问题排查流程--导入

背景

本文主要描述导入问题的排查思路和常见问题解决方法。这里简单阐述下不同导入方式的流程,方便大家理解导入流程和排查问题,具体的可参考文档导入章节。

排查流程

Stream Load

Stream load内部调用链路

Stream Load是一种同步执行的导入方式。用户通过 HTTP 协议发送请求将本地文件或数据流导入到 StarRocks中,并等待系统返回导入的结果状态,从而判断导入是否成功。

暂时无法在文档外展示此内容

  1. 通过返回值判断
{
    "Status":"Fail",
    "BeginTxnTimeMs":1,
    "Message":"too many filtered rows",
    "NumberUnselectedRows":0,
    "CommitAndPublishTimeMs":0,
    "Label":"4682d766-0e53-4fce-b111-56a8d8bef390",
    "LoadBytes":69238389,
    "StreamLoadPutTimeMs":4,
    "NumberTotalRows":7077604,
    "WriteDataTimeMs":4350,
    "TxnId":33,
    "LoadTimeMs":4356,
    "ErrorURL":"http://192.168.10.142:8040/api/_load_error_log?file=__shard_2/error_log_insert_stmt_e44ae406-32c7-6d5c-a807-b05607a57cbf_e44ae40632c76d5c_a807b05607a57cbf",
    "ReadDataTimeMs":1961,
    "NumberLoadedRows":0,
    "NumberFilteredRows":7077604
}
  1. 返回值Status:非Success,

a. 存在ErrorURL

Curl ErrorURL,例如

curl "http://192.168.10.142:8040/api/_load_error_log?file=__shard_2/error_log_insert_stmt_e44ae406-32c7-6d5c-a807-b05607a57cbf_e44ae40632c76d5c_a807b05607a57cbf"

b.不存在ErrorURL

{
    "TxnId":2271727,
    "Label":"4682d766-0e53-4fce-b111-56a8d8bef2340",
    "Status":"Fail",
    "Message":"Failed to commit txn 2271727. Tablet [159816] success replica num 1 is less then quorum replica num 2 while error backends 10012",
    "NumberTotalRows":1,
    "NumberLoadedRows":1,
    "NumberFilteredRows":0,
    "NumberUnselectedRows":0,
    "LoadBytes":575,
    "LoadTimeMs":26,
    "BeginTxnTimeMs":0,
    "StreamLoadPutTimeMs":0,
    "ReadDataTimeMs":0,
    "WriteDataTimeMs":21,
    "CommitAndPublishTimeMs":0
}

查看本次导入的load_id和调度到的be节点

grep -w $TxnId fe.log|grep "load id"

#输出例子:
2021-12-20 20:48:50,169 INFO (thrift-server-pool-4|138) [FrontendServiceImpl.streamLoadPut():809] receive stream load put request. db:ssb, tbl: demo_test_1, txn id: 1580717, load id: 7a4d4384-1ad7-b798-f176-4ae9d7ea6b9d, backend: 172.26.92.155

在对应的be节点查看具体原因

grep $load_id be.INFO|less
I0518 11:58:16.771597  4228 stream_load.cpp:202] new income streaming load request.id=f145be377c754e94-816b0480c5139b81, job_id=-1, txn_id=-1, label=metrics_detail_1652846296770062737, db=starrocks_monitor, tbl=metrics_detail
I0518 11:58:16.776926  4176 load_channel_mgr.cpp:186] Removing finished load channel load id=f145be377c754e94-816b0480c5139b81
I0518 11:58:16.776930  4176 load_channel.cpp:40] load channel mem peak usage=1915984, info=limit: 16113540169; consumption: 0; label: f145be377c754e94-816b0480c5139b81; all tracker size: 3; limit trackers size: 3; parent is null: false; , load_id=f145be377c754e94-816b0480c5139b81

如果查不到具体原因,可以继续查看线程上下文,比如上文的 4176

grep -w 4176 be.INFO|less

Broker Load

用户在提交导入任务后,FE 会生成对应的 Plan 并根据目前 BE 的个数和文件的大小,将 Plan 分给多个 BE 执行,每个 BE 执行一部分导入任务。BE 在执行过程中会通过 Broker 拉取数据,在对数据预处理之后将数据导入系统。所有 BE 均完成导入后,由 FE 最终判断导入是否成功。

