@Rocky while downscaling the cluster size, all the nodes have become follower and there is no leader currently. getting the following error:

ERROR 1064 (HY000): (JE 18.3.20) Could not determine master from helpers

It sounds like your FE (FrontEnd) cluster has lost its quorum during the downscaling process. This error typically occurs when the remaining FE nodes cannot reach a majority to elect a new Leader, often because the metadata still expects the nodes you removed to be part of the election group. To recover from this state, you need to force one of the remaining Follower nodes to become the Leader and reset the election group. Recovery Steps (The "Measure of Last Resort") Follow these steps on one of your remaining FE Follower nodes (preferably the one with the most up-to-date metadata): 1. Stop all FE nodes in the cluster. 2. Modify the configuration of the chosen FE node: Edit

fe/conf/fe.conf

and add the following line:

properties
    bdbje_reset_election_group = true

3. Restart this FE node:

bash
    ./fe/bin/start_fe.sh --daemon

4. Verify the status: Log into this FE node via MySQL client and run:

sql
    SHOW FRONTENDS;

Confirm that this node has the role

LEADER

and

Alive

is

true

. 5. Clean up the configuration: Once the node is successfully running as the Leader, remove the

bdbje_reset_election_group = true

line from

fe.conf

. You should restart this node one more time without that flag to ensure it's running in a normal state. 6. Re-add other nodes: For the other FE nodes you want to keep: * Stop the FE process. * Crucial: Clear their metadata directory (the path defined by

meta_dir

in

fe.conf

, default is

fe/meta

). * Restart them using the

--helper

flag pointing to the new Leader:

bash
        ./fe/bin/start_fe.sh --helper <leader_ip>:<edit_log_port> --daemon

Why did this happen? In StarRocks, FE Followers use a protocol (BDBJE) that requires a majority (quorum) to elect a leader. If you had 3 followers and removed 2 without the cluster correctly updating its internal voting list, the remaining 1 node cannot form a majority (which would be 2 out of 3) and thus cannot become a Leader. Note: For future downscaling, ensure you use

ALTER SYSTEM DROP FOLLOWER "ip:port";

while the cluster is still healthy so the metadata is updated correctly before nodes are taken offline. References • administration/Meta_recovery.md • faq/Deploy_faq.md