...
- usecase: need to exchange info about "index of chunks managed by the worker" between cmsd and xrootd
- considering shared memory (if we can keep it simple)
- ~1000 elements, easy, can lock entire block, auto release lock
- don't want to do it through xrootd directly, or via database. Messes up scheduling I/O heavy requests
- having it will allow incremental updates to the structure of the index of chunks without restarting the services
- create stories for Andy
- created DM-2507, DM-2508
Deleting tables
- related story: DM-1896
- worry about race condition: one czar marks table delete-pending, other czar schedules query related to that table
- decided to introduce small delay, rely on fast synchronization between czars and CSS
- considered stamping schema with "schema version", incremented when table deleted etc. Worry, it can stop perfectly valid queries that need tables not affected
- considered cross-czar communication
...