Beyond multitenancy
container-based application factory
Damien Metzler
dmetzler@nuxeo.com
https://github.com/dmetzler/
Thierry Delprat
tdelprat@nuxeo.com
https://github.com/tiry/
About This talK
This is our technical adventure
- multi-tenancy with advanced per-tenant customization
-
build a Software factory using VMs, Docker and now CaaS
This is our story but, hopefully, this should also make sense for
- applications that are modular and configurable
- applications moving to the cloud / SaaS model
a long road
2012/2013
2014
now
initial idea and first tentative
first viable version
current development
and future architecture
2012/13 - IdeA & first try
Multi-Tenants, Cloud & Application Factory
Some Context
Nuxeo
we provide a Platform that developers can use to
build highly customized Content Applications
we provide components, and the tools to assemble
everything we do is open source
our customers are people building software
in-house -
software vendors -
SaaS
Nuxeo Platform
Repository
Services
Workflows, Conversions, Diff, Notifications, Activity ...
Building ON TOP of a Platform
Building ON TOP of a Platform
Building ON TOP of a Platform
Building ON TOP of a Platform
Building ON TOP of a Platform
Project Management
Building ON TOP of a Platform
Digital Assets Management
Building ON TOP of a Platform
Case Management
Building ON TOP of a Platform
Collaboration
Building ON TOP of a Platform
Customer
1 Customer
Custom deployment
More Customers
Industrialization
Customers
PLM Solution
Customer/Tenant 1
Customer/Tenant 2
Styles
Screen Layouts
Meta-Data
Workflows
Business Logics
Connectors
Styles
Screen Layouts
Meta-Data
Workflows
Business Logics
Connectors
APPLICATION LEVEL MULTI-TENANTS
Document Store
Security
Life Cycle
Indexing
Versioning
all clients share the same application
application manages data and configuration partitionning
Application Level Multi-Tenants
Shallow isolation
- quota management is not efficient
-
customization options are limited
Monolithic
- same version, same component set
-
same upgrade and maintenance policy
Not even simple
- scale out is not that easy (i.e. move a tenant)
- per-tenant Backup/Restore is not easy
-
Heterogeneous deployment units
VM level / JVM level / App level
Can not leverage OSGi / Extension Point model
Not "Cloud Native approach"
Container Level Multi-tenants
rely on infrastructure to provide tenants isolation
application does not need to be impacted
Flexible
Unlimited Customization
Full Isolation & Quotas
Application Factory
Create "on demand" application for each customer
- use Container level isolation
- provision infrastructure from the Cloud
- custom assembly for each customer
Application Factory
Create "on demand" application for each customer
- use Container level isolation
- provision infrastructure from the Cloud
- custom assembly for each customer
Application Factory
Create "on demand" application for each customer
- use Container level isolation
- provision infrastructure from the Cloud
- custom assembly for each customer
Application Factory
Create "on demand" application for each customer
- use Container level isolation
- provision infrastructure from the Cloud
- custom assembly for each customer
Application Factory
Create "on demand" application for each customer
- use Container level isolation
- provision infrastructure from the Cloud
- custom assembly for each customer
Let's do this ...
... for our own use cases !
Nuxeo Trials Use Case
We build a platform and customization tools
We want people to be able to taste the full experience
- choose their components
- configure business rules
- run the app they build
Online demo site
with full per-tenant customization
PaaS vs Automation
Early testing with CloudFoundry / OpenShift
- Fast moving ground
- Very opinionated
- Our app does not fit in Java PaaS
Go with Deployment Automation on IaaS
- Seems easier
- Better match for sys-admin / devops
Some Challenges
Prospects do not pay
Prospects want to access all features
We want them to have a great experience
free customers with high expectations !
Cost is a major concern
Optimize hosting infrastructure
Cutting Costs
EC2 Spot instances
Start/Stop instances when needed
Leverage AWS Services
Early Results
A lot of moving parts
- no clear deployment unit
-
lot of scripting
Slow
- provisioning is slow
- start/stop too slow to be usable
works but ...
DEAD END ?
We need lighter foot print
We need faster startup
Hope Comes from
Some developers have started playing
with lightweight containers because
VMs are too fat & slow !
and Docker !
the developers
[dev] Cool new stuff on cloud related techs
...
http://www.docker.io/ :
a command line tool for launching and managing
arbitrary processes using LXC.
Open Source project contributed by the dotcloud guys.
...
03/2013
2014 - Nuxeo.io v1
Build the Application Factory on Docker
Arken
First target is to power Nuxeo Trials
need a smooth UX
but also build a generic infrastructure
publish work as opensource
Assign a team of java developers
not system-administrators
Expect results in 3 months !
rebuild / refactor every 3 months
Build Your Own Application
Build Your Own Application
Select
target Platform
Build Your Own Application
Pick additional components
Build Your Own Application
Build Your Own Application
Build new components
Build Your Own Application
Select custom Components
Build Your Own Application
Define Application Model
Build Your Own Application
Build Your Own Application
Choose deployment
environment
Build Your Own Application
Deploy & Run !
Make your own container recipe
- choose your ingredients
- assemble
- run
Available to the web user
Build Your Own Application
Nuxeo.io Factory
Principles
Docker containers !
