Re: Assigning account numbers, intuiting open and close dates

I think what you are missing is a viable workflow that accounts for errors, missing information, etc. My approach would be,

Grab all data and attempt to import it into a temporary table without no keys and three additional fields: 1) a flag that says 'processed|unprocessed|failed' that defaults to 'unprocessed' and 2) a unique account id 3) Long error message
Run a utility (who cares what it is written in -- it can be a stored proc if you prefer) that scans the prospective record and fills in the blanks -- the account id. If no suitable account id can be produced flag the record as 'failed' and update the long error message field with the reason. If it passes move the record to its final destination.
Finally, log every transaction with a timestamp when it was processed, how long it took to run, etc.

You'll also need a hardfile log to capture instances when the database is unavailable or if somebody changes permissions on the tables so as to make them unwriteable/unreadable/etc.

Celebrate Intellectual Diversity

Comment on Re: Assigning account numbers, intuiting open and close dates

Replies are listed 'Best First'.
Re^2: Assigning account numbers, intuiting open and close dates by Voronich (Hermit) on Apr 09, 2014 at 19:02 UTC
Well yeah. A lot of that I just left out because it's not particularly germane to the problem I'm trying to solve. I've got standard process boilerplate surrounding this thing. processed/unprocessed doesn't have sustained value over time, as a matching run will produce simple deterministic results and there will either be data or not. Not means failed, and processed is a function of a monthly run being done or not. So there isn't sufficient justification to add additional columns to denote purely derivative information. Only collect data that can conceivably be useful. I have overall process timing and logging in my boilerplate perl. But most of that stuff isn't really useful.	[reply]

Replies are listed 'Best First'.

Re^2: Assigning account numbers, intuiting open and close dates
by Voronich (Hermit) on Apr 09, 2014 at 19:02 UTC

Well yeah. A lot of that I just left out because it's not particularly germane to the problem I'm trying to solve. I've got standard process boilerplate surrounding this thing.

processed/unprocessed doesn't have sustained value over time, as a matching run will produce simple deterministic results and there will either be data or not. Not means failed, and processed is a function of a monthly run being done or not. So there isn't sufficient justification to add additional columns to denote purely derivative information.

Only collect data that can conceivably be useful. I have overall process timing and logging in my boilerplate perl. But most of that stuff isn't really useful.

[reply]