HomeStore
HomeStore copied to clipboard
Storage Engine for block and key/value stores.
Each replication request is tracked as repl_req structure. In followers they are created upon leader sending data either over raft or data channel (whichever comes first). There is a possibility...
With the revised Device Manager, we support partial boot up, as in even if one of the drives are unreadable or some regions are unreadable, we can still recover other...
ReplDev doesn't distinguish whether alloc writes are from same shard or not. From HomeStore we want to parallel those allocation on different chunks, and only sequentialize for allocation/writes on same...
At present crash recovery of root node split and merge doesn't work correctly, because of presence of another node pointing to root node not persisted. We need to fix that.
@xiaoxichen reported this issue while testing end to end flow. Possibly because bits avail isn't calculated correctly.
We need a generic mechanism to simulate crash and subsequent recovery for all unit tests to run. Without this, we are relying on actual crashing and then restart with some...