SQL Query Issues

Error: "timeout when waiting for send fragments RPC"

This error indicates a communication timeout during RPC fragment transmission. To resolve this, follow these steps:

1. Verify network connectivity between nodes.
2. Consider upgrading to version 2.x, as this issue occurs less frequently compared to pre-2.0 versions.
3. Adjust the following configuration parameters in be.conf and fe.conf to optimize thread pools and timeout settings:

# be.conf
fragment_pool_thread_num_max = 2048
fragment_pool_queue_size = 4096
brpc_num_threads = 256

# fe.conf
remote_fragment_exec_timeout_ms = 30000

Syntax Support: INSERT INTO with WITH Clause

If you encounter compatibility issues with INSERT INTO ... WITH ... SELECT, ensure you are using the correct syntax patterns. Apache Doris supports the following structures:

1. INSERT INTO table_name WITH label xxx ...
2. INSERT INTO table_name (column_list) ...

Refer to the official documentation for specific syntax details regarding the INSERT statement.

Data Import and Operations

Error: "get tableList write lock timeout"

This error usually stems from insufficient JVM memory allocation for the Frontend (FE) or resource contention caused by co-deploying FE and Backend (BE) on the same node.

To diagnose:

1. Check FE memory settings.

2. If the error persists, capture a thread dump for analysis: ``` jstack -l [pid] > thread_dump.log

Stream Load Warning: "[PUBLISH TIMEOUT] transaction commit successfully, BUT data will be visible later"

This message indicates that the data import was committed successfully but visibility is delayed. This behavior was optimized in Apache Doris version 2.0.4. Its recommended to upgrade to version 2.0.4 or later to mitigate this issue.

Stream Load CSV: Handling Default Timestamp Columns

When importing CSV data via Stream Load in to a table with a default timestamp column (e.g., insertTime), you may encounter an error stating "actual column number in csv file is less than schema column number."

To fix this, you must explicitly specify all columns in the request header, including the default column with its corresponding function:

-H "columns: col_a, col_b, col_c, insertTime=current_timestamp()"

Operations and Maintenance

Backend (BE) Fails to Start: RocksDB Lock Error

If the BE fails to start with an error indicating IO error: While lock file: .../LOCK: Resource temporarily unavailable, perform the following checks:

1. Inspect the disk health where the storage directory resides.
2. Check for process conflicts. If an automated restart process is triggering repeated starts:
   • Stop the BE process.
   • Remove or move the lock file located at be/storage/meta/LOCK.
   • Restart the BE service.

Cross-Datacenter Deployment Feasibility

Deploying an Apache Doris cluster across two data centers is generally not recommended. Cross-datacenter latency typically falls in the millisecond range, whereas same-rack latency is usually under 0.1ms. Higher latency increases the risk of brpc timeout errors.

For disaster recovery or active-active scenarios, consider using the Cross-Cluster Replication (CCR) feature available in Apache Doris 2.0.

FE Cluster Clock Sync Error

If fe.log shows Clock delta: xxxx ms... exceeds max permissible delta, it means the time difference between the Master FE and the Follower/Observer FE exceeds the default threshold of 5 seconds.

Solution: Enable NTP (Network Time Protocol) on all nodes to ensure system clocks are synchronized within the limit defined by max_bdbje_clock_delta_ms.

Handling Failed Transaction Rollbacks

If a transaction rollback fails, the system's internal Garbage Collection (GC) mechanism will automatically detect and clean up the residual data.

Architecture: Data Layering and Real-time Computing

When implementing data layering and real-time computing with Apache Doris, consider the following architectural patterns:

1. External Scheduling Tools: Use schedulers like DolphinScheduler to run micro-batch jobs at minute-level intervals.
2. Materialized Views: Utilize materialized views for real-time synchronization and pre-aggregation. Version 2.1 introduces multi-table materialized views, allowing for asynchronous scheduled layering internally without external schedulers.
3. Flink Integration: Implement layering within a Flink application (e.g., ODS -> Kafka -> Flink -> DW -> Kafka -> Flink -> Doris).
4. Future CDC Support: Future plans for Doris Binlog will enable patterns like ODS -> Doris <-> Flink. Currently, the first two methods are the most common choices for scenarios tolerating a 3-5 minute data latency.