暂时无法在文档外展示此内容

目前一个Broker Load的任务流程会经过PENDING–>LOADING–>FINISHED(或CANCELLED)的流程,当状态为CANCELLED的时候需要介入排查。

  1. Show load查看任务状态,状态为CANCELLED的时候进一步跟进
  2. 如果URL不为空,则curl $URL查看具体报错信息
  3. 如果URL为空,通过fe日志查看load id和be
  4. 检查hdfs文件路径是否指定正确,可以指定到具体文件也可以指定某目录下的所有文件
  5. hdfs导入请检查一下是否有k8s认证,并进行配置
grep $JobnId fe.log
  1. be中查看具体异常
grep $load_id be.INFO

ErrorMsg中的type取值:

  • USER-CANCEL: 用户取消的任务

  • ETL-RUN-FAIL: 在ETL阶段失败的导入任务

  • ETL-QUALITY-UNSATISFIED: 数据质量不合格,也就是错误数据率超过了 max-filter-ratio

  • LOAD-RUN-FAIL: 在LOADING阶段失败的导入任务

  • TIMEOUT: 导入任务没在超时时间内完成

  • UNKNOWN: 未知的导入错误

Routine Load

                 +-----------------+
fe schedule job  |  NEED_SCHEDULE  |  user resume job
     +-----------+                 | <---------+
     |           |                 |           |
     v           +-----------------+           ^
     |                                         |
+------------+   user(system)pause job +-------+----+
|  RUNNING   |                         |  PAUSED    |
|            +-----------------------> |            |
+----+-------+                         +-------+----+
|    |                                         |
|    |           +---------------+             |
|    |           | STOPPED       |             |
|    +---------> |               | <-----------+
|   user stop job+---------------+    user stop job
|
|
|               +---------------+
|               | CANCELLED     |
+-------------> |               |
system error    +---------------+

上图表示的是routine load的任务状态机

show routine load for db.job_name
MySQL [load_test]> SHOW ROUTINE LOAD\G;
*************************** 1. row ***************************
                  Id: 14093
                Name: routine_load_wikipedia
          CreateTime: 2020-05-16 16:00:48
           PauseTime: 2020-05-16 16:03:39
             EndTime: N/A
              DbName: default_cluster:load_test
           TableName: routine_wiki_edit
               State: PAUSED
      DataSourceType: KAFKA
      CurrentTaskNum: 0
       JobProperties: {"partitions":"*","columnToColumnExpr":"event_time,channel,user,is_anonymous,is_minor,is_new,is_robot,is_unpatrolled,delta,added,deleted","maxBatchIntervalS":"10","whereExpr":"*","maxBatchSizeBytes":"104857600","columnSeparator":"','","maxErrorNum":"1000","currentTaskConcurrentNum":"1","maxBatchRows":"200000"}
DataSourceProperties: {"topic":"starrocks-load","currentKafkaPartitions":"0","brokerList":"localhost:9092"}
    CustomProperties: {}
           Statistic: {"receivedBytes":162767220,"errorRows":132,"committedTaskNum":13,"loadedRows":2589972,"loadRowsRate":115000,"abortedTaskNum":7,"totalRows":2590104,"unselectedRows":0,"receivedBytesRate":7279000,"taskExecuteTimeMs":22359}
            Progress: {"0":"13824771"}
ReasonOfStateChanged: ErrorReason{code=errCode = 100, msg='User root pauses routine load job'}
        ErrorLogUrls: http://172.26.108.172:9122/api/_load_error_log?file=__shard_54/error_log_insert_stmt_e0c0c6b040c044fd-a162b16f6bad53e6_e0c0c6b040c044fd_a162b16f6bad53e6, http://172.26.108.172:9122/api/_load_error_log?file=__shard_55/error_log_insert_stmt_ce4c95f0c72440ef-a442bb300bd743c8_ce4c95f0c72440ef_a442bb300bd743c8, http://172.26.108.172:9122/api/_load_error_log?file=__shard_56/error_log_insert_stmt_8753041cd5fb42d0-b5150367a5175391_8753041cd5fb42d0_b5150367a5175391
            OtherMsg:
1 row in set (0.01 sec)

当任务状态为PAUSED或者CANCELLED的时候需要介入排查

任务状态为PAUSED时:

  1. 可以先查看ReasonOfStateChanged定位下原因,例如“Offset out of range”
  2. 若ReasonOfStateChanged为空,查看ErrorLogUrls可查看具体的报错信息
curl ${ErrorLogUrls}

