503 Service Unavailable is an HTTP status code indicating that the server is temporarily unable to handle a request. Visitors should avoid repeated refreshes and submissions. Before restarting, operators should correlate the time, scope, recent changes, logs, resources and dependencies.
503 means temporarily unavailable. Identify the source and start with reversible recovery actions.
Common causes include traffic spikes, exhausted CPU, memory or connections, planned maintenance, application or database outages, and failed health checks after deployment. Before changing settings, determine whether the response comes from the CDN, proxy, application or database layer.
| Meaning | The service is temporarily unable to handle a request. |
|---|---|
| Common causes | Overload, exhausted resources, maintenance, upstream or database outages, and configuration errors. |
| For visitors | Avoid repeated refreshes; wait and retry once after checking whether the request completed. |
| For operators | Record the scope and time, then review changes, logs, monitoring and dependencies. |
| SEO | A correct response can signal a temporary outage; avoid leaving it in place for a prolonged period. |
| Reviewed | July 29, 2026 |
What 503 Service Unavailable means
RFC 9110 defines a 503 response as the server being temporarily unable to handle a request because of overload or scheduled maintenance. If the server can suggest when to try again, it may send a Retry-After header.
A 503 does not necessarily mean that the server itself has failed. Between the browser and application may be a CDN, WAF, load balancer, reverse proxy, container and database. Correlate response headers, logs and monitoring to determine which layer generated the response.
503 describes the result, not the underlying cause.
Worker exhaustion, depleted database connections, an unhealthy backend and a CDN block call for different responses. Identify the source before changing settings.
How 500, 502, 503, 504 and 429 differ
5xx codes indicate server errors, while 429 indicates rate limiting. Use the table for initial triage; confirm the cause from the configuration and logs.
| Code | Meaning | Typical situation | Where to look first |
|---|---|---|---|
| 500 | Internal Server Error | An unexpected failure while handling a request, such as an exception, configuration error or permission failure. | Application and web server error logs, and recent changes. |
| 502 | Bad Gateway | A proxy received an invalid or incomplete response from an upstream server. | Reverse proxy, upstream application, connections and certificates. |
| 503 | Service Unavailable | Temporary overload, maintenance or no healthy backend. | Load, workers, health checks and dependencies. |
| 504 | Gateway Timeout | A proxy did not receive an upstream response in time. | Upstream processing time, timeouts and database or API latency. |
| 429 | Too Many Requests | A client or API key exceeded a configured rate limit. | WAF, API gateway and rate-limiting rules. |
Avoid repeating payment, booking or form requests
An upstream service may have completed a request even when the screen shows an error. Check your history or confirmation email before retrying to avoid duplicate orders or submissions.
Common causes of 503 errors
Investigate the request path in layers: entry points, application, data, external dependencies and delivery, instead of treating every incident as a slow server.
Traffic spikes and concurrency limits
Campaigns, viral traffic, crawlers or abusive requests can exhaust workers or connections and prevent new work from being accepted.
Exhausted CPU, memory, disk or connection pools
Check memory, file descriptors, PHP workers, database connections and queue depth as well as CPU.
Database, cache or external API outages
An application may be running but still return 503 when it cannot reach its database, cache, payment API or identity provider.
Deployment, configuration or health-check mismatches
Changes to environment variables, ports, certificates, routing or readiness conditions can make new instances unhealthy.
Scheduled maintenance
Some systems intentionally return 503 during maintenance. Check that the maintenance page is not being served with a 200 status.
CDN, WAF or load balancer controls
Rate limits, bot protection, an unreachable origin or no healthy backends can cause a layer in front of the application to generate a 503.
WordPress or PHP backlogs
Heavy plugins, scheduled tasks, backups, external calls, too few PHP-FPM workers and slow database queries can combine to produce 503 responses.
Hosting or regional incidents
If multiple sites or networks are affected without a recent change, check your provider's status page and support channels.
What visitors can do
A 503 is usually generated on the service side. Check the situation and avoid duplicate transactions before changing settings on your device.
-
Wait, then refresh once
Retry-Afterindicates when to retry. Wait as instructed, and avoid repeated refreshes that can add load. -
Check official incident and maintenance notices
Look at the service's status page, website or official channels. A widespread incident cannot be fixed by changing your device.
