Data import FAQ

• General questions

  • How do I choose a load method?

  • What factors affect load performance?

  • How do I resolve the "close index channel failed" or "too many tablet versions" error?

  • How do I resolve the "Label Already Exists" error?

  • How do I resolve the "ETL_QUALITY_UNSATISFIED; msg:quality not good enough to cancel" error?

  • How do I troubleshoot a remote procedure call (RPC) timeout that occurs during data loading?

  • How do I resolve the "Value count does not match column count" error?

  • How do I resolve the "ERROR 1064 (HY000): Failed to find enough host in all backends. need: 3" error?

  • How do I resolve the "Too many open files" error found in the BE service logs during data loading?

  • How do I resolve the "increase config load_process_max_memory_limit_percent" error?

  • Why does disk space usage spike during data loading?

  • How do I resolve errors that occur when reading data from an OSS external table using JindoSDK?

  • How do I resolve the "transmit chunk rpc failed" error during data loading?

  • What should I do if the memory used by the primary key index exceeds the limit during data loading?

  • Why do I get a cancellation error when reading data using a Spark connector?

  • How do I troubleshoot slow transaction processing when running a load job with the Spark connector?

• Stream load

  • Does stream load support identifying column names in the first row of a text file, or can I skip the first row?

  • The data type for my partition key is not a standard DATE or INT. Do I need to convert the data when loading it into StarRocks using stream load?

  • How do I resolve the "body exceed max size: 10737418240, limit: 10737418240" error?

  • How do I resolve the "Cancelled because of runtime state is cancelled" error that occurs during a stream load?

• Routine load

  • How do I improve load performance?

  • What do I do if the status of a load job changes to PAUSED or CANCELLED after I run the SHOW ROUTINE LOAD command?

  • Does using routine load to consume data from Kafka and write it to StarRocks guarantee consistency semantics?

  • How do I resolve the "Broker: Offset out of range" error?

• Broker load

  • Does broker load support rerunning a load job that has finished successfully and is in the FINISHED state?

  • How do I fix garbled characters that appear during a broker load?

  • When using broker load to load HDFS data, the date field is 8 hours ahead of the correct time. How do I resolve this?

  • How do I resolve the "ErrorMsg: type:ETL_RUN_FAIL; msg:Cannot cast '' from VARCHAR to ARRAY" error when loading ORC-formatted data using broker load?

  • Why does a broker load job report no errors, but I cannot query any data?

  • How do I resolve the "failed to send batch" or "TabletWriter add batch with unknown id" error?

  • How do I resolve the "LOAD-RUN-FAIL; msg:OrcScannerAdapter::init_include_columns. col name = xxx not found" error?

  • How to troubleshoot issues such as a load job that takes too long to finish?

  • How do I configure access to an Apache HDFS cluster in high availability (HA) mode?

  • How do I configure Hadoop ViewFS Federation?

  • How do I resolve the "Can't get Kerberos realm" error when accessing a cluster with Kerberos authentication?

• INSERT INTO

  • When loading data with an INSERT INTO statement, each row takes about 50-100 ms to insert. Can the performance be optimized?

  • How do I resolve the "index channel has intolerable failure" error when loading data using an INSERT INTO SELECT statement?

  • How do I resolve the "execute timeout" error when loading a large amount of data using an INSERT INTO SELECT statement?

• Real-time sync from MySQL to StarRocks

  • How do I resolve the "Could not execute SQL statement. Reason:org.apache.flink.table.api.ValidationException: One or more required options are missing" error when running a Flink job?

  • How do I automatically restart failed tasks in Flink?

  • How do I manually stop a Flink job and then restore it to its previous state?

• Flink connector

  • How do I resolve a load failure when using the exactly-once transactional API?

  • The timestamp is 8 hours behind when sinking data to StarRocks using flink-connector-jdbc_2.11. How do I resolve this?

  • Why are the BE service's memory and CPU fully utilized even when no queries are running?

  • Why doesn't the BE service release its allocated memory back to the operating system?

