[CDCSDK] Restarting connector task can lose track of tablet list in cases of tablet split

[adithya-kb]

Connector model today is to divide the list of tablets to tasks and then let the tasks do processing. However, the list which is received by these tasks is immutable, so let’s say we have a task restart, it will again try to read the same list.

Task 0: A, B, C
Task 1: D, E

Consider the following example:

1. Task 0 receives the tablet list as A, B, C
2. Tablet A splits down the line into A_1, A_2
3. Task 0 is restarted
4. Upon restart, it will read the original immutable list again (A, B, C) which is the stale list now
5. At this point, there will be an error since the connector doesn’t know that it needs to find the children (or possibly grandchildren) tablets

Jira Link: DB-4992