Hacker News Reader

Comment by mostlyjason | Hacker News Reader

Imagine customer support where you really need a record of the individual transactions for that particular customer. A monitoring solution won't give you that level of detail.

View on HN

Individual transactions aren't found in application logs, they are found in the data store. Also, again, it's important to stress that central logging should be available, just not on all the time. If it's a specific sequence of events that triggers a problem, turn on logging and then ask them to do the sequence again. If they don't need to do the sequence again, then it wasn't really of consequence. And if it happens for multiple users, then customer service should be able to turn on let the developers know to try and find the cause which they can do by adding more metrics or perhaps some temporary central logging.

A better compromise would be always having logging on but only retaining very recent logs. You need the infrastructure to support logging and correlation anyway and if you have a short enough retention period you don't have to worry about dealing with growing infrastructure just to support it.

Yeah I think most people store them for a limited time, such as a week or a month. Most issues that pop up will only be relevant that long. You can have permanent archives for individual data since storage is cheap, or even summary metrics. Better to start with ground truth because you often don't know what data you'll need in hindsight.

I think that's the difference between a log and an audit trail.

Parent makes a good point imo. Systems are now sufficiently complex that logging everything is a useless indicator. And if you're only using a subset of your logging anyway (by generating metrics), why collect anything more before you know there's a problem.

Tl;dr: avoid premature log verbosity

Reply on news.ycombinator.com