@antweiss
● Misconfigured Fallback Options
● Retry Storms
● Outages When a Downstream is Overloaded
● Cascading Failures When a SPOF crashes
The Traps of Distributed SysteMS
@antweiss
The BirdWatch Loop
apiVersion: apiextensions.k8s.io/v1beta1
kind: CustomResourceDefinition
metadata:
name: birdwatches.otomato.link
WAIT
FOR
CANARY
Found Canary:
Service: aleph
Version: 0.4
Release
some
traffic to
the canary
Canary
Healthy?
Canary Flies to
Production
yes no
Canary Dies
https://github.com/otomato-gh/birdwatch-otomator
@antweiss
SOME HARD QuestIONS:
● Won’t DevOps get Stuck in the Mesh?
● Who’s Gonna Build the Otomators?
● Why Python and not Golang?
@antweiss
SUMMARIZE:
● Service Mesh Protects Us From the Traps
● Solves the Shared Libraries Issues
● A Good Fit if you like Centralised Control
● Smarter Pipelines -> Dumber Services
● Without Otomators The Mesh Can Get Clogged
● Dumb Services in a Clogged Mesh = Catastrophe
● Therefore Real Ninjas use Otomators
@antweiss
The Future:
● Alternatives:
○ Linkerd 2.0 (from Bouyant)
○ Consul Connect (Hashicorp)
● Commercial Options
○ Aspen Mesh (F5 incubator)
○ AWS, Azure, GCP
● SMI Spec - the Standard Service Mesh API