nexus/sreweekly/markdown/393/01-github-teivah-sre-roadmap-an-opinionated-roadmap-to-become-an-sre-conc.md

# GitHub – teivah/sre-roadmap: An Opinionated Roadmap to Become an SRE (Concepts > Tools)

- **期号**: SRE Weekly Issue #393（2023-10-08）
- **作者**: Teiva Harsanyi
- **链接**: https://github.com/teivah/sre-roadmap

## 简介

This repo contains a path to learn SRE, in the form of a list of concepts to familiarize oneself with.

## 正文

An opinionated roadmap to become an SRE (Concepts > Tools)

- *Concepts*  - Fallacies of distributed computing
  - Synchronous vs. asynchronous
  - Event log vs. message queue
  - Exactly-once delivery
  - Different types of message failure
  - Orchestration vs. choreography
  - Causality
  - CDN
  - *Hashing*    - Consistent hashing
    - Geohashing
    - Perfect hashing
  - Read-heavy vs. write-heavy impacts
  - Federation
  - *Latency*    - Latency, throughput, goodput
    - Latency numbers every programmer should know
    - How to prevent latency variability
    - Tail latency
  - How to reduce sharing
  - Idempotency
  - *Load balancer*    - Concepts
    - Layer 4 vs. layer 7 load balancer
  - Liveness vs. safety properties
  - Microservices: pros and cons
  - REST
  - gRPC
  - Service mesh
  - Source of truth
  - Stateful vs. stateless
  - Total vs. partial order
  - Why can't we rely on the system clock in distributed systems
  - Vector clock
- *Cache*  - When to use a cache
  - Cache-aside vs. read-through
  - Eviction policy
  - Refresh-ahead
  - Write-through vs. write-back
  - Distributed cache
  - Performance cache vs. capacity cache
- *Databases*  - *Different types of databases*    - NoSQL vs. SQL databases
    - Relational vs. document
    - Column-oriented databases
    - Graph databases
    - Vector database
    - Objects-based storage
  - ACID
  - *Partitioning*    - Criteria
    - Methods
    - Replication vs. partition
  - Hotspot
  - CALM theorem
  - CAP theorem
  - PACELC theorem
  - Cardinality
  - Chain replication
  - Consensus
  - Concurrency control
  - Consistency models
  - Isolation levels
  - Serializability
  - Linearizability
  - CRDT
  - *Indexes*    - Tradeoff
    - Primary vs. secondary indexes
  - Denormalization
  - View & materialized view
  - Transaction
  - Distributed transactions downsides
  - Strategies to handle rebalancing
  - Leader election
  - MVCC
  - N+1 select problem
  - Quorum
  - Raft
  - Read repair
  - Single-leader, multi-leader, leaderless replication
  - Split-brain
  - 2PC
  - 3PC
  - WAL
  - Write and read amplification
- *Data structure*  - *Probabilistic data structures*    - Bloom filter
    - Count-min sketch
    - HyperLogLog
  - *Storage*    - LSM tree
    - B-tree
    - SSTable

- *Concepts*  - Difference between availability, resiliency, robustness, fault-tolerance, and reliability
  - Why is it wrong to target 100% availability
  - Blast radius
  - Failure domain
  - Cascading failures
  - Hard vs. soft dependencies
  - *Scalability*    - Concepts
    - Knee point
    - Ceiling
  - Number one source of outages
  - Tail tolerance
  - Toil
- *Patterns/Anti-patterns*  - Bulkhead pattern
  - Circuit breaker
  - Exponential backoff
  - Jitter
  - Graceful degradation
  - Load shedding
  - Retry amplification
  - Backpressure
  - Rate limiting
  - Request hedging
- *Practices*  - Chaos engineering

- *Concepts*  - What's the difference between monitoring and observability
  - Trace vs. metric vs. log
  - Golden signals
  - Observer effect
  - Percentile
  - Streetlight anti-method
  - Time-series based monitoring lies
  - USE method
  - Main metrics for cache
  - Why should we be careful about average performance metrics
- *Alerting*  - Alerting strategy
  - Alerting fatigue concept
  - Characteristic of a good alert
  - Slow vs. fast burn alert

- *Concepts*  - Bake time
  - Feature flag
  - Feature freeze
  - Rollout supervision
- *Rollout types*  - Blue green rollout
  - Canary rollout
  - Progressive rollout
  - Shadow rollout

- *Concepts*  - SLI vs. SLO vs. SLA
  - Error budget
- *SLO*  - Difference between KPIs and SLOs
  - Benefits of having alerts based on SLOs
  - Why is exceeding an SLO not necessarily a good thing
  - SLO for data (freshness, completeness, consistency, etc.)
  - SLO for mobiles
  - SLO for services

- Container
- Container orchestration

- Scripting
- Filesystem
- Memory
- Processes
- Resource utilization
- Network

- ARP protocol
- Bandwidth
- BGP
- CoDel
- CORS
- DNS
- Ping vs. heartbeat
- *TCP*  - TCP vs. UDP
  - Congestion control
  - Connection backlog
  - Flow control
  - Handshake
- HTTP
- HTTP/2
- Head of line blocking
- Health checks: passive vs. active
- Internet model
- NTP
- OSI model
- Routers
- Switch
- Network topologies
- What happens if you type google.com in your browser

- Authentication
- Certificate
- Certificate authority
- Cipher
- Confidentiality
- Encryption
- TLS
- PKI
- Signature

- Core analysis loop
- Correlation vs. causation
- First principle
- Five whys technique
- *Incident management*  - How to address an incident (assess, mitigate, resolve)
  - Incident roles
  - How to write a postmortem
  - 3C principles (Coordinate, Communicate, maintain Control)

- SRE role
- Version control

- *Communication*  - Writing
  - Oral
  - Presentation
  - The XY problem
- Collaboration
- Problem solving
- Curiosity
- Navigating ambiguity
- Staying humble