Faults that do not reproduce on demand.
WHY THEY ARE HARD
You cannot test a fix if you cannot reproduce the fault.
WHAT CAUSES THEM
Resource limits reached only under load Race conditions with concurrent requests Caching serving inconsistent content A scheduled task interfering periodically External service failures One server in a set behaving differently Time-dependent code
THE APPROACH
Log more. Add logging around the suspected area, including a timestamp and a request identifier, so you can correlate.
Correlate with resource usage graphs. A fault occurring only at peak times is a resource problem.
Correlate with scheduled tasks. A fault every hour at the same minute is a cron job.
Ask affected users for specifics: exact time, what they were doing, which browser, which device.
CHECK CACHING
Inconsistent behaviour between visitors is frequently a caching problem: one visitor served a cached version, another not.
PATIENCE
Intermittent faults take longer. Gather data before theorising, and resist changing several things at once.
