Server Changes in Fluid Server Cluster Discovery Feature

[tianzhu007]

To implement server discovery, we need to add the following changes to the server side:

1. New interface (ISession): Used to store the Fluid server cluster’s URLs and session alive flag. This session information will be returned by the discovery call in create session and join session flow.
2. Database changes:
   a. Add the session information to the document metadata table on container creation.
   b. When the session ends, we will mark the isSessionAlive flag as false in session information in the deli lambda.
3. New API to get session information: A GET API called during the join flow to return the correct session information to the client. This api will return the session information if there is an existing session and will create a session and update the session information and then return the session details if there is no alive session.
4. We need to disable and update the cache, the first time the client creates the new session. Otherwise, the client might read the outdated summary from a cache in the old cluster. For example:
      - Client A opens a document in the US, generates some ops and closes the session. The final summary with the sequence # 20 is cached.
      - Client B then opens the same document and creates a session in the EU, generates some ops, and closes the session. The final summary with the sequence # 30 is cached in EU.
      - Client A reopens the document from the US and a new session is created. If we do not disable the cache, client A reads the summary with the summary sequence number of 20 which would be incorrect. Therefore, we need to let client A read the summary directly from the storage, instead of the cache, and update the cache with the fetched summary (sequence number 30).

[hedasilv]

Hey @tianzhu007, thanks for working on this PR! I just left a few follow-up comments, the most important one related to our discussion regarding caching the summary in historian.

Also, curious: what was your testing strategy after you pushed the changes to address comments? Just want to make sure we test this thoroughly as this is an important change.

Finally, it would be good to get @pradeepvairamani and/or @znewton sign off besides mine before merging just in case I missed something 😄

[tianzhu007]

Hey @tianzhu007, thanks for working on this PR! I just left a few follow-up comments, the most important one related to our discussion regarding caching the summary in historian.

Also, curious: what was your testing strategy after you pushed the changes to address comments? Just want to make sure we test this thoroughly as this is an important change.

Finally, it would be good to get @pradeepvairamani and/or @znewton sign off besides mine before merging just in case I missed something 😄

Thanks for reviewing it. :)
I used the driver with/without the discovery feature test the server-side change locally.
Afterward, I will deploy both r11s, and historian images to one of the dev clusters, and test it by both drivers with/without the discovery feature.