3,971 karma · joined October 25, 2011
> For a web service, that usually means one log event at the end of every request.
The point of a log, as I have seen it used, is to understand what's going on during a request. Yes, some of that is timing / success / failure information, but you also need:
- Logical tracing - at a high level, what decisions did we make (authz, business logic) when serving this request
- Partial failures or warnings - the request may have succeeded, but we may have failed or timed out a call to a dependent service and delivered a partially degraded response as a result
- Tracepoints to support active development - how often do we hit a certain codepath (to help developers roll out or deprecate features)
It's useful to distinguish between "context" (things we know about a request), timing, success, and plain old logging (using human language to describe to an operator what the system is doing). Context should be included with every line - I strongly agree with OP's proposal of wide events, don't make operators ever JOIN if you can avoid it - but it won't always be the same because we learn things about the request as we process it. Yes, if you wait until the end of a request, you will have maximal context, but at the loss of fine-grained detail during processing.
Unfortunately, the US and many other countries have chosen the other path (sporadic enforcement with severe punishment) largely because it's easier to implement. There's a lot of momentum to change this but it's politically difficult at least in America.
And presumably they wouldn’t be shy about telling us if they had.
Also HN: No, not like that
https://www.explore.com/1804742/not-divine-story-miracle-fli...
sleep 0.001 # Delay to allow the receiver to process the signal1. The "lethal trifecta" is also the "productive trifecta" - people want to be able to use LLMs to operate in this space since that's where much of the value is; using private / proprietary data to interact with (do I/O with) the real world.
2. I worry that there will soon be (if not already) a fourth leg to the stool - latent malicious training within the LLMs themselves. I know the AI labs are working on this, but trying to ferret out Manchurian Candidates embedded within LLMs may very well be the greatest security challenge of the next few decades.
Whether or not those changes actually change the "character" of the company is a different question (IMHO Newsmax is morally defunct and cannot be saved) but no company anywhere would just shrug something like this off as "the cost of doing business".