@Rocky what are steps to downgrade from 4.1.4 to 4.0.14?

Downgrading from StarRocks v4.1.4 to v4.0.14 is a supported minor version downgrade because your target version (v4.0.14) is higher than v4.0.6, which is the minimum required version for downgrades from the v4.1 branch. Due to internal data layout and metadata changes introduced in v4.1 (such as tablet splitting and distribution mechanisms), downgrading to any version earlier than v4.0.6 is not supported and will cause metadata incompatibility. Phase 1: Compatibility & Preparation Before starting, ensure all schema change jobs are completed and disable the cluster's balancing tasks to prevent data movement during the process. 1. Check Schema Changes: Ensure no

ALTER TABLE

jobs are running.

sql
    SHOW PROC '/jobs/<db_name>/schema_change';

2. Disable Tablet Balancing:

sql
    ADMIN SET FRONTEND CONFIG ("tablet_sched_max_scheduling_tablets" = "0");
    ADMIN SET FRONTEND CONFIG ("tablet_sched_max_balancing_tablets" = "0");
    ADMIN SET FRONTEND CONFIG ("disable_balance" = "true");
    ADMIN SET FRONTEND CONFIG ("disable_colocate_balance" = "true");

3. Create Metadata Snapshot: This ensures the Leader FE saves the latest state before the downgrade.

sql
    ALTER SYSTEM CREATE IMAGE;

Verify the image is created and synchronized to other FEs by checking the

fe.log

for a "push image... successful" message. Phase 2: Downgrade FE (Frontends) You must downgrade Follower FEs first, then the Leader FE. 1. Stop the FE node:

bash
    cd <fe_dir>/fe
    ./bin/stop_fe.sh

2. Replace Binaries: Replace the

bin

,

lib

, and

spark-dpp

directories with those from the v4.0.14 package.

bash
    # Backup current folders
    mv lib lib.bak && mv bin bin.bak && mv spark-dpp spark-dpp.bak
    # Copy v4.0.14 folders
    cp -r /path/to/StarRocks-4.0.14/fe/lib .
    cp -r /path/to/StarRocks-4.0.14/fe/bin .
    cp -r /path/to/StarRocks-4.0.14/fe/spark-dpp .

3. Start the FE node:

bash
    ./bin/start_fe.sh --daemon

4. Verify: Check

fe.log

to ensure the FE joined the cluster. Repeat for all Followers, then the Leader. Phase 3: Downgrade BE/CN (Backends/Compute Nodes) Once FEs are stable on v4.0.14, proceed with the BE/CN nodes one by one. 1. Stop the BE/CN node:

bash
    ./bin/stop_be.sh # or stop_cn.sh --graceful

2. Replace Binaries: Replace

bin

and

lib

with v4.0.14 versions (similar to the FE step). 3. Start the node:

bash
    ./bin/start_be.sh --daemon

Phase 4: Post-Downgrade Once the entire cluster is running v4.0.14 and all nodes show as

Alive

in

SHOW FRONTENDS

and `SHOW BACKENDS`: 1. Re-enable Balancing:

sql
    ADMIN SET FRONTEND CONFIG ("tablet_sched_max_scheduling_tablets" = "10000");
    ADMIN SET FRONTEND CONFIG ("tablet_sched_max_balancing_tablets" = "500");
    ADMIN SET FRONTEND CONFIG ("disable_balance" = "false");
    ADMIN SET FRONTEND CONFIG ("disable_colocate_balance" = "false");

Troubleshooting * Metadata Incompatibility: If an FE fails to restart with an

UNKNOWN Operation Type

error, you may need to temporarily set

metadata_ignore_unknown_operation_type = true

in

fe.conf

, restart, create a new image via

ALTER SYSTEM CREATE IMAGE

, and then remove the setting. References • release_notes/release-4.0.md • deployment/downgrade.md • release_notes/release-3.3.md