Leverage AWS infrastructure
Principles
Passivation
- Less than 5% of trials are active at a given time
- Fast start/stop (no data)
- High density hosting on AWS
Principles
Dual state orchestration
- Expected vs Actual
-
Decoupling & Monitoring
Use a distributed registry
CASTING
Containers infrastructure:
Docker + CoreOS
Scheduler:
Fleet
Distributed registry:
etcd
Monitoring:
DataDog
CASTING
CASTING
CASTING
The Containers:
Nuxeo Servers
Manager.io (Nuxeo +AngularJS)
Passivator (Go Service)
ArkenCtl (Go Cmd)
Dynamic reverse proxy Gogeta
(1 per tenant)
Big Picture
Big Picture
Big Picture
Big Picture
Deploy new tenant
Big Picture
Register new tenant
Big Picture
Deploy more tenants
Big Picture
Route request to Customer X
Big Picture
Route request to Customer X
PASSIVATE
ACTIVATE
ACTIVATE
ACTIVATE
ACTIVATE
started
Monitoring
Some Good Results
1000+ instances/month managed on 4 EC2 VMs (m3.2xlarge)
Production hosting for some Nuxeo based applications
Eventually stable
Docker and Go are really great
The vision is good !
All Good ?
Almost !
Miseries
- CoreOS updates
- btrfs fragmentation and IO issues
- etcd stability
- fleet "shortest path" scheduling
- ...
Lots of moving parts + Lots of young solutions
Experience a new type of failure every day !
Structural Issues
Lot of boring glue code
- Networking, Port Mapping, provisionning ...
-
developers are sick of shell scripts
Storage management is an issue
- asymmetrical model & scalability
- Security concerns on shared infrastructure
CoMplexity
Troubleshooting is tricky
Command line tools
Still too scary for a customer
the system is complex to setup
now - Nuxeo.io v2
Leverage our experience and Docker evolutions
Drivers
Docker ecosystem evolved
Kubernetes, Swarm, Compose, Rancher ...
We believe in the initial vision is good
Customers starts to like the idea of Containers
We have learned a lot from Nuxeo.io v1
Time for a reboot
Additional contraints
Cluster configurations
1 node, 2 nodes, 7 nodes ...
Production hosting
take over Nuxeo-Cloud
Customer compliant
avoid or hide and package
the glue code
X
Remove as much code as we can
Leverage Docker evolution
Networking in Docker / Rancher / Kubernetes
- No more "manual" port & DNS mapping
- Use Software Defined Network
s/port-mapping/SDN/g
s/Ansible/Swarm/g
s/Scripting/Compose/g
Clustering in Docker-Swarm / Rancher / Kubernetes
- No more scripting automation
Stack templating in Docker-Compose / Rancher / Kubernetes
- Manage an application as a set of containers
Free from the Shell !
less glue code to write / debug
focus on application level
Leverage Docker evolution
Volumes in Docker
storage can now be provisionned as a container
containers can now be statefull
Can provision Storage nodes exactly as Processing nodes
Streamline architecture: everything is container
all tenant resources are provisioned the same way
CaaS ?
- Avoid building a CaaS infrastructure
- Focus on Application Customization & Templating
- Keep some flexibility on Cluster Orchestration
RANCHER
Caas Choice
Nice high level REST API
abstraction on CaaS provider
Swarm/Kubernetes/Cattle
Administration UI
- Infrastructure vs Application view
-
Monitoring and SSH access
Close to Docker
- Built on top of Docker and Compose
- Contribute to Docker
CaaS Choice
Application Management
-
Stack definition & templating
Provide additional features
- Health Monitoring
- Load Balancing and DNS
- Rolling upgrades
Volume plugins :
convoy: NFS / GlusterFS
Nuxeo.io with Rancher
Nuxeo.io with Rancher
Nuxeo.io with Rancher
Nuxeo.io with Rancher
Nuxeo.io with Rancher
Nuxeo.io with Rancher
Nuxeo.io with Rancher - DEMO
Direct Gains
Rancher & Docker
do the heavy lifting
Deploying Arken is easy
One unique API
to deploy new tenants
Arken Contents
Application Templating
Package Selection : Wizard + Config + Docker File
Deployment template : Compose + Rancher
Passivation Management
Passivation aware Routing
State management
API & Adapter
Independent
Go Based
OpenSource
multi-tenants ?
Application Level Multi-Tenancy
- data only isolation
Container Level Multi-Tenancy - Nuxeo.io v1
- processing isolation - shared data sources
Container Level Multi-Tenancy - Nuxeo.io v2
- complete data and process isolation
multi-tenants ?
Tenants sharing an Application
Tenants sharing Infrastructure
+ dedicated applications
processing isolation : docker containers
- reusable execution units
data isolation : docker and volumes
- stateful containers
application configuration: Rancher + Arken + Nuxeo
- software defined tenant
Application Factory: Rancher + Arken
Breaking the Monolith ?
Stack of OSGi Services = MicroServices
Break deployment by tenants
Break deployment by containers
Deploy different service stacks Different containers
Converters, Async Processing, IDE , Account management ...
Leverage Arken architecture
Introduce new granularity level
Any Questions ?
Thank You !
http://www.nuxeo.com/careers/
We are hiring !
New York, Paris, Lisboa
Beyond Multitenancy
By Thierry Delprat
Beyond Multitenancy
O'Reilly Software Architecture 2016
- 5,490