The database which is the value of the event tracing property event_trace_rdb_url will automatically create two tables JOB_EXECUTION_LOG and JOB_STATUS_TRACE_LOG and several indexes.
| Column name | Column type | Required | Describe |
|---|---|---|---|
| id | VARCHAR(40) | Yes | Primary key |
| job_name | VARCHAR(100) | Yes | Job name |
| task_id | VARCHAR(1000) | Yes | Task name, create new tasks every time the job runs. |
| hostname | VARCHAR(255) | Yes | Hostname |
| ip | VARCHAR(50) | Yes | IP |
| sharding_item | INT | Yes | Sharding item |
| execution_source | VARCHAR(20) | Yes | Source of job execution. The value options are NORMAL_TRIGGER, MISFIRE, FAILOVER. |
| failure_cause | VARCHAR(2000) | No | The reason for execution failure |
| is_success | BIT | Yes | Execute successfully or not |
| start_time | TIMESTAMP | Yes | Job start time |
| complete_time | TIMESTAMP | No | Job end time |
JOB_EXECUTION_LOG records the execution history of each job.
There are two steps:
JOB_EXECUTION_LOG, and all fields except failure_cause and complete_time are not empty.is_success, complete_time and failure_cause(if the job execution fails).| Column name | Column type | Required | Describe |
|---|---|---|---|
| id | VARCHAR(40) | Yes | Primary key |
| job_name | VARCHAR(100) | Yes | Job name |
| original_task_id | VARCHAR(1000) | Yes | Original task name |
| task_id | VARCHAR(1000) | Yes | Task name |
| slave_id | VARCHAR(1000) | Yes | Server’s name of executing the job. The valve is server’s IP. |
| execution_type | VARCHAR(20) | Yes | Type of job execution, the value options are NORMAL_TRIGGER, MISFIRE, FAILOVER. |
| sharding_item | VARCHAR(255) | Yes | Collection of sharding item, multiple sharding items are separated by commas. |
| state | VARCHAR(20) | Yes | State of job execution, the value options are TASK_STAGING, TASK_RUNNING, TASK_FINISHED, TASK_KILLED, TASK_LOST, TASK_FAILED, TASK_ERROR. |
| message | VARCHAR(2000) | Yes | Message |
| creation_time | TIMESTAMP | Yes | Create time |
JOB_STATUS_TRACE_LOG record the job status changes.
Through the task_id of each job, user can query the life cycle and running track of the job status change.