You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Restart Core after an RPC whitelist change in dashmate, and drop drive-abci's selective whitelist probe #5820
drive-abci's wait_for_core_to_sync v0 (source) probes one Core RPC at startup, getspecialtxes through the credit pool balance read (added with the protocol 14 withdrawal limit, #5237). If Core refuses it, drive-abci logs "restart core so it loads the rpc whitelist of drive's consensus user" and refuses to start.
The reason is real: dashmate's config migrations add methods to the consensus user's rpcwhitelist (5.0.0-beta.2: getspecialtxes, 5.0.0-rc.2: getdeploymentinfo, and protocol 15 adds protxinfo), but Core reads the whitelist only when Core itself starts. A node whose Core was not restarted after such a migration answers every other call and then refuses every block once the new method is needed.
Problem with the current check
Wrong layer. drive-abci encodes dashmate's operational knowledge (how Core is configured, that a restart reloads the whitelist).
Selective. It checks one method, the one added at the time, and not the other methods drive-abci calls.
Proposal
dashmate owns it: when a config migration changes what a service's rendered config contains (for Core, dash.conf and its rpcwhitelist), dashmate marks that service for restart, and the update path restarts it. This belongs with the dashmate update-path work.
Then either remove drive-abci's probe, or replace it with one generic startup check of every Core RPC method drive-abci calls, from a single list, with a neutral error that does not refer to dashmate.
#5775 (protocol 15 operator payouts) does not add a probe for protxinfo, for this reason.
Activity