问题
I am trying to write an RDD into a Cassandra table. As shown below TableWriter wrote 0 rows several times and finally writes to Cassandra.
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.171 s.
18/10/22 07:15:50 INFO Executor: Finished task 4.0 in stage 0.0 (TID 4). 622 bytes result sent to driver
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.220 s.
18/10/22 07:15:50 INFO Executor: Finished task 1.0 in stage 0.0 (TID 1). 665 bytes result sent to driver
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.194 s.
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.224 s.
18/10/22 07:15:50 INFO Executor: Finished task 6.0 in stage 0.0 (TID 6). 708 bytes result sent to driver
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.231 s.
18/10/22 07:15:50 INFO Executor: Finished task 5.0 in stage 0.0 (TID 5). 622 bytes result sent to driver
18/10/22 07:15:50 INFO Executor: Finished task 3.0 in stage 0.0 (TID 3). 622 bytes result sent to driver
18/10/22 07:15:50 INFO TableWriter: Wrote 0 rows to log_by_date in 0.246 s.
18/10/22 07:15:50 INFO Executor: Finished task 0.0 in stage 0.0 (TID 0). 708 bytes result sent to driver
18/10/22 07:15:50 INFO TaskSetManager: Finished task 4.0 in stage 0.0 (TID 4) in 418 ms on localhost (executor driver) (1/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 2.0 in stage 0.0 (TID 2) in 433 ms on localhost (executor driver) (2/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 5.0 in stage 0.0 (TID 5) in 426 ms on localhost (executor driver) (3/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 3.0 in stage 0.0 (TID 3) in 433 ms on localhost (executor driver) (4/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 0.0 in stage 0.0 (TID 0) in 456 ms on localhost (executor driver) (5/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 1.0 in stage 0.0 (TID 1) in 436 ms on localhost (executor driver) (6/8)
18/10/22 07:15:50 INFO TaskSetManager: Finished task 6.0 in stage 0.0 (TID 6) in 424 ms on localhost (executor driver) (7/8)
18/10/22 07:15:50 INFO **TableWriter: Wrote 1 rows to log_by_date in 0.342 s.**
Why it is failing to save it sevaral times prior, how to tune it for production?
回答1:
This is not a failure as noted by user10465355. When Spark breaks a job into Tasks it is possible that the work is not evenly distributed or that there isn't enough work for every task to have work to do. This results in some tasks being empty, so when they are processed by the Spark Cassandra Connector they write 0 rows.
For example say;
- You read 100 records into 10 Spark Partitions/Tasks
- You do a filter which eliminate values with a filter so now only 30 records remain in 5 tasks. The other 5 are empty.
- When you write you will now only see records written for 5 tasks, and 5 tasks will report they had no rows written.
来源：https://stackoverflow.com/questions/52928181/why-cassandra-tablewriter-writing-0-records-and-how-to-fix-it