Refactored leveled_sst gen_fsm into gen_statem #401
Conversation
but this function is deprecated... put in for backward compatibility
There was a problem hiding this comment.
There are some code coverage issues. I'm loathe to give up on 100% code coverage, as I'm worried about missing a regression in code coverage if we tolerate anything less. There is some code here which I think can never be triggered, but also one case (update_blockindex_cache being cast in a race in a switch to delete_pending) which probably needs a test to get coverage.
The casting of timeout sin leveled_sst I think will end when we merge in the latest fix from develop-3.0
| cdb_clerkcomplete(Pid) -> | ||
| gen_fsm:send_all_state_event(Pid, clerk_complete). | ||
| gen_statem:cast(Pid, clerk_complete). | ||
|
|
There was a problem hiding this comment.
The clerk_complete message can only be sent when in the reader state. It is send only by the leveled_iclerk, and the leveled_iclerk only examines files in this state. The state cannot be transitioned to delete_pending without the leveled_iclerk first having called clerk_complete - so there should be no race. So handling this message should be done in the reader state only.
There is an issue otherwise with code coverage, in that there will no longer be 100% code coverage (and so spotting regression of code coverage will be hard).
| {keep_state_and_data, [hibernate]}; | ||
| starting(info, _Msg, _State) -> | ||
| {keep_state_and_data, []}. | ||
|
|
There was a problem hiding this comment.
If there are no info messages supported should we remove handling of them. Again code coverage being the issue.
| end. | ||
| end; | ||
| starting(cast, {update_blockindex_cache, BIC}, State) -> | ||
| handle_update_blockindex_cache(BIC, State); | ||
|
|
There was a problem hiding this comment.
This message cannot be received in the startup state, as it is prompted from get_slots which can only occur in the reader state
| {stop, normal, State}; | ||
|
|
||
| delete_pending(cast, {update_blockindex_cache, BIC}, State) -> | ||
| handle_update_blockindex_cache(BIC, State); |
There was a problem hiding this comment.
This is interesting. get_slots may be prompted which results in update_blockindex_cache but a set_for_delete message may happen in-between. Hence we do need support for update_blockindex_cache in delete_pending state - but this is never triggered in test, so we perhaps need a unit test to explicitly trigger it.
|
Replaced by #404 |
No description provided.