Mas 3.0 dialyzer #800
Mas 3.0 dialyzer #800
Conversation
Work through dialyzer issues. Should just be one unresolved issue left
Add logging of socket
Otherwise if the last partition is a soft_retry_limit failure will never report as complete with errors
maybe_complete_fullsync function requires further mocking
Dialyzer has an error if we call terminate from system_terminate - appears to be related to the exit/1 funciton within terminate. This change brings a new function to log and return the reason for failure (previously part of terminate), so that the exit can happen in the parent function without causing a big DRY issue
| {noreply, State4#state{error_exits = ErrorExits1, | ||
| purgatory = Purgatory, | ||
| dropped = Dropped}}; | ||
| maybe_complete_fullsync(Running, |
There was a problem hiding this comment.
This is more than just a refactoring...
There was a problem hiding this comment.
You are right, I was a bit lazy combining the two. Shall I revert out the commits associated with #799 and put in a separate PR?
There was a problem hiding this comment.
Thanks for the reference. This is fine.
| @@ -47,7 +47,7 @@ | |||
| cluster, | |||
| fullsync_worker, | |||
| work_dir = undefined, | |||
| strategy :: keylist | aae, | |||
| strategy :: keylist | aae | undefined, | |||
There was a problem hiding this comment.
The documentation says:
%% If the protocol version >= 2.0, and the cluster has aae enabled and the replication
%% configuration stanza enables the aae strategy (the default is keylist),
In other words, it assumes a default. I would expect the init function to then implement that default by initialising the state with keylist alternatively to follow the logic.
At this moment there is a possible race condition when the server is started and a call legacy_status comes in, this may crash the server (line 107). Dializer seems not to spot this, though.
Co-Authored-By: Thomas Arts <[email protected]>
Co-Authored-By: Thomas Arts <[email protected]>
Dialyzer fixes.
In gen_leader.erl the fix to dialyzer error on system_terminate/4 is clunky. It is not clear why system_terminate/4 had no return, and how moving the code around in this way means there is no no problem.
Rolled in for src/riak_repl2_fscoordinator.erl is a fix to the intermittently failing test #799. This had been incorrectly ignored in previous releases (given that it was accepted there were known problems in aae full-sync in these test cases) - but know following this fix has predictable (passing) behaviour