Looking for performance feedback on NetBox v4.5.2

[jeremystretch]

I'm happy to announce that NetBox v4.5.2 was just released. It includes numerous performance improvements resulting from an intentional spike in performance-focused work over the past couple weeks. (You may have noticed our new issue type.)

These improvements include:

• Caching object types and custom fields for each model for the duration of a request. This avoids redundant SQL queries without breaking custom objects functionality. (#21259, #21300)
• Being more clever about serializing object change data (#21260, #21303)
• Prefetching related objects for the response when creating/updating objects via the REST API (#21263)
• Other miscellaneous improvements

In my (admittedly limited) local testing, I've observed pretty significant reductions in response times. For instance, a REST API request to create 100 IP addresses assigned to interfaces took around 2400ms on v4.5.1, but only ~930ms on v4.5.2: a ~60% reduction. (Bear in mind this is on a development instance with debugging enabled.) Various other testing suggested similarly substantial improvements.

Of course, lab benchmarks are only so useful. I'm interested to see any real-world metrics people have comparing v4.5.2 performance to previous releases. If you'd like to share, please remember to include:

• The type(s) of request being made
• How the requests are being generated (i.e. curl, pynetbox, Terraform, etc.)
• An estimate of the number of objects involved
• The previous version of NetBox to which you're comparing
• Any other data that might be relevant for someone to run a local comparison

While this release represents a renewed focus on application performance, we're only just getting started. I believe there are still many opportunities to improve performance generally without sacrificing functionality or data integrity. This will remain an area of focus as we build toward NetBox v4.6.