  • Why can't Flink connector dependencies be resolved?

  • Does the sink.buffer-flush.interval-ms parameter take effect in the Flink connector for StarRocks?

  • How do I resolve the "Caused by: Could not resolve host for client socket" error when a Flink job writes to StarRocks?

• DataX writer

  • Does loading data with DataX support updates?

  • How do I handle keywords used in names to prevent errors when synchronizing data with DataX?

• Spark load

  • How do I resolve the "When running with master 'yarn' either HADOOP-CONF-DIR or YARN-CONF-DIR must be set in the environment" error?

  • How do I resolve the 'Cannot run program "xxx/bin/spark-submit": error=2, No such file or directory' error when submitting a Spark job with the spark-submit command?

  • How do I resolve the "File xxx/jars/spark-2x.zip does not exist" error?

  • How do I resolve the "YARN client does not exist in path: xxx/yarn-client/hadoop/bin/yarn" error?

"Close index channel failed" or "too many tablet versions" error

• Cause

  This issue is caused by frequent data imports. When new data versions are created faster than the system can merge them through compaction, the number of uncompacted versions exceeds the system limit.

• Solution

  The default limit for uncompacted versions is 1,000. You can resolve this issue with the following methods:

  • Increase the batch size for each data import and reduce the import frequency.

  • Modify the following parameters in the BE configuration file be.conf to accelerate compaction.

    cumulative_compaction_num_threads_per_disk = 4
    base_compaction_num_threads_per_disk = 2
    cumulative_compaction_check_interval_seconds = 2

The "Label Already Exists" error

• Problem

  This error indicates that a load job with the same label is already running or has successfully completed in the same database on your StarRocks cluster.

• Cause

  Stream Load submits load jobs over HTTP, and most HTTP clients have a built-in retry mechanism. When the StarRocks cluster receives the initial request, it starts the load job. However, if the cluster fails to respond before the client's request times out, the client may automatically resend the request. The cluster is already processing the original request, so it rejects the duplicate and returns the Label Already Exists error.

• Solution

  Check for label conflicts between different load methods or for duplicate load job submissions. Troubleshoot the issue as follows:

  • Search the primary FE logs for the job label. If the same label appears twice, this confirms the client resubmitted the request.

    Note

    In a StarRocks cluster, load job labels are unique within a database, regardless of the load method. Therefore, a conflict can occur if different load jobs use the same label.

  • Run the SHOW LOAD WHERE LABEL = "xxx" command to check whether a load job with the same label exists in the FINISHED state. Replace xxx with the label you are checking.

  As a best practice, estimate the load time based on the data volume of your request. Then, set the client's request timeout to be longer than that estimate. This prevents the client from prematurely resubmitting the request.

The "ETL_QUALITY_UNSATISFIED" error

Error 1064: Failed to find enough host

When creating a table, add "replication_num" = "1" to the table properties.

Resolve 'Too many open files' error

The "increase config load_process_max_memory_limit_percent" error

Troubleshoot RPC timeouts

Check the write_buffer_size setting in the BE configuration file. This parameter controls the size threshold for memory blocks on a BE, and its default value is 100 MB. If the threshold is too large, it may cause a remote procedure call (RPC) timeout. In this case, adjust write_buffer_size along with the tablet_writer_rpc_timeout_sec parameter. For more on BE parameters, see parameter configuration.

"Value count does not match column count" error

• Problem

  An import job returns the "Value count does not match column count" error when the number of columns parsed from the source data does not match the number of columns in the destination table.

• Cause

  This error occurs when the column delimiter specified in the import command or import statement does not match the one in the source data. For example, if a three-column file in CSV format uses a comma (,) as its delimiter, but the import command or import statement specifies a tab character (\t), the file is incorrectly parsed as a single column.

• Solution

  Change the column delimiter in the import command or import statement to match the one in your source data, and then retry the import.

Choose an import method

For guidance on choosing an import method, see data import.

Factors affecting import performance

Handle header rows during Stream Load

Transform non-standard partition key data

Yes. StarRocks lets you transform data during the load process.

