Building Multi Tenant Java Applications

Rajesh Venkatesan Senior Architect, HCL Technologies [email protected]

Multi Tenancy An Overview

Time Share ASP End User Web Apps

What?

Ability to cater to multiple customers using a shared instance of Software/Hardware

When?

Why?

How?
Thats what this session is about

Inability of SOHO and SMB segments to adopt IT Non IT Businesses getting entrenched in managing IT

Multi Tenancy Impact in the real world

Shared Infrastructure Lower Cost Higher Complexity of Construction Dedicated Infrastructure Higher Cost Tailor Made Construction relatively easy Heavy Customization Support Non Standard Requirements Configuration over Customization Driving Standardization

Single Vs MultiTenancy
Higher Scale Lower Cost

Shared Vulnerability

Shared Upgrades
Scalability is bound to target customer size Minimized Vulnerability

Customized Upgrades

Architectural Facets of Multi Tenancy in the Software World

Virtualized Hardware Database Application Servers Inbound Outbound

Shared Infrastructure

Integration

Configuration over Customization

Standardization of UI Data Model Business Logic

Security

Data Security Application Security

Shared Infrastructure Database

Typically Multi Tenancy at the database level has 3 standard patterns Separate Database
Traditional Isolated Database Instance Per Customer

Shared Database Separate Schema

Customers get their own schema but are co-hosted in the same database

Shared Database Shared Schema

Drives the highest efficiency. All Customers data is stored in the same database and schema with a tenant id qualifier

Isolated

Shared

Isolated

Separate DB

Separate Schema

Shared Schema

Shared

Source: Multi Tenant Data Architecture, Frederick Chong, Gianpaolo Carraro, and Roger Wolter Microsoft Corporation

Database Multi Tenancy Patterns Pros and Cons

Separate Database
Easier to Maintain Allows Customization Higher Security Easy DB Upgrades High Cost to Customer

Shared Database Separate Schema

Easier to Maintain Allows Customization Relatively Higher Security Slightly Complex DB Upgrades Average Cost to Customer

Trade Off Considerations

Compliance/Regulatory Cost Operations Time to Market Liability

Shared Database Shared Schema

Lowest Cost

Complex Upgrade Process Availability impacts multiple customers Data Security delegated to application layer

Database Multi Tenancy Implementation

Isolated Database and Shared Database Separate Schema

Standard Data Access simply returns the appropriate connection based on tenant context
From a JDBC Perspective this implies different connection strings based on the customer. Typical Tenant Context is set by an intercepting filter and obtained at the DAO layer possibly via a ThreadLocal variable

Shared Database Shared Schema

Approach 1 Business Logic and Data Access is aware of multi tenant context and therefore query appropriately Pros Easy to build Cons High Probability of bugs leading to data leakage Approach 2 Abstract Multi Tenancy concern to the Data Access Layer and write business logic without tenant context. Data Access Layer automatically adds tenant context to all data calls

For Hibernate implement a Tenant aware ConnectionProvider and switch off the second level cache.

For Hibernate Use Filters Use Hibernate Shards

Integration

Typical integration concerns when applications move out of customer premises include

How can I receive notification

Is there standard integration

How can I push data to the application

How do I orchestrate my business process

Familiar? SOA?
8

Integration Contd

Fundamentally the application must support well defined interfaces for inbound Integration as well as Outbound Integration Inbound Integration Expose services Technology Independent Standards Based High Security Multi Tenant Aware
WSS4J Axis, XFire

Implementation SOAP Well Defined Standard WSS for Multi Tenant Security (Username/Token, X509 Tenant Certificate, SAML, Kerberos) REST Easy Integration Simplicity Security to be built on top.

JAX-WS

Integration Contd

Outbound Integration Allow Tenants to register for integration events. Push Vs Pull Push Synchronous Data can pushed to waiting WS endpoints Publish Standard Web Service Interfaces that customers can implement. Multi Tenant aware integration layer appropriately calls out the tenant specific interface. Problem with availability of customer endpoints Push Asynchronous Expose Secure Asynchronous Messaging Infrastructure. Heavy Vs Light Weight Events For security reasons and other reasons, push non-critical information alone into the message. The listening party then calls back via standard web service inbound interface for the actual message. Push the entire message with all relevant information. The Infrastructure is absolutely secure. The messaging infrastructure takes responsibility of ensuring delivery.

