Logging sucks (loggingsucks.com)

Article summary

The article discusses the limitations of traditional logging practices, which were designed for a different era of monolithic systems and single servers. It argues that modern systems, with multiple services and distributed architectures, require a new approach to logging, known as wide events or canonical log lines, which capture all relevant context for a request in a single log event. This approach allows for more effective debugging and analytics. The article also discusses the importance of structured logging, high-cardinality data, and sampling to manage costs.

Main themes

• Logging practices
• Distributed systems
• Wide events
• Structured logging
• Observability
• System architecture

What commenters say

• The concept of observability and wide events was not invented by a single person, but rather has been a topic of discussion in the industry for a long time.
• Some commenters argue that the complexity of modern systems is a result of poor architecture, while others see it as a necessary consequence of scale and complexity.
• There is disagreement about the effectiveness of microservices architecture, with some arguing it is overcomplicated and others seeing it as a necessary approach to manage complexity.
• Some commenters emphasize the importance of questioning assumptions and simplifying systems, while others argue that complexity is unavoidable in large-scale systems.
• There are different opinions on what constitutes sensitive data, with some considering customer lifetime value to be sensitive and others seeing it as important information for prioritization and understanding customer value.
• Some commenters argue that traditional logging practices are still effective for local processes, but inadequate for understanding distributed systems, which require transaction tracing and other tools to provide context.