-H "columns: NO,DATE_1, VERSION, PRICE, DATE=LEFT(DATE_1,6)"

In this statement, DATE_1 is a temporary variable that holds the raw data. The LEFT() function transforms this data, and the result is assigned to the DATE column in the StarRocks table. Note that you must first list temporary names for all columns in the CSV file and then define the transformation expressions. Supported functions for column transformation include scalar functions, including non-aggregate functions and window functions.

"body exceed max size: 10737418240, limit: 10737418240" error

• Cause

  The source data file exceeds the 10 GB size limit for a Stream Load job.

• Solution

  • Split the data file into smaller files using seq -w 0 n.

  • Run curl -XPOST http://be_host:http_port/api/update_config?streaming_load_max_mb=<file_size> to adjust the streaming_load_max_mb BE configuration item. For more information about BE configuration items, see parameter configuration.

Improve import performance

min(alive_be_number, partition_number, desired_concurrent_number, max_routine_load_task_concurrent_num)

I0325 20:27:50.410579 15259 data_consumer_group.cpp:131] consumer group done: 41448fb1a0ca59ad-30e34dabfa7e47a0. consume time(ms)=3261, received rows=179190, received bytes=9855450, eos: 1, left_time: -261, left_bytes: 514432550, blocking get time(us): 3065086, blocking put time(us): 24855

Troubleshoot PAUSED or CANCELLED Routine Load jobs

Routine Load and exactly-once semantics

Routine Load guarantees exactly-once semantics.

Each import job is a single transaction. If an error occurs during execution, the transaction aborts, and the FE does not update the offsets for the relevant partitions. When the FE reschedules the job, it resumes consumption from the last saved offsets, ensuring exactly-once semantics.

Error: Broker: Offset out of range

Can a finished Broker Load job be rerun?

No, you cannot rerun a Broker Load job that is in the FINISHED state. To ensure exactly-once semantics, you cannot reuse the label of a completed load job. As a workaround, use SHOW LOAD to find the original job, copy its configuration, assign a new label, and submit a new load job.

Garbled content in Broker Load

Reason: column count mismatch, expect=6 real=1. src line: [$交通]; zcI~跟团+v];   count mismatch, expect=6 real=2. src line: [租erD食休闲娱乐

This issue is caused by an incorrect FORMAT AS parameter in the import job request. To fix it, set the FORMAT AS parameter to match your source file's format and retry the import job.

Date fields are 8 hours ahead with Broker Load

• Cause

  The cause is a timezone mismatch. The StarRocks table and the Broker Load job are configured with China Standard Time, but the server uses UTC. This mismatch shifts the date field forward by 8 hours during the load.

• Solution

  Remove the timezone parameter during table creation.

Broker Load error: Cannot cast '<slot 6>' from VARCHAR to ARRAY<VARCHAR(30)>

• Cause

  This error occurs because the column names in the source data file do not match those in the StarRocks table. When the SET clause is executed, StarRocks performs type inference, but the data type conversion fails when the cast function is called.

• Solution

  Ensure that the column names in the source file and the target table are identical. This eliminates the need for a SET clause. As a result, StarRocks does not call the cast function for data type conversion, which allows the import to succeed.

Missing data after successful Broker Load

Broker Load is an asynchronous load method. Although a statement to create a load job may complete without errors, the job itself may not have succeeded. Run the SHOW LOAD statement to check the result status and errmsg of the load job. You can then adjust the parameters and retry the job.

The "failed to send batch" or "TabletWriter add batch with unknown id" error

This error is caused by a data write timeout. To resolve it, modify the system variable query_timeout and the BE configuration item streaming_load_rpc_max_alive_time_sec. For more information about BE configuration items, see parameter configuration.

Resolve the "LOAD-RUN-FAIL; msg:OrcScannerAdapter::init_include_columns. col name = xxx not found" error

When importing data in Parquet or ORC format, ensure the column names in the file header match those in the StarRocks table.

