the Edge
A vendor study of 22,000 developers shows throughput up and deployments down. A small simulation shows why, and what to measure before your review queue falls off the cliff.
- Software Engineering
SL#126 - Singleton: The Most Hated Pattern, Judged Fairly
Singleton gets blamed for sins that belong to global state. Here is what the pattern actually guarantees, the one thing it gets wrong, and where "exactly one" is still the right answer.
- design patterns
SL#125 - AWS Services (17/120) - AWS Storage Gateway: The Hybrid Bridge, a Working S3 File Gateway, and the Day You Should Use Something Else
We will deploy an S3 File Gateway on EC2, mount it over NFS, and watch a file written to a "disk" turn into an S3 object. Plan for 60 to 90 minutes and under a dollar.
- Storage Gateway
- Tutorial
- AWS Services