gaojun2048 commented on issue #3416: URL: https://github.com/apache/incubator-dolphinscheduler/issues/3416#issuecomment-698852848
> > > > **Resource files are no longer stored on tenants, but are stored by project** > > > > Tenants is necessary . because every user run task, upload files to hdfs must use a hadoop user to do this. we know every file in hdfs must ower by a hadoop user(tenants). If one user can can multiple tenants, when he upload resource file to hdfs, which hadoop user(tenants) is the file ower? Do we need to select tenants when saving resource files? > > > > > > > > > I'm talking about weakening tenants, not deleting them > > > Now the files are stored in HDFS, which is divided by tenant > > > In fact, if divided by project, it can also meet the needs > > > Now the implementation of the middle platform system, resources storage by project is more in line with the needs of tenants > > > The tenant is only responsible for the execution rights of tasks, not for storage > > > > > > If users and tenants are one-to-one, resource files can be organized by tenants and by projects. Of course, it is best to organize by projects. > > If the user and the tenant are not one-to-one, the user must choose which tenant to upload when uploading files to hdfs. > > So I think user and the tenant are one-to-one is necessary. Based on this premise, the organization of resource files by project is the best. > > What if a user wants to have the rights of multiple tenants?? > > For example, execute Spark Program with spark tenant > > Execute hive program with hive user > > The root user is required to execute the system program > > Today's programs are stored on tenants > > Cross tenant execution procedures, resource files have to be authorized separately > > If there are 100 projects, resource files refer to each other > > How to manage resource files? > > If the executive authority of the management program is ＂??? > > Looking forward to the development of dispatching... After talking, we found that the key question is whether the resource file needs to have permission, whether the resource file in a project is shared by all users, whether the resource file between different projects is shared, if so, then we can use a tenant (this tenant is not associated with any users and is only used by ds to manage resource files) to organize resource files by project, and all users have common permissions to these resource files. Then the tenant associated with the user is only used to execute the task, so that the user and the tenant can be many-to-many, and the user can select the tenant and the yarn queue in the tenant to run the task. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