-
Check the result of payments, bookings and form submissions first
Review purchase or booking history and confirmation emails. Do not resubmit completed requests; contact the operator if the result is unclear.
-
If only you are affected, try another network
Different results on Wi-Fi and mobile data or in a private window may point to CDN, WAF or network-path differences.
Clearing the cache is not the first priority
A 503 is usually a server-side response. A cached error page is possible, but first wait and check official notices; resetting device settings or reinstalling an app is not normally necessary.
Safe recovery steps for operators
Under pressure it is tempting to change several settings at once. Preserve the evidence and work through observation, diagnosis, reversible mitigation and verification in order.
-
Record the scope and first detection time
Record affected URLs, methods, devices, networks, regions, status codes and first detection. Determine whether the issue also affects the admin interface or APIs.
-
Pause changes and review recent deployments and maintenance
Pause additional deployments, configuration changes, load tests and batch jobs. Compare the first error time with releases and certificate, environment-variable and DNS changes.
-
Identify the layer returning 503
Inspect Server and Via headers, CDN and request IDs, and error-page content. A browser message alone does not establish an origin outage.
-
Correlate logs and metrics on the same timeline
Compare access, web server, application, load balancer, database and external API logs. Review memory, connections, queues, workers, error rates and latency as well as CPU.
-
Make one controlled change at a time
Apply a recorded, reversible action such as an approved rollback, traffic limit, pause of heavy jobs, static cache or switch to healthy instances.
-
Confirm stability before declaring recovery
Monitor successful responses on important URLs, error rates, p95 and p99 latency, database success rates and queue depth over a suitable period. One successful 200 is not enough.
First check response headers and timing
A browser display alone may not reveal the status or the source. Fetch headers with a GET request and compare URLs and networks.
terminal
# Show response headers without saving the body
curl -sS -D - -o /dev/null https://example.com/
# Inspect status, total time and remote IP
curl -sS -o /dev/null \
-w 'status=%{http_code} total=%{time_total}s remote=%{remote_ip}\n' \
https://example.com/
curl -I sends a HEAD request, which may behave differently from GET. To approximate a normal visit, keep GET and discard only the response body as shown above.
Example: returning Retry-After
For planned maintenance with a useful recovery estimate, send a delay in seconds or an HTTP-date. Follow the official documentation for your server or CDN.
http response
HTTP/1.1 503 Service Unavailable
Retry-After: 300
Cache-Control: no-store
Content-Type: text/html; charset=UTF-8
Treat automatic retries according to the method
Do not retry idempotent operations such as GET in the same way as state-changing POST or PATCH requests. Bound the number and duration of retries and design upstream failover to prevent duplicate processing.
What to check in each environment
Investigation differs for WordPress, reverse proxies, CDNs and containers. Start with the components used in your configuration.
WordPress Shared hosting and PHP
- Provider incident notices and CPU, memory and concurrency limits.
- PHP-FPM workers, PHP error logs and web server error logs.
- Recently updated plugins, themes and PHP versions.
- Heavy work such as backups, image processing, wp-cron and external API calls.
- Database connections, slow queries, options table growth and cache state.
- Maintenance mode or temporary files left after a failed update.
NGINX / Apache Reverse proxy
- Upstream connection refusals, timeouts, invalid responses and healthy backend count.
- Proxy and FastCGI destinations, ports, UNIX sockets and permissions.
- Worker connections, file descriptors and keepalive settings.
- Failover conditions and retry limits; do not retry non-idempotent requests without bounds.
- Whether an error page is incorrectly returned with status 200.
CDN / WAF Delivery and protection
- CDN status, origin reachability, certificates and DNS.
- Whether WAF rules, bot protection or rate limits block legitimate users or Googlebot.
- Whether identifying headers show that the CDN or origin generated the error.
- The lifetime of cached 503 responses and what to purge after recovery.
- Load balancer health-check URLs, expected status codes and timeouts.
Containers / serverless Cloud environments
- Healthy instance count, readiness, liveness and startup failures.
- Maximum instances, concurrency, CPU and memory limits, and autoscaling delay.
- Listening port, environment variables, secrets and IAM.
- Traffic split between revisions and rollback availability.
- Database connection growth and connection spikes during scale-out.
How 503 errors affect SEO
Returning the correct 503 during brief planned maintenance is an appropriate way to signal a temporary outage. It is more accurate than returning 200 with an error page or 404 for a temporary interruption.
Google says its crawlers temporarily reduce crawling when they encounter 5xx or 429 responses. Previously indexed URLs are initially retained, but persistent errors may eventually cause removal. Crawling can gradually recover after successful responses return.
SEO checks during maintenance
- Return HTTP 503 while the service is temporarily unavailable.
- Include Retry-After when a useful recovery estimate is available.
- Do not return an error page with status 200.
- Do not leave the entire site returning 503 for an extended period.
- After recovery, check that important URLs return 2xx.
- Use Search Console and server logs to check that 5xx errors are decreasing.
SEO checks during a 503 incident
- Can crawlers fetch robots.txt, XML sitemaps, CSS and JavaScript as well as the homepage?
- Are WAF or rate-limiting rules accidentally blocking Googlebot?
- What status does the same URL return from the CDN and the origin?
- Are you returning the correct 503 instead of relying on noindex on an error page?
- After recovery, have you checked URL Inspection and the page indexing report?
Preventing future 503 errors
Prevention requires more than a larger server. Combine early detection of incoming load, backlogs, dependency outages and failed deployments with measures that limit impact.
Monitor saturation
Set thresholds for memory, connections, workers, queues, database pools, error rates and p95 and p99 latency as well as CPU.
Smooth traffic
Use CDNs, caches, queues, asynchronous work, rate controls and staggered batch schedules to absorb spikes.
Scale safely
Plan autoscaling limits, startup time and database connection limits together so added application instances do not exhaust the database.
Roll out changes gradually
Use canary releases, health checks, rollback and database migration compatibility checks to avoid simultaneous failure across instances.
Maintain a recovery runbook
Document contacts, logs, decision criteria, rollback, notices and verification so a new operator can follow the same process.
Prevent duplicate execution
Use idempotency keys or duplicate checks for payment, booking and registration APIs so retries after 503 do not repeat a transaction.
Recovery checklist
Before closing the incident, check the items below. Verify sustained service stability rather than relying on one successful page load.
- Affected URLs, APIs and the admin interface consistently return successful responses.
- 5xx and timeout rates have returned to normal.
- p95 and p99 response times are no longer elevated.
- CPU, memory, connections, queues and workers have spare capacity.
- Database, cache and external API success rates have recovered.
- Healthy backend counts have recovered for the CDN and load balancer.
- Payments, bookings and forms have not been processed twice.
- The maintenance page and temporary Retry-After header have been removed.
- Search Console and monitoring show no continuing growth in 5xx errors.
- The cause, decisions, changes, recovery time and preventive actions are recorded.
Frequently asked questions about 503 errors
Will a 503 error go away on its own?
A temporary traffic spike or planned maintenance may resolve on its own. Repeated errors can indicate unresolved resource, dependency or configuration problems; operators should inspect logs and metrics.
How does 503 differ from 502 and 504?
503 means a service is temporarily unable to handle a request. 502 means a gateway received an invalid upstream response, while 504 means it did not receive an upstream response in time.
Will clearing my browser cache fix it?
A 503 usually originates at the server, CDN or proxy, so clearing the cache alone often will not help. Check official notices and wait before retrying once.
Can a 503 error affect SEO?
A correct 503 during brief maintenance is appropriate. Persistent 5xx responses can reduce crawling, and URLs that continue returning errors may eventually be removed from the index.
What should Retry-After contain?
Use a recovery time as an HTTP-date or a non-negative integer number of seconds. It is useful when a meaningful estimate is available, such as during planned maintenance.
Will restarting the server fix a 503 error?
A restart may restore service temporarily, but it can also erase logs or connection evidence. Review changes, logs, resources and dependencies first, and use a documented, reversible recovery procedure.
Official documentation and references
Use the RFC for HTTP semantics and the official documentation for your services and web server for crawler behavior and implementation details.
- RFC 9110: 503 Service Unavailable
- RFC 9110: Retry-After
- Google: How HTTP Status Codes Affect Google's Crawlers
- Google Search Central: CDNs and crawling
- NGINX: ngx_http_proxy_module
- Apache HTTP Server: mod_proxy
This article summarizes general procedures using HTTP specifications and official documentation available on July 29, 2026. Follow your hosting, cloud, CDN and web server documentation and backup policies for actual recovery operations.