Spark Load

查看任务

show load order by createtime desc limit 1\G

或者

show load order where label="$label"\G
*************************** 1. row ***************************
         JobId: 76391
         Label: label1
         State: FINISHED
      Progress: ETL:100%; LOAD:100%
          Type: SPARK
       EtlInfo: unselected.rows=4; dpp.abnorm.ALL=15; dpp.norm.ALL=28133376
      TaskInfo: cluster:cluster0; timeout(s):10800; max_filter_ratio:5.0E-5
      ErrorMsg: N/A
    CreateTime: 2019-07-27 11:46:42
  EtlStartTime: 2019-07-27 11:46:44
 EtlFinishTime: 2019-07-27 11:49:44
 LoadStartTime: 2019-07-27 11:49:44
LoadFinishTime: 2019-07-27 11:50:16
           URL: http://1.1.1.1:8089/proxy/application_1586619723848_0035/
    JobDetails: {"ScannedRows":28133395,"TaskNumber":1,"FileNumber":1,"FileSize":200000}
  1. 先查看ErrorMsg,判断是否可直观判断
  2. 如果上一步得不到具体异常,则在fe/log/spark_launcher_log/下面查找日志,日志命名为spark-launcher-{load-job-id}-{label}.log
  3. 如果上一步还查不到具体异常,可以在fe.WARNING日志中查找日志
  4. 如果上一步查不到异常,则访问spark ui查到对应executor日志

Insert Into

Insert into也是大家目前遇到问题比较多的导入方式。目前Insert into支持以下两种方式:

  • 方式一:Insert into table values ();
  • 方式二:Insert into table1 xxx select xxx from table2

方式一不建议在线上使用

由于insert into导入方式是同步的,执行完会立即返回结果。可以通过返回结果判断导入成功或失败。

Flink-connector

写入StarRocks是封装的stream load,内部流程可参考Stream Load导入

无法复制加载中的内容

由于Flink-connector底层走的是stream load的方式,所以可以参考stream load排查方式进行。

  1. 首先从Flink日志中搜索"_stream_load"关键字,确认成功发起了stream load任务
  2. 然后排查搜索对应stream load的label,搜索该label的导入返回结果,如下图
{
    "Status":"Fail",
    "BeginTxnTimeMs":1,
    "Message":"too many filtered rows",
    "NumberUnselectedRows":0,
    "CommitAndPublishTimeMs":0,
    "Label":"4682d766-0e53-4fce-b111-56a8d8bef390",
    "LoadBytes":69238389,
    "StreamLoadPutTimeMs":4,
    "NumberTotalRows":7077604,
    "WriteDataTimeMs":4350,
    "TxnId":33,
    "LoadTimeMs":4356,
    "ErrorURL":"http://192.168.10.142:8040/api/_load_error_log?file=__shard_2/error_log_insert_stmt_e44ae406-32c7-6d5c-a807-b05607a57cbf_e44ae40632c76d5c_a807b05607a57cbf",
    "ReadDataTimeMs":1961,
    "NumberLoadedRows":0,
    "NumberFilteredRows":7077604
}
  1. 接下来参考stream load排查流程即可

Flink-CDC

写入StarRocks是封装的stream load,内部流程可参考Stream Load导入

  1. Flink任务没有报错的时候

第一步:确认binlog是否开启,可以通过 SHOW VARIABLES LIKE 'log_bin’查看;

第二步:确认flink、flink-cdc、flink-starrocks-connector和mysql版本(MySQL版本为5.7和8.0.X)是否满足要求,flink、flink-cdc和flink-starrocks-connector的大版本需要一致,例如都是1.13版本

第三步:逐步判断是查源表还是写starrocks的问题,这里利用下面的sql文件演示一下,该文件是Flink-cdc中第7步生成的flink-create.1.sql

CREATE DATABASE IF NOT EXISTS `test_db`;

CREATE TABLE IF NOT EXISTS `test_db`.`source_tb` (
  `id` STRING NOT NULL,
  `score` STRING NULL,
  PRIMARY KEY(`id`)
 NOT ENFORCED
) with (
  'username' = 'root',
  'password' = 'xxx',
  'database-name' = 'test',
  'table-name' = 'test_source',
  'connector' = 'mysql-cdc',
  'hostname' = '172.26.92.139',
  'port' = '8306'
);

