Running multiple jobs using one config file

We often have this kind of use case:

• Read src_accounts.csv file and write to accounts table on PostgreSQL
• Read src_payments.csv file and write to payments table on PostgreSQL
• Read src_requests.csv file and write to requests table on PostgreSQL

In this case, we need multiple configuration files and multiple embulk run commands. Idea here is to add a simple multi-transaction feature to embulk.

For example, the configuration file can be like this:

jobs:
    accounts:
    payments:
    requests:
in:
    type: file
    path_prefix: src_${job}
    parser:
        type: csv
out:
    type: postgresql
    table: table_${job}

Discussion points are guess, transaction, next config diff and resume.

As the result of guess, we can output this config file:

jobs:
    accounts:
        in:
            parser:
                type: csv
                columns:
                    - {name: id, type: long}
                    - {name: surename, type: string}
                    - {name: givenname, type: string}
    payments:
        in:
            parser:
                type: csv
                columns:
                    - {name: id, type: long}
                    - {name: timestamp, type: timestamp, format: format: '%Y-%m-%d %H:%M:%S.%N'}
    requests:
        in:
            parser:
                type: csv
                columns:
                    - {name: request_id, type: string}
                    - {name: account_id, type: long}
                    - {name: timestamp, type: timestamp, format: format: '%Y-%m-%d %H:%M:%S.%N'}
in:
    type: file
    path_prefix: src_${job}
    parser:
        type: csv
out:
    type: postgresql
    table: table_${job}

About transaction, an idea is to commit jobs one by one when all jobs are completed. For example, embulk runs transactions in this order:

1. job accounts begins (1 task)
2. job payments begins (2 tasks)
3. job requests begins (2 tasks)
4. job accounts task-1 completes
5. job payments task-1 completes
6. job requests task-1 completes
7. job payments task-2 completes
8. job requests task-2 completes
9. job requests commits
10. job payments commits
11. job accounts commits

A concern is that requests and payments won't be rolled back when commit of job accounts fails.

DSL (https://github.com/embulk/embulk/issues/131) does similar thing. Any thoughts?