File-based Payment Backend
Last updated
1.2
Last updated
TBD
This table will hold the status of files dispatched
file_name
status
NEW/DISPATCHED
This table will hold the status of files received from bank.
file_name
status
NEW/PROCESSED
This block will be like a cronjob that runs periodically and does the following
Read all the payment instructions that have not been processed yet (status: NEW) via direct query of DB
Query ID Account Mapper if translation from ID to bank account information is required.
Create an output file specific to the format accepted by a bank.
Write the file in the File Storage
Update the status of all processed instructions as FILED.
This functional block may be implemented as a script that is run as cronjob. The features will be implemented:
Unique file names -- could be composed of the batch ID with a few more fields or it could be completely UUID based.
Full in-memory processing
Process data in a batch size as expected by the bank, provided the data can be accommodated in the RAM. (This should be fine because typically banks will not expect very large files. The number of rows may be in the order of a few thousand).
Entire processing and updates in DB should happen atomically. In case there is a failure, the script should be able to discard the incomplete file and start all over again.
The batch size should be configurable
The file format will be specific to a bank. The script should be designed such that several file formats can be configured without having to make major changes in the code.
This will typically be large S3-compliant storage (like MinIO or AWS S3) to store all the payment instruction files. It will have two buckets:
Outgoing: For outgoing payment files and another for received payment files.
Incoming: For files fetched from bank(s)
The File Dispatcher is like a cronjob that does the following:
Picks an unprocessed (un-sent) file from the bucket and using bank interface transfers the file to the bank. This could be a mechanism like SFTP. This would be very bank-specific.
Features:
The implementation may be done via script called periodically.
The file dispatch mechanism will be specific to a bank. The script should be designed such that several dispatch mechanisms can be configured without having to make major changes in the code
The File Fetcher is like a cronjob that does the following:
Fetch fresh files form bank's systems and stores them in Incoming bucket
Features:
The implementation may be done via script called periodically.
The file dispatch mechanism will be specific to a bank. The script should be designed such that several fetch mechanisms can be configured without having to make major changes in the code
This block will be like a cronjob that runs periodically and does the following
Parse the file to read the status of each transaction. Translate the status to a common standard code if required.
Update the status in DB as PAID or FAILED.
Update error code and error messages
This functional block may be implemented as a script that is run as cronjob. The features will be implemented:
Full in-memory processing
Entire processing and updates in DB should happen atomically. In case there is a failure, the script should be able to discard the incomplete process and start all over again.
G2P payments are typically not real-time and can happen over several hours or even days. Assuming that SSD disks are used for DB and file storage, and all processing happens in-memory it should be possible to read a batch of 10, 000 to 100,000 from DB, process it, and write into a file in a few seconds (hypothesis needs to be tested). Thus_,_ several million transactions processing should be possible in a day. A barebone calculation: if we assume 10,000 records can be read from DB, processed and written in a file in 10 seconds, then 3.6 million records can be processed in 1 hour.
The response time of ID Account Mapper also needs to be considered. The query to ID Account Mapper should be in Sync mode to keep the design simple.
The Payment File Creator, Payment File Reader, File Dispatcher, and File Fetcher must be designed such that they can run on independent CPUs.
The DB must be suitably indexed for fast reads and updates.
With this capacity, a single instance of each of the above would suffice. There may not be a need to have multiple instances of these blocks running in parallel as it will significantly increase the complexity of the design.
TBD
TBD
TBD
Update status in the table as NEW.
Configuration for bucket ID, location, and credentials for accessing .
Update status in table in the DB as DISPATCHED.
Update status in table in the DB as NEW.
Read an unprocessed file from File Store -> Incoming bucket. The list of unprocessed files is obtained by reading the table.
Update the status in the table as PROCESSED.
Configuration for bucket ID, location, and credentials for accessing .
