Phantom Friday checkout crashes traced to forgotten canary deployment instance
A checkout service began throwing intermittent server errors most Friday afternoons, with all symptoms disappearing by Monday, making the bug extremely difficult to reproduce or diagnose. Initial investigations ruled out long-running queries, connection leaks in a newly shipped reporting endpoint, and rogue scheduled jobs as potential causes. The real culprit turned out to be a forgotten third service instance left over from a canary deployment experiment months earlier, which was no longer receiving live traffic and therefore invisible in standard metrics. This dormant instance had a faulty health-check routine that opened a fresh database connection every few seconds without reusing the pool, steadily exhausting the connection limit. An auto-scaling policy restarted the instance every Monday, resetting the leak and explaining the precise weekly pattern engineers had struggled to pin down.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in