Slide 1

Slide 1 text

Effective DevOps Team Topologies Or, how to scale people @hashfyre Sportskeeda, Tech Talks Unconference - Apr 5, 2019

Slide 2

Slide 2 text

Devs and Ops Silos ● Throw-it-over-the-wall ● DONE ○ means feature complete ○ May crash in Prod ● Dev-local env is miles from Prod ● Ops don’t think they need to get involved early Devs Ops

Slide 3

Slide 3 text

DevOps Team Silo ● Dev: “We need to get DevOps people” ● Dev: “They’ll magically solve cloud and containers for us” ● Ops: “Don’t dare commit to my terraform repo and break aws” ● Makes sense in 12-18 mo, then needs to be disassembled Devs Ops DevOps

Slide 4

Slide 4 text

Dev Don’t need Ops ● Dev: “I wrote terraform; I’m fullstack.” ● Ops: “Plan fails on prod, also can’t we have this CICD’ed?” ● Dev: Walks away distracted ● Might need to evolve into Ops-as-IaaS or DevOps-as-a-Service Devs Ops DevOps

Slide 5

Slide 5 text

Rebranded SysAdmin ● Dev: “We need someone who knows linux!” ● DevOps: “But K8S is not just linux” ● Dev: “Linux all the things!” ● Ops: “I got a promotion!” Devs Ops DevOps

Slide 6

Slide 6 text

DevOps as Toolchain Team ● Dev: “I’m going to build a PubSub on Kafka” ● Ops: “But… NATS, EMQTT, RabbitMQ!” ● Dev: “We will write our own API Gateway” ● Ops: “But… Kong!” ● Who owns it when things perform half as bad? ● Early Ops involvement, we need infra-spec-reviews, please! Devs Ops DevOps

Slide 7

Slide 7 text

Embedded Ops ● Narrow app scoped operations ● AppOps teams conflict with each other on prod ● Duplicated effort due absence of high-level unified solutions ● Resource constraints, re-prioritizations, burnouts ● “We have our own prometheus and ELK” Devs Ops DevOps

Slide 8

Slide 8 text

Devs - Ops Collab ● Zion! ● Complementary specializations! ● Dev: “What Datastore do I use for optimal Document Search at projected scale?” ● Ops: “Hey, this method seems to have a memory-leak and no logging, let sit down and fix.” ● Ops: “Let’s implement tracing together!” Devs Ops

Slide 9

Slide 9 text

Shared-Ops ● Fully integrated Ops into Product teams ● Complete overlap ● Dev: “Hey, remember that bug in the prometheus-operator? I fixed it in upstream!” ● Ops: “Hey, I noticed a small method with a query doing cartesian joins, I created an index and re-wrote the query for you.” Devs Ops

Slide 10

Slide 10 text

Ops-as-IaaS ● Dev: “I need on-demand infra” ● DevOps: “I’ll create a self-serve platform API wrapper on top of AWS for that.” ● Ops: “We’ll write automation on top of the custom IaaS APIs and handle training, knowledge-sharing.” Devs Ops DevOps

Slide 11

Slide 11 text

DevOps with Expiry ● Starts from a DevOps Team Silo ● Gradually fuse the teams together. ● Reduce only to knowledge translation and training and then make them vanish! Devs Ops DevOps

Slide 12

Slide 12 text

DevOps Evangelists ● For orgs with large gaps. ● You need well-respected Devs and Ops architects in this team. ● Change leadership. Devs Ops DevOps

Slide 13

Slide 13 text

Google Model ● Devs prove code is good to go with tests and benchmarks ● DevOps helps on all envs but prod, does lifecycle management ● SRE only allows code that satisfy SLO to prod Devs Ops DevOps SRE

Slide 14

Slide 14 text

Bibliography Team Topologies, by ● Matthew Skelton - @matthewpskelton ● Manuel Pais - @manupaisable

Slide 15

Slide 15 text

Fin. @hashfyre on Twitter, github Ops-as-IaaS/SRE’esque person @Hotstar