CREATE TABLE IF NOT EXISTS `test_db`.`sink_tb` (
  `id` STRING NOT NULL,
  `score` STRING NULL
  PRIMARY KEY(`id`)
 NOT ENFORCED
) with (
  'load-url' = 'sr_fe_host:8030',
  'sink.properties.row_delimiter' = '\x02',
  'username' = 'root',
  'database-name' = 'test_db',
  'sink.properties.column_separator' = '\x01',
  'jdbc-url' = 'jdbc:mysql://sr_fe_host:9030',
  'password' = '',
  'sink.buffer-flush.interval-ms' = '15000',
  'connector' = 'starrocks',
  'table-name' = 'test_tb'
);

INSERT INTO `test`.`sink_tb` SELECT * FROM `test_db`.`source_tb`;

安装的Flink目录下执行下面语句进入flink-sql

bin/sql-client.sh

首先验证读取source表是否正常

#分别把上面的sql粘贴进来判断是查询源表的问题还是写入到starrocks的问题
CREATE DATABASE IF NOT EXISTS `test_db`;

CREATE TABLE IF NOT EXISTS `test_db`.`source` (
  `id` STRING NOT NULL,
  `score` STRING NULL,
  PRIMARY KEY(`id`)
 NOT ENFORCED
) with (
  'username' = 'root',
  'password' = 'xxx',
  'database-name' = 'test',
  'table-name' = 'test_source',
  'connector' = 'mysql-cdc',
  'hostname' = '172.26.92.139',
  'port' = '8306'
);

#验证source是否正常
select * from `test_db`.`source_tb`;

再验证写入starrocks是否正常

CREATE TABLE IF NOT EXISTS `test_db`.`sink_tb` (
  `id` STRING NOT NULL,
  `score` STRING NULL
  PRIMARY KEY(`id`)
 NOT ENFORCED
) with (
  'load-url' = 'sr_fe_host:8030',
  'sink.properties.row_delimiter' = '\x02',
  'username' = 'root',
  'database-name' = 'test_db',
  'sink.properties.column_separator' = '\x01',
  'jdbc-url' = 'jdbc:mysql://sr_fe_host:9030',
  'password' = '',
  'sink.buffer-flush.interval-ms' = '15000',
  'connector' = 'starrocks',
  'table-name' = 'test_tb'
);

INSERT INTO `test`.`sink_tb` SELECT * FROM `test_db`.`source_tb`;
  1. Flink任务出错

第一步:确认flink集群是否有启动,可能有的同学本地下载的flink没有启动,需要./bin/start-cluster.sh启动下flink

第二步:根据具体的报错再具体分析

DataX

写入StarRocks是封装的stream load,内部流程可参考Stream Load导入

无法复制加载中的内容

由于DataX底层也是走的stream load方式,所以可以参考stream load排查方式进行。

  1. 首先从datax/log/YYYY-MM-DD/xxx.log日志中搜索"_stream_load"关键字,确认成功发起了stream load任务

A. 如果没有stream load生成,具体查看datax/log/YYYY-MM-DD/xxx.log日志,分析异常解决

B. 如有stream load生成,在datax/log/YYYY-MM-DD/xxx.log搜索对应stream load的label,搜索该label的导入返回结果,如下图

{
    "Status":"Fail",
    "BeginTxnTimeMs":1,
    "Message":"too many filtered rows",
    "NumberUnselectedRows":0,
    "CommitAndPublishTimeMs":0,
    "Label":"4682d766-0e53-4fce-b111-56a8d8bef390",
    "LoadBytes":69238389,
    "StreamLoadPutTimeMs":4,
    "NumberTotalRows":7077604,
    "WriteDataTimeMs":4350,
    "TxnId":33,
    "LoadTimeMs":4356,
    "ErrorURL":"http://192.168.10.142:8040/api/_load_error_log?file=__shard_2/error_log_insert_stmt_e44ae406-32c7-6d5c-a807-b05607a57cbf_e44ae40632c76d5c_a807b05607a57cbf",
    "ReadDataTimeMs":1961,
    "NumberLoadedRows":0,
    "NumberFilteredRows":7077604
}
  1. 接下来参考stream load排查流程即可

常见问题

  1. “Failed to commit txn 2271727. Tablet [159816] success replica num 1 is less then quorum replica num 2 while error backends 10012”,

这个问题具体原因需要按照上面排查流程在be.WARNING中查看具体异常

  1. close index channel failed/too many tablet versions

导入频率太快,compaction没能及时合并导致版本数过多,默认版本数1000

降低频率,调整compaction策略,加快合并(调整完需要观察内存和io),在be.conf中修改以下内容

base_compaction_check_interval_seconds = 10
cumulative_compaction_num_threads_per_disk = 4
base_compaction_num_threads_per_disk = 2
cumulative_compaction_check_interval_seconds = 2
  1. Reason: invalid value ‘202123098432’.

