-
Type:
Task
-
Status: Complete
-
Priority:
Medium
-
Resolution: Done
-
Affects Version/s: None
-
Fix Version/s: 1.6.79
-
Component/s: None
-
Labels:None
-
Epic Link:
-
Sprint:Ev 18.22, Ev 18.23
-
Documentation Impact:Unset
As of now, if master primary crashed (stopped) and re-started, it will get last_pp_seq_no from other Nodes and continue correct ordering.
For backup primaries this information is not propagated, so backup primary will start sending PRE-PREPARES from 1, and other nodes will reject it. It may eventually cause a memory leak, since ordering on this backup is stopped for some time.
Acceptance criteria
- Create a new key-value DB for persisting current states
- Persist last pp_seq_no for each Instance in the pool
- Use persisted last_pp_seq_no as a new pp_seq_no for a Replica if it's a primary in one of the Instances (including Master?)