Oh FFS! Someone's going to get canned for this.

[DS23]

maybe they turned the history off for an entirely valid reason such as to halt the flood of error messages...

[MarillionFan]

I would have thought you would have been able to pick up the changes in the dimension. This is the first thing I would be checking week on week as it is vital to historical reporting. I am assuming you mean by turned off SCDs that the dimension was being updated directly rather than a new row being inserted with previous rows having their end date set (SCD type 1 instead of type 2)? I can see this being the fault across the board - from the developer all the way to management.

Correct. All dimensions were updating properly except one. One we are not using for any reporting at present as we had no output requirements but once populated would yield results.

News in. The offshore developer who wrote the ETL code has just been taken off the account by the consultancy! What a surprise!!!

[NorthWestPerm2Contr]

Correct. All dimensions were updating properly except one. One we are not using for any reporting at present as we had no output requirements but once populated would yield results.

News in. The offshore developer who wrote the ETL code has just been taken off the account by the consultancy! What a surprise!!!

That does not seem right - there should have been some testing in place to validate the work of the developer - especially since he is offshore.... Again sounds like several people could be at fault, especially testers....

[minestrone]

That does not seem right - there should have been some testing in place to validate the work of the developer - especially since he is offshore.... Again sounds like several people could be at fault, especially testers....

DBA I think, testers sign off on product prior to release and this is a post release issue. I would instruct a DBA to check the tables every day and if the audit tables were not changing then they should have flagged.

[MarillionFan]

That does not seem right - there should have been some testing in place to validate the work of the developer - especially since he is offshore.... Again sounds like several people could be at fault, especially testers....

The breakdown has been in unit testing. The offshore team has not been testing their own work, preferring to deploy and through over to me. Already said it was unacceptable and the quality was poor.

Going to have to have that revisited.

[NorthWestPerm2Contr]

DBA I think, testers sign off on product prior to release and this is a post release issue. I would instruct a DBA to check the tables every day and if the audit tables were not changing then they should have flagged.

Definitely some Kimball subsystems were not in place. Perhaps a design flaw? MF to blame!

[Sysman]

I would instruct a DBA to check the tables every day and if the audit tables were not changing then they should have flagged.

That's certainly a way forward from this point. Horse already bolted and all that.

[minestrone]

That's certainly a way forward from this point. Horse already bolted and all that.

I would have turned on the functionality for a couple of days, tell them it will take a few days to get the data as resources are pushed, use the data from post release and the data from the last couple of days and write a script that matches up the missing dates with improvised data, randomise it using cos but keep a general extrapolation of the range.

[MarillionFan]

Definitely some Kimball subsystems were not in place. Perhaps a design flaw? MF to blame!

It's an odd contract this one.

I came in and was asked to design a data model and follow Kimball. Did that. Left.

Four months later, got a call. Came back in. They'd offshored developed the model, ETL and handed over to their support organization. Could I now define the framework model (have someone else develop it), and the output (have yet another person develop it), test it, manage the client(not the development) and then get it rolled out, before being told to bugger off.

Client sees everything as a production line. They will literally have a contractor come in for five days to do a piece of work(like ETL), not test it properly internally, through it at the client to test, let the developer go, wait a few weeks and when it doesnt work go back out to see if the contractor is available or get another.

I am under strict instructions not to do any development.

Doomed.

[MarillionFan]

I would have turned on the functionality for a couple of days, tell them it will take a few days to get the data as resources are pushed, use the data from post release and the data from the last couple of days and write a script that matches up the missing dates with improvised data, randomise it using cos but keep a general extrapolation of the range.

(If we choose to, it looks like we can recover the whole model from the landing tables)

The simple truth is, someone ('now removed') turned off the updating after a week.

Bearing in mind, last week on a deployment to production(after signoff), I logged in to find that one of the developers had left a truncate table on one of the updates.