[security-dev] [PicketLink IDM] - File-based Identity Store

Stuart Douglas sdouglas at redhat.com
Sun Jan 20 20:53:28 EST 2013

Pedro Igor Silva wrote:
> Hi All,
>      Would like to know your opnion about how we're storing identity information using the File-based Identity Store and discuss possible alternatives.
>      Just for background, the motivation behind the file-based store is to provide a fast, ready-to-use and simple store, ideally for test and development scenarios/environments. The configuration is minimal and requires the file system.
>      Today we're basically serializing objects (JDK Serialization API) and storing them into files. The layout is quite simple:
>          /tmp/pl-idm/:
>          total 4
>          drwxrwxr-x. 2 pedroigor pedroigor 140 Jan 18 15:20 65d62693-953c-43a6-ac43-4b655174bbb4 ---->  Each Partitions has its own directory
>          -rw-rw-r--. 1 pedroigor pedroigor 554 Jan 18 15:20 pl-idm-partitions.db ---->  Serialized data for partitions
>          -rw-rw-r--. 1 pedroigor pedroigor    0 Jan 18 15:20 pl-idm-relationships.db ---->  Serialized data for Relationships
>          /tmp/pl-idm/65d62693-953c-43a6-ac43-4b655174bbb4: ---->  Partition directory.
>          total 8
>          -rw-rw-r--. 1 pedroigor pedroigor  789 Jan 18 15:20 pl-idm-agents.db ---->  Serialized data for Agents
>          -rw-rw-r--. 1 pedroigor pedroigor 1134 Jan 18 15:20 pl-idm-credentials.db ---->  Serialized data for Credentials
>          -rw-rw-r--. 1 pedroigor pedroigor    0 Jan 18 15:20 pl-idm-groups.db ---->  Serialized data for Groups
>          -rw-rw-r--. 1 pedroigor pedroigor    0 Jan 18 15:20 pl-idm-roles.db ---->  Serialized data for Roles
>       Serialization provides us a fast way to store data, but I have some concerns that I want to share:
>          - As we're serializing objects, we may have to ensure compatibility with prior versions. I think Version Control is a option here (btw, Stuart Douglas gave me some tips about that).
>          - Is better to use JBoss Marshalling instead of using the JDK Serialization API directly ? Mainly considering the JBoss ecosystem ?
>          - Is there a better format to store data ? Such as XML ...

I think that long term XML or a custom file format will give better 
results than just using JDK serialization to dump objects to disk, as it 
gives you much more flexibility in terms of versioning.

The flip side is of course that it does involve more work. A potential 
compromise could be to use serialization, but only to write primitive 
fields, for example instead of something like:

for(User user: users) {

Do something like:

out.write(1); //file version number
for(User user: users) {

This is still very quick to code, and does not reference any Picketlink 
classes in the serialized data.


>          - I had some discussions with Shane about using Infinispan. We agreed that the IDM cache will be ISPN-based, that is fine. But maybe a ISPN-based store can fits well too. ISPN allows to store data using different CacheStore implementations, transaction support, indexing, distributable or local storage, etc.
>       Wdyt ?
> Regards,
> Pedro Igor
> _______________________________________________
> security-dev mailing list
> security-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/security-dev

More information about the security-dev mailing list