Remaining concerns for upgrade_client
implementation
#391
Labels
O: reliability
Objective: cause to improve trustworthiness and consistent performing
Milestone
In continuation of #349
Summary
The investigation conducted around the upgrade client implementation (ADR06) left some unanswered questions, primarily arising from the
ibc-go
implementation approach. Before enabling this feature, these concerns should be addressed:Details
Considering the point made here, though decreasing the unbounding period can be a risky action, there isn't any check for this purpose in
ibc-go
Upgrading a frozen client or one where the upgrade message is outside the trust period, both are accepted. Is there any case related to these situations in which we should also avoid preserving connections? (at any cost)
There have been breaking changes in how client upgrades work in the 02-client-refactor PR. Specifically, in v5.0.1 (the version ibc-rs is tracking), the upgraded ClientState doesn’t get its “custom fields” zero’d, but in main, it does. This is a protocol breaking change, right? So which of the 2 versions should ibc-rs replicate? And even within ibc-go chains, doesn’t that cause problems if e.g. a chain connected to the hub upgrades to the latest ibc-go, but not the hub?
The text was updated successfully, but these errors were encountered: