503 Service Unavailable

HTTP 503 Service Unavailable tells the client that the server is currently unable to handle the request, but that the condition is expected to be temporary rather than permanent. RFC 9110 explicitly ties 503 to the server's capacity or maintenance state, not to anything wrong with the request itself, and recommends the server include a Retry-After header telling the client when to try again, either as a number of seconds or an HTTP date. Unlike 500, which signals an actual fault, a well-formed 503 is often a deliberate, controlled response the server chose to send rather than a symptom of something broken.

Planned maintenance is one of the cleanest uses of 503: many frameworks and CMS platforms, such as WordPress, Rails, and Laravel, ship a maintenance-mode flag that, when enabled, makes every request return 503 with a Retry-After header while a deployment or database migration runs, then automatically clears once the flag is removed. This is preferable to letting requests hit a half-migrated database or a server mid-restart, because it tells clients and search engines explicitly that the outage is intentional and temporary rather than leaving them to infer that from a timeout or a 500.

The other major source of 503s is overload: too many concurrent requests for the available application workers, database connection pool exhaustion, or a load balancer that has no healthy backend to route to because every instance failed its health check at once. Autoscaling groups can make this worse during a traffic spike if new instances take longer to boot and pass health checks than the traffic takes to arrive, producing a wave of 503s until capacity catches up; this is why readiness probes, connection queuing, and gradual traffic ramp-up matter as much as raw instance count.

503 also has a specific meaning for search engines that 500 does not: Google's documentation states that Googlebot treats a 503, ideally with Retry-After, as a signal to come back later without dropping the page from the index, whereas a prolonged run of 500s or an unclear error can eventually be interpreted as the page being gone. This makes 503 the correct status for planned maintenance windows on public sites, since it protects search rankings, but only if it is removed once the site is back; a site accidentally stuck serving 503 for days can still lose ranking despite the status code's intent.