(tmp_c1,tmp_c2)
SET
(
   id=tmp_c2,
   name=tmp_c1
)

When loading an ORC file generated directly by Apache Hive with a table header of (_col0, _col1, _col2, ...), the load job may fail with an "Invalid Column Name" error. In this case, use the SET clause to define column conversion rules.

Troubleshoot a slow load job

In the FE log file fe.log, find the load job ID by searching for its label. Then, in the BE log file be.INFO, use this ID to find the root cause from the log context.

Configure access to an HDFS HA cluster

(
    "username"="user",
    "password"="passwd",
    "dfs.nameservices" = "my-ha",
    "dfs.ha.namenodes.my-ha" = "my_namenode1,my_namenode2",
    "dfs.namenode.rpc-address.my-ha.my-namenode1" = "nn1-host:rpc_port",
    "dfs.namenode.rpc-address.my-ha.my-namenode2" = "nn2-host:rpc_port",
    "dfs.client.failover.proxy.provider.my-ha" = "org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider"
)

Add the HDFS cluster configurations to the hdfs-site.xml file. When a broker process reads information from the HDFS cluster, it requires only the cluster's file path and authentication information.

Configure Hadoop ViewFS Federation

Copy the core-site.xml and hdfs-site.xml configuration files to the broker/conf directory.

If you use a custom file system, also copy the corresponding .jar files to the broker/lib directory.

"Can't get Kerberos realm" error

1. Verify that the /etc/krb5.conf file is configured on every machine that hosts a broker.

2. If an error still occurs after configuration, you need to add -Djava.security.krb5.conf:/etc/krb5.conf to the end of the JAVA_OPTS variable in the broker's startup script.

Slow performance of single-row INSERT INTO

The INSERT INTO statement is designed for batch writes. This means a single write incurs roughly the same overhead as a batch write. Therefore, for OLAP workloads, avoid using INSERT INTO statements for single-row insertions.

Resolve the "index channel has intolerable failure" error

This error occurs because the streaming load RPC times out. To resolve this issue, adjust the relevant RPC timeout parameters in the configuration file.

Resolve "execute timeout" error with INSERT INTO SELECT

This error indicates a query timeout. To resolve this, adjust the session variable query_timeout. This parameter defaults to 600 seconds.

set query_timeout = xx;

Flink job error: "One or more required options are missing"

• Cause

  In the StarRocks-migrate-tools (SMT) configuration file config_prod.conf, multiple rule groups such as [table-rule.1] and [table-rule.2] are configured, but necessary configuration is missing.

• Solution

  Check whether the database, table, and Flink connector information is configured for each rule group, such as [table-rule.1] and [table-rule.2].

Automatic restart of failed tasks

Flink uses checkpointing and a restart strategy to automatically restart failed tasks.

# unit: ms
execution.checkpointing.interval: 300000
state.backend: filesystem
state.checkpoints.dir: file:///tmp/flink-checkpoints-directory

Manually stop and restore a Flink job

When you stop a Flink job, you can manually trigger a Savepoint. A Savepoint is a consistent snapshot of the execution state of a streaming job, created through the Checkpointing mechanism. You can later restore the Flink job from a specified Savepoint.

Stop a job and create a Savepoint. This automatically triggers a Savepoint for the specified Flink job before stopping it. You can also specify a target directory to store the Savepoint.

bin/flink stop --type [native/canonical] --savepointPath [:targetDirectory] :jobId

./flink run -c com.starrocks.connector.flink.tools.ExecuteSQL -s savepoints_dir/savepoints-xxxxxxxx flink-connector-starrocks-xxxx.jar -f flink-create.all.sql

Exactly-once transaction load failure

• Problem: You receive the following error message:

  com.starrocks.data.load.stream.exception.StreamLoadFailException: {
      "TxnId": 3382****,
      "Label": "502c2770-cd48-423d-b6b7-9d8f9a59****",
      "Status": "Fail",
      "Message": "timeout by txn manager",
      "NumberTotalRows": 1637,
      "NumberLoadedRows": 1637,
      "NumberFilteredRows": 0,
      "NumberUnselectedRows": 0,
      "LoadBytes": 4284214,
      "LoadTimeMs": 120294,
      "BeginTxnTimeMs": 0,
      "StreamLoadPlanTimeMs": 7,
      "ReadDataTimeMs": 9,
      "WriteDataTimeMs": 120278,
      "CommitAndPublishTimeMs": 0
  }

• Cause: The sink.properties.timeout is shorter than the Flink checkpoint interval, leading to a transaction timeout.

