http://blog.christianposta.com/istio-workshop/slides/#/agenda

Intro

microservices architecture shifts complexity from the space of code design and implementation into system operations
pushing the complexity from one place (software design and development) to another (operations) is just like hiding some tricky parts in somewhere else
k8s lets you to automate things and solve the problems easily
istio gives the rest of the needed tools
use helm to automate things

Problems

Communication

Direct communication

client never calls the service
use API gateway
handle auth, etc

Slow communication

use cache for GET
fail fast when POST, PUT, DEL
envoy will be sidecar which can be configured somehow

Security (NEED TO TEST)

kubernetes has NetworkPolicy
whats does istio give for us?
ingress/egress

Improve reliability

Healthz

proper timeouts

Graceful start / shutdown

tear down connections in the good order (to avoid data loss)
idempotency
avoid draining connection pool -> graceful stop is important

Circuit breaker & rate limitation (NEED TO TEST)

limit the impact of failures, latency spikes, and other undesirable network effects

Change management

services can come and go
rapid provisioning of new resources
should be able to deploy a new service quickly
automate everything (repo, ci, deployment)
use helm to pack/deploy apps

Deployment strategies

k8s deployment .spec.strategy.type==Recreate will kill all pods before new ones are created
0 downtime deployment - .spec.strategy.type==RollingUpdate controls the process with maxUnavailable and maxSurge
blue/green
canary (monitoring needed, allows to test in production, dark launch -> flippers)
use versioned deployments with istio request routing

Monitoring (NEED TO TEST)

loosely-coupled services collaborating
detect serious problems
fail and detect problems fast -> write test where it is necessary
prometheus
HPA + heapster

Fault injection

test resilience of your application
network latency -> delay
crash failure -> abort