I’ve been working on the Redshift Serverless PDF for the last two or three weeks.
The reason it’s taking so long is firstly I’m with a client, but secondly, boto3, which is normally difficult to work with, is particularly difficult to work with, for Redshift Serverless.
For example, if you delete a workgroup, and then you poll until the workgroup is not listed any more, you still cannot delete namespace the workgroup was using.
So you discover all these issues, one after the other, and it takes forever, because workgroup spin up/down times are lengthy.
I’m currently running what I hope will be the final development run of the test code.
I’m missing one potential test, which is to try to provoke a CSC-serviced query.
On a manual WLM cluster, I found yesterday (as part of experimenting
for the Serverless PDF - I otherwise have almost no knowledge of
experience of CSC can easily and directly induce CSC use - configure WLM
to be one queue with one slot, issue a long-running query, and then
issue a select. The select will go to CSC.
Workgroups I think are using AutoWLM, and I’ve no personal experience of AutoWLM, but given the latest rumours about how it works, provoking CSC could be more work than I have time for today - I have some other to get done this afternoon, and it’s 1pm now and I need three hours today for gym and errands, too.
Actually, I think, it’s more that on an emotional level, getting boto3 working with Redshift Serverless has been so enervating that I can’t face another substantial Redshift task right now.
I’d like to do it, but also I don’t feel it’s necessary. There’s a ton of other evidence to support the conclusion of the PDF.