Pay attention to error verbosity & handling; your future self will thank you for it.

  • fail noisily and informatively
  • mission-critical killed process not reporting (error) logs? Hard to manage every failure edge case, especially from external sources? Solution: check for “absence of expected completion” rather than “wait for observable failure.”
  • try to contain (rather than solve) problems
  • in production, 85% of time is spent on error handling, issue reporting & triaging (only 15% of time is spent on code)
    • reducing debugging time by 5% is far more valuable than reducing model dev time by 5%