• Solution: Increase the value of this parameter to exceed the Flink checkpoint interval.

8-hour time lag with flink-connector-jdbc_2.11

• Problem: The localTimestamp function in Flink generates correct timestamps, but they lag by 8 hours when written to a StarRocks sink table. This issue occurs even though both the Flink and StarRocks servers are set to the Asia/Shanghai time zone. The environment uses Flink 1.12 and the flink-connector-jdbc_2.11 driver.

• Solution: To resolve this, explicitly configure the time zone by setting the 'server-time-zone' = 'Asia/Shanghai' parameter in your Flink sink table definition and appending &serverTimezone=Asia/Shanghai to the url parameter. The following example demonstrates the correct configuration.

  CREATE TABLE sk (
      sid int,
      local_dtm TIMESTAMP,
      curr_dtm TIMESTAMP
  )
  WITH (
      'connector' = 'jdbc',
      'url' = 'jdbc:mysql://192.168.**.**:9030/sys_device?characterEncoding=utf-8&serverTimezone=Asia/Shanghai',
      'table-name' = 'sink',
      'driver' = 'com.mysql.jdbc.Driver',
      'username' = 'sr',
      'password' = 'sr123',
      'server-time-zone' = 'Asia/Shanghai'
  );

High BE memory and CPU usage at idle

The high CPU usage is temporary and occurs when the BE periodically collects statistics. High memory usage is expected. To optimize performance, a BE retains up to 10 GB of memory by default, managing it internally instead of releasing it back to the operating system. You can adjust this limit using the tc_use_memory_min parameter.

The tc_use_memory_min parameter specifies the minimum amount of memory retained by TCmalloc. The default value is 10737418240 bytes (10 GB). StarRocks releases idle memory back to the operating system only when its memory usage exceeds this value. You can configure this parameter in the be.conf file on the Configure tab of the StarRocks service in the EMR console. For more information about BE configuration, see Parameter Configuration.

BE memory is not released to the OS

The database requests large blocks of memory from the operating system. To avoid the high overhead of large-scale memory allocation, it pre-allocates extra memory and delays its release for reuse. To observe this behavior, monitor memory usage in a test environment over an extended period to verify that the memory is eventually released.

Unresolved Flink connector dependencies

• Cause: The dependencies cannot be resolved because the mirror section in /etc/maven/settings.xml is not configured with the Alibaba Cloud mirror address.

• Solution: Set the Alibaba Cloud public repository address to https://maven.aliyun.com/repository/public.

Behavior of sink.buffer-flush.interval-ms

• Problem: If sink.buffer-flush.interval-ms is 15s and the checkpoint interval is 5 minutes, does sink.buffer-flush.interval-ms still take effect?

  +----------------------+--------------------------------------------------------------+
  |         Option       | Required |  Default   | Type   |       Description           |
  +-------------------------------------------------------------------------------------+
  |  sink.buffer-flush.  |  NO      |   300000   | String | the flushing time interval, |
  |  interval-ms         |          |            |        | range: [1000ms, 3600000ms]  |
  +----------------------+--------------------------------------------------------------+

• Solution: A flush is triggered when any of the following parameters reaches its threshold. This behavior is independent of the checkpoint interval setting. The checkpoint interval applies only to exactly once semantics; at least once semantics rely on sink.buffer-flush.interval-ms.

  sink.buffer-flush.max-rows
  sink.buffer-flush.max-bytes
  sink.buffer-flush.interval-ms

Data updates with DataX

DataX supports data updates with the primary key model. To enable this, add the _op field to the reader section of your JSON configuration file.

