Bug #91150 | Force primary to stop accept writes on low number of members in the group | ||
---|---|---|---|
Submitted: | 5 Jun 2018 16:02 | Modified: | 1 Aug 2018 0:17 |
Reporter: | Artem Danilov | Email Updates: | |
Status: | Verified | Impact on me: | |
Category: | MySQL Server: Group Replication | Severity: | S4 (Feature request) |
Version: | 8.0.11 | OS: | Any |
Assigned to: | CPU Architecture: | Any |
[5 Jun 2018 16:02]
Artem Danilov
[6 Jun 2018 4:28]
MySQL Verification Team
Hello Artem, Thank you for the feature request! Thanks, Umesh
[1 Aug 2018 0:17]
Artem Danilov
One other use case which came to my mind in addition to bugs in automation is the group replication bootstrap. The group replication is bootstrapped with a single node and more nodes join later. However, if someone was able to send writes before any extra node is joined then these writes are not replicated and could result in data loss if the first node dies hard before any other one joined and caught up. Probably this use case can be solved by some extra hoops like starting server always with read_only=1 and make it writable only after more nodes are joined. But my point is that the replication of any committed writes is an invariant which MySQL server should hold for any production environment which takes data losses problem serious. Usually quality and testing of some external automation scripts is not as good as MySQL server and bugs of interacting multiple scripts do happen in huge deployments, especially races appearing in some very rare conditions.