Common Operational, Data, and SQL Issues in Apache Doris – FAQs and Solutions

Part 1: Common Operational Issues

Q1. Why do some tablets remain after DECOMMISSIONing a BE node?

Tablets may stay because they belong to recently deleted objects still in the recycle bin (adjust catalog_trash_expire_second) or because migration tasks failed (check show proc "/cluster_balance").

Q2. How to set priority_network ?

Set it explicitly in CIDR format (e.g., 10.168.0.0/16) to ensure the correct NIC IP is chosen.

Q3. What are FE Master, Follower, and Observer?

FE has only Follower and Observer roles; the Master is a selected Follower. Observers do not participate in elections.

Q4. New disks added to a node are not balanced.

Doris balances at the node level, not per‑disk. Solutions: recreate the table, use DECOMMISSION with

admin set frontend config("drop_backend_after_decommission" = "false")

, or manually migrate via HTTP API.

Q5. How to read FE/BE logs correctly?

Key FE logs: fe.log, fe.warn.log, fe.out, fe.audit.log. Example line:

2021-09-16 23:13:22,502 INFO (tablet scheduler|43) [BeLoadRebalancer.selectAlternativeTabletsForCluster():85] cluster is balance: default_cluster with medium: HDD. skip

. BE logs: be.INFO, be.WARNING, be.out. Example line:

I0916 23:21:22.038795 28087 task_worker_pool.cpp:1594] finish report TASK. master host: 10.10.10.10, port: 9222

.

Q6. How to troubleshoot a crashed FE/BE node?

Check be.out for stack traces, use dmesg for OOM, examine be.INFO for fatal logs, and consider minidump analysis.

Q7. Storage medium mismatch error.

Ensure default_storage_medium matches the path suffix (e.g., .SSD) or modify BE/F​E configs accordingly.

Q8. Nginx load‑balancing FE Web UI login issues.

Use ip_hash in the upstream block to keep sessions sticky.

Q9. FE fails to start with "wait catalog to be ready. FE type UNKNOWN".

Check priority_network IP mismatch or start enough Follower nodes for master election.

Q10‑Q14. Various version, JDK, ODBC, and parser table errors.

Adjust JDK versions, use compatible ODBC drivers, or copy missing parser files as instructed.

Part 2: Data Operation Problems

Q1. Stream Load redirected to internal IP.

FE redirects to a BE’s internal IP; use a load balancer or ensure internal network access.

Q2. Can column names be modified?

No; column names are stored in data files and cannot be changed without rewriting data.

Q3. Can materialized views be created on Unique Key tables?

No; Unique Key tables do not support materialized views due to lack of detailed row information.

Q4. Tablet writer failures (-235, -215, -238).

-235: tablet version count exceeds max_tablet_version_num. Reduce ingest rate or check compaction. -238: too many segments per rowset; increase max_segment_num_per_rowset or reduce batch size.

Q5‑Q9. Various replica, disk, RPC, and connection errors.

Use show backends, show tablet, and admin repair commands; adjust parameters like brpc_socket_max_unwritten_bytes, tablet_writer_ignore_eovercrowded, or brpc_max_body_size as needed.

Part 3: SQL Issues

Q1. "Failed to get scan range, no queryable replica found".

Check tablet replica health with show tablet and show proc "/cluster_balance".

Q2. Incomplete information from show backends/frontends .

Execute on the Master FE or set set forward_to_master=true;.

Q3. Invalid cluster ID errors.

Ensure all nodes belong to the same cluster; adjust or delete cluster_id files as needed.

Q4. Inconsistent results on Unique Key tables.

Duplicate keys with different values cause replica divergence; consider using a sequence column for deterministic ordering.