Handle keywords in names for DataX

Enclose the corresponding field in backticks ().

Error: HADOOP-CONF-DIR or YARN-CONF-DIR not set

This error occurs when using Spark Load because the HADOOP-CONF-DIR environment variable is not set in the spark-env.sh file on the Spark client.

Error: Cannot run program "xxx/bin/spark-submit": error=2, No such file or directory

This error indicates that when using Spark Load, the spark_home_default_dir parameter is either unspecified or points to an incorrect Spark client root directory.

"File xxx/jars/spark-2x.zip does not exist" error

This Spark Load error indicates that the spark-resource-path configuration option does not point to the packaged ZIP file. Verify that the file path and file name are correct.

Resolve "YARN client does not exist" error

This error occurs when using Spark Load because the yarn-client-path configuration option does not correctly point to the YARN executable file.

High disk usage triggers safe mode

• Problem: When the disk usage of a BE node in the cluster exceeds the threshold, the system automatically enters safe mode, blocking new data write operations. This issue often results from the accumulation of large amounts of garbage data in the trash directory after large-scale delete or update operations.

• Cause: Insufficient disk space.

  When disk usage exceeds the threshold, the system triggers safe mode to block further data write operations and prevent the disk from becoming full.

  Safe mode is triggered when the available space on a disk drops below a calculated threshold. This threshold is the minimum of the following two values:

  • 10% of the total capacity of a single disk (i.e., 10% * diskInfo.getTotalCapacityB()).

  • The value of the safe_mode_check_disk_space configuration parameter, which defaults to 100 GB (107,374,182,400 bytes).

  The formula is as follows:

  double safeModeCheckDiskCapacity = Math.min(
      0.10 * diskInfo.getTotalCapacityB(), Config.safe_mode_check_disk_space);

• Solutions:

  • Option 1 (Recommended): Expand disk capacity.

    If you have insufficient disk space, we recommend expanding the disk capacity. This resolves the issue at its root and prevents safe mode from being triggered frequently. For more information, see Disk Expansion.

  • Option 2: Adjust the safe mode threshold.

    If you cannot immediately expand the disk, you can adjust the safe mode threshold as a temporary solution.

    On the Parameter Settings page, add the FE configuration parameter safe_mode_check_disk_space.

    Note

    You can lower this value based on your business requirements, but we do not recommend setting it below 50 GB (53,687,091,200 bytes). Lowering this threshold increases the risk of the disk becoming full, which can make the instance unavailable.

    After the adjustment, the minimum required available space is total capacity * 10% for a disk with less than 1000 GB of total capacity, and the adjusted value of safe_mode_check_disk_space for a disk with 1000 GB or more.

Queries canceled when reading with Spark Connector

• Problem: When you use the Spark Connector to read data from StarRocks, a query might be unexpectedly canceled. This error typically occurs with long-running queries when an automatic cleanup process deletes the expired data versions they need.

• Cause: If a query's execution time exceeds the configured retention period for data versions, the versions that the query requires may be garbage-collected before it completes. As a result, the task is canceled because it cannot find the necessary data.

• Solution: On the Parameter Settings page, increase the lake_autovacuum_grace_period_minutes parameter. The default value is 30 minutes. You can set a higher value based on your actual query requirements, for example, 60 minutes or more, to prevent the premature deletion of data required by long-running queries.

Flink job fails with "Could not resolve host"

To resolve this issue, set the sink.version=V1 parameter in your Flink job. By default, the connector uses version V2, which has stability and compatibility issues. Version V1 is more stable. For more information, see the open source StarRocks documentation for Apache Flink.

Handle excessive memory consumption of primary key index

Consider the following optimization solutions:

• Create a partitioned table: For tables that use the primary key model, creating partitions prevents the primary key index for the entire table from being loaded into memory at once during data writes. This reduces memory pressure.