导入文件某列和表中的类型不一致导致

  1. the length of input is too long than schema

导入文件某列长度不正确,比如定长字符串超过建表设置的长度、int类型的字段超过4个字节。

  1. actual column number is less than schema column number

导入文件某一行按照指定的分隔符切分后列数小于指定的列数,可能是分隔符不正确。

  1. actual column number is more than schema column number

导入文件某一行按照指定的分隔符切分后列数大于指定的列数,可能是分隔符不正确。

  1. the frac part length longer than schema scale

导入文件某decimal列的小数部分超过指定的长度。

  1. the int part length longer than schema precision

导入文件某decimal列的整数部分超过指定的长度。

  1. the length of decimal value is overflow

导入文件某decimal列的长度超过指定的长度。

  1. There is no corresponding partition for this key

导入文件某行的分区列的值不在分区范围内。

  1. Caused by: org.apache.http.ProtocolException: The server failed to respond with a valid HTTP response

Stream load端口配置错误,应该是http_port

  1. flink demo,按要求建立了测试库表,然后程序没有任何报错日志,数据也无法sink进去,请问有什么排查思路呢

可能是无法访问be导致,当前flink封装的stream load,fe接收到请求后会redirect $be:$http_port,一般本地调试的时候,能访问fe+http_port,但是无法访问be+http_port,需要开通访问be+http_port的防火墙

  1. Transaction commit successfully,But data will be visible later

该状态也表示导入已经完成,只是数据可能会延迟可见。原因是有部分publish超时,也可以调大fe配置publish_version_timeout_second

  1. get database write lock timeout

可能是fe的线程数超了,建议可以调整下be配置:thrift_rpc_timeout_ms=10000(默认5000ms)

  1. failed to send batch 或 TabletWriter add batch with unknown id

请参照章节导入总览/通用系统配置/BE配置,适当修改 query_timeout 和 streaming_load_rpc_max_alive_time_sec

  1. LOAD-RUN-FAIL; msg:Invalid Column Name:xxx
  • 如果是Parquet或者ORC格式的数据,需要保持文件头的列名与StarRocks表中的列名一致,如 :
(tmp_c1,tmp_c2)
SET
(
   id=tmp_c2,
   name=tmp_c1
)

表示将Parquet或ORC文件中以(tmp_c1, tmp_c2)为列名的列,映射到StarRocks表中的(id, name)列。如果没有设置set, 则以column中的列作为映射。

注意:如果使用某些Hive版本直接生成的ORC文件,ORC文件中的表头并非Hive meta数据,而是

(_col0, _col1, _col2, ...) , 可能导致Invalid Column Name错误,那么则需要使用set进行映射。

  1. Can't get Kerberos realm

A:首先检查是不是所有的broker所在机器是否都配置了

/etc/krb5.conf 文件。

如果配置了仍然报错,需要在broker的启动脚本中的

JAVA_OPTS 变量最后,加上

-Djava.security.krb5.conf:/etc/krb5.conf

  1. orc数据导入失败ErrorMsg: type:ETL_RUN_FAIL; msg:Cannot cast ‘<slot 6>’ from VARCHAR to ARRAY<VARCHAR(30)>

导入源文件和starrocks两边列名称不一致,set的时候系统内部会有一个类型推断,然后cast的时候失败了,设置成两边字段名一样,不需要set,就不会cast,导入就可以成功了