10

Security

Facets of Security

Physical Security

Security
Application Security Data Security

11

Data Security

JCA/JCE Use tenant specific encryption when required. Decouple encryption awareness from the data layer allowing data leaks to still be harmless
TradeOffs

Data at Rest

Database functions cannot be applied on encrypted fields Performance

Tokenization of Data Only a token reference is stored in the database. Actual data has to come from a high security data protection server

Data in Transit

Use Secure means of transfer (https) and add authentication/ authorization layer on top. Use In Wire Encryption for highly critical data JSSE

12

Application Security

Application Security is not different from traditional applications but some aspects become a lot more critical. Exposing the application on the web brings about a gamut of application security threats. Be Aware of possible security vulnerabilities and address them. The OWASP Top Ten Project (http://www.OWASP.org) is a good place to look. A1: Injection

A2: Cross-Site Scripting (XSS)

A3: Broken Authentication and Session Management A4: Insecure Direct Object References A5: Cross-Site Request Forgery (CSRF) A6: Security Misconfiguration

A7: Insecure Cryptographic Storage

A8: Failure to Restrict URL Access A9: Insufficient Transport Layer Protection A10: Unvalidated Redirects and Forwards

13

Application Security Contd

Some of the security best practices for applications Encrypt all communication between the browser and server via SSL. Strong password policy enforcement using configurable password policy. Passwords are stored after one way encryption in the database. It is impossible to know user passwords. Auto-Generated Passwords automatically expire after xx hours. Use of token based authentication with zero trust on server side Sessions. All access to the application is authenticated and is either secured by an authentication token or via certificates. Decoupled Authentication and Authorization and consolidation of concerns in order to establish a single point of control of user access. RBAC ensuring there are no super-users who get access to the system. Extensive Logging Capability ensuring every action is traceable to the user, request and session along with the actual change to the database. Database credentials created with named permissions. OS credentials created with named permissions All Inbound and Outbound interface points must be secured by default. (SSL) Additional Tenant Aware Security measures like Tenant Specific Certificates

14

Application Security Federated Identity

Tenant 1

Corporate LDAP

Multi Tenant Application Tenant n

Corporate LDAP

With applications moving outside of customer premise, corporate users are forced to have multiple identities one corporate and other in-cloud application identity. This poses a security problem for customers since a person moving out of the company still has access to corporate data. Therefore it becomes necessary to allow identity to be federated from the corporate context. Therefore the application has to be ready to De-Couple Identity Management and Authentication Support delegation of IdM and Authentication to corporate systems through established standards like SAML.

15

Configuration Over Customization

In order to drive efficiency, an application must standardize its features. However this results in not being able to accommodate customers with alternate business processes. This results in an architectural requirement: How to support customization via configuration?

Database

Allow extension of existing entities

Business Logic

Business Logic Templates Allow pluggable business logic. Allow small changes to business process Metadata driven UI

UI

Customize Look and Feel Layout Content

16

UI Customization

Depending on requirements UI customization is done at various depths Look and Feel The ability to change the font, color and style of existing UI Layout The ability to switch component layouts Content The ability to choose what content goes where. Two Approaches

Template/Skin Based Allow tenants to choose different themes and ability to write new themes is restricted but possible. Standard mechanism followed by most websites (Blogger, Wordpress, Liferay)

Complete Customization Allows tenants to customize the UI as per their requirements. How much they can customize is left to the application. Drag and Drop UI to Customize

Both approaches require a metadata layer that can understand the customization done be specific tenants. UI Rendering must take into account a standard layout as well as the metadata for rendering. Accommodate tenant specific UI Data models that can extensions to standard data models.
17

Business Logic & Database Customization

Business Process Customization

Enable an application to be flexible in allowing changes to business logic Allow different workflows to be configured per tenant. Reference: At the application design level Follow a highly de-coupled, pluggable component based design. Standard IoC Pattern to plug new implementations
Multi Tenant Data Architecture, Frederick Chong, Gianpaolo Carraro, and Roger Wolter Microsoft Corporation

Database Customization
Ability to extend the schema as per specific requirement In the Shared Database Separate Schema and Separate Database pattern, this becomes trivial as the customization can be done directly. In the Shared Database-Shared Schema, the following approaches are standard To have a pre-determined set of fields for specific data models that can be used as extensions. To have a generic extension schema that can accommodate customization to any entities and a data access and business logic layer that can bring in the tenant context when querying.

At the functional level Decide on the smaller variations that a business process/logic can take. Make these configurable. Allow ability to plugin newer processes as the application evolves. Accommodate generic data models during processing to cater to extended schemas Again a metadata layer is required to understand the configuration done by tenants at the business process level as well as newer business process that is available.

Spring

18

Scalability

Data

In case of a RDBMS, Shared Database Shared Schema use partitioning by tenantid (SHARD) Give a thought about NoSQL Databases if dealing with multiples of TB of data(ACID vs BASE)

Clustering Make services as stateless as possible. Session Replication is a nightmare. Avoid file system for data. Use a central datastore

Hadoop HBASE

Application Server

De-Coupled Components Conceptualize application features that can be de-coupled and scaled separately. Allows a resource hogging feature to be separated out and scale strategy planned differently. Cache data where possible (memory IS cheap) Plan for failure Auto Recovery.

With the current scope of browser capabilities (HTML5) pushing state to the browser has become easier.

UI

Also frameworks like GWT has enabled complex applications to sit on the client side. For applications using more sophisticated RIA clients (OpenLAZLO, FLEX or Silverlight), the same principle applies

19

Questions?

20