• Enable the persistent primary key index: This feature persists the primary key index on disk instead of storing it in memory, which reduces memory consumption. Run the following SQL statement to enable it.

       ALTER TABLE xxx SET ("enable_persistent_index"="true");

Jindosdk read errors from OSS external tables

• Problem: An error occurs when you use JindoSDK to read StarRocks external table data from OSS.

• Cause: This issue typically occurs because the pread implementation of JindoSDK does not function as expected, which results in errors when you read data.

• To resolve this issue, on the Parameter Settings page, adjust the configuration in the jindosdk.cfg file to optimize the memory buffer and reduce the occurrence of errors.

  • fs.oss.memory.buffer.size.watermark: The watermark for the memory buffer. Set this value to 0.6 (the default value is 0.3).

  • fs.oss.memory.buffer.size.max.mb: Increases the maximum memory buffer size. We recommend that you set this value to 6144 (6 GB). The default value is 6144 MB.

Disk space spike during data import

• Problem: When you import data into StarRocks, you might observe an unexpected increase in disk space usage, particularly in the partition marked as "other". This can affect system stability and performance.

• Possible cause: This issue is typically caused by lingering temporary or obsolete files from the data import process. The most common reason is an excessive buildup of files in the trash.

• Solution: Adjust the trash_file_expire_time_sec parameter to accelerate the cleanup of files from the trash. This parameter defines how long, in seconds, files remain in the trash before being permanently deleted. Reducing this value releases disk space more quickly.

Slow transactions with the Spark connector

• Problem: You may experience slow transaction processing when running a data import job with the Spark connector on a fully managed StarRocks cluster in Alibaba Cloud E-MapReduce. This issue typically occurs when the number of pending transactions reaches the system's default limit, which significantly slows subsequent data imports.

• Solution: To optimize performance, set the lake_enable_batch_publish_version parameter to true.

Resolve the "transmit chunk rpc failed" error

• Problem: You encounter the following error message when you import data.

  transmit chunk rpc failed [dest_instance_id=5a2489c6-f0d8-11ee-abf6-061aec******] [dest=172.17.**.**:8060]
  detail: brpc failed, error=Host is down, error_text=[E112]Not connected to 172.17.**.**:8060 yet,
  server_id=904 [R1][E112]Not connected to 172.17.**.**:8060 yet, server_id=904 [R2][E112]Not connected to 172.17.**.**:8060 yet, server_id=904 [R3][E112]Not connected to 172.17.**.**:8060 yet, server_id=904
  

• Possible cause: This error typically indicates congestion during RPC communication.

• Solution: You can resolve this issue using one of the following two methods.

  • Adjust the unwritten data threshold. The current state may be OVER CROWDED, which indicates that the RPC source has a large amount of unwritten data that exceeds the default threshold. The default value of brpc_socket_max_unwritten_bytes is 1 GB. If the amount of unwritten data exceeds this value, this error may occur. We recommend that you increase this value appropriately to avoid the OVER CROWDED error.

  • Increase the RPC message body size. Another issue occurs when the RPC body size exceeds the brpc_max_body_size limit, which has a default value of 2 GB. This can happen if a query contains oversized strings or bitmap type data. You can resolve this issue by increasing the BE parameter brpc_max_body_size.

Resolve the "Cancelled because of runtime state is cancelled" error

• Problem: The error "Cancelled because of runtime state is cancelled" occurs when you run a Stream Load job without setting the sink.buffer-flush.interval-ms parameter.

• Possible cause: This issue typically occurs during data loading into StarRocks. The sink.buffer-flush.interval-ms parameter defines the flush interval for the data buffer, determining how often accumulated data is written to the target storage. If you do not explicitly set this parameter, the system uses a default value. This default value may be too large, causing data to remain in memory for an extended period without being flushed. This can lead to the runtime state being canceled due to a timeout or other resource constraints, which triggers the error.

• Solution: Explicitly set the sink.buffer-flush.interval-ms parameter in your code to override the default value. This lets you adjust the buffer flush frequency for your specific use case, ensuring a more efficient and stable data transfer.