  1. No source file in this table

表中没有文件

  1. cause by: SIMPLE authentication is not enabled. Available:[TOKEN, KERBEROS]

kerberos认证失败。klist检查下认证是否过期,并且该账号是否有权限访问源数据

  1. Reason: there is a row couldn’t find a partition. src line: [];

导入的数据在starrocks表中无指定分区

1赞

主键模型表,使用stream load导入数据,导入10条数据成功,导入1.5万条数据失败。
以下是fe.log , txn id 26089。
2022-03-21 01:01:58,852 INFO (thrift-server-pool-16|6450) [DatabaseTransactionMgr.beginTransaction():295] begin transaction: txn id 26089 with label 20220321_010158 from coordinator BE: 10.2.24.33, listner id: -1
2022-03-21 01:01:58,853 INFO (thrift-server-pool-12|2578) [FrontendServiceImpl.streamLoadPut():843] receive stream load put request. db:dsj_dwd, tbl: dws_md_user_log_total_test, txn id: 26089, load id: f44b923a-69d8-aacd-7828-f82f5ed2be90, backend: 10.2.24.33
2022-03-21 01:01:58,890 INFO (thrift-server-pool-6|162) [FrontendServiceImpl.loadTxnCommit():692] receive txn commit request. db: dsj_dwd, tbl: dws_md_user_log_total_test, txn id: 26089, backend: 10.2.24.33
2022-03-21 01:01:58,891 WARN (thrift-server-pool-6|162) [DatabaseTransactionMgr.commitTransaction():517] Failed to commit txn [26089]. Tablet [16568] success replica num is 0 < quorum replica num 2 while error backends 10085,10086,10061
2022-03-21 01:01:58,891 WARN (thrift-server-pool-6|162) [FrontendServiceImpl.loadTxnCommit():706] failed to commit txn: 26089: Failed to commit txn 26089. Tablet [16568] success replica num 0 is less then quorum replica num 2 while error backends 10085,10086,10061
2022-03-21 01:01:58,892 INFO (thrift-server-pool-17|6451) [FrontendServiceImpl.loadTxnRollback():794] receive txn rollback request. db: dsj_dwd, tbl: dws_md_user_log_total_test, txn id: 26089, reason: Failed to commit txn 26089. Tablet [16568] success replica num 0 is less then quorum replica num 2 while error backends 10085,10086,10061, backend: 10.2.24.33

以下为be.log:
grep ‘f44b923a-69d8-aacd-7828-f82f5ed2be90’ be.WARNING ,无内容。

grep ‘f44b923a-69d8-aacd-7828-f82f5ed2be90’ be.INFO
I0321 01:01:58.856958 26789 stream_load_executor.cpp:56] begin to execute job. label=20220321_010158, txn_id=26089, query_id=f44b923a-69d8-aacd-7828-f82f5ed2be90
I0321 01:01:58.857122 26789 plan_fragment_executor.cpp:70] Prepare(): query_id=f44b923a-69d8-aacd-7828-f82f5ed2be90 fragment_instance_id=f44b923a-69d8-aacd-7828-f82f5ed2be91 backend_num=0
I0321 01:01:58.889406 26622 tablet_sink.cpp:865] Closed olap table sink load_id=f44b923a-69d8-aacd-7828-f82f5ed2be90 txn_id=26089, node add batch time(ms)/wait lock time(ms)/num: {10085:(10)(0)(1)} {10086:(8)(0)(1)} {10061:(9)(0)(1)}

be节点状态均正常,查询正常。

你好,确认下be有重启吗

发下这个be.INFO完整的日志

============================33==========================
I0321 10:30:57.710168 26708 tablet_manager.cpp:806] Found 0 expired tablet transactions
I0321 10:30:57.711711 26708 tablet_manager.cpp:831] Reported all 493 tablets info
I0321 10:31:12.169165 26658 load_channel_mgr.cpp:328] Cleaning timed out load channels
I0321 10:31:12.169199 26658 load_channel_mgr.cpp:361] Memory consumption(bytes) limit=4047532167 current=0 peak=437051176

W0321 10:31:16.946296 26758 tablets_channel.cpp:309] Fail to close tablet writer, tablet_id=16568 transaction_id=26114 err=Cancelled: primary key size exceed the limit.

主键模型的表很多吗?主键列多吗?另外请问下是用的哪个版本呢?

2.0.1 ,
字段1 varchar(8)
字段2 varchar(64)
字段3 varchar(64)
字段4 varchar(64)
字段5 varchar(400)

请问下您的be节点内存是多大的,可以curl http://be_host:http_port/metrics|grep be_update_primary_index_bytes_total看下主键索引所占内存。建议减少下主键的数量

Routine Load导入kafka数据时,kafka分区比较多但be节点少,所以需要拆成多个Routine Load任务,这种情况下能否指定消费kafka的分区?

可以指定分区导入,具体见文档create routine load