proxy_ignore_client_abort: need conditional behavior based on response status code

Can proxy_ignore_client_abort be conditional?

No. In standard Nginx distributions, proxy_ignore_client_abort is a binary switch (on or off). It cannot be conditioned on the upstream response status code, content length, or any other runtime attribute of the response.

Technical Explanation

The limitation exists because proxy_ignore_client_abort determines how Nginx handles the connection before the upstream server has sent the response headers. If the directive is off and the client disconnects, Nginx immediately closes the connection to the upstream server to save resources. Because the status code is only known after the upstream begins responding, Nginx cannot use that code to decide whether to ignore the abort in real-time.

Alternative Patterns for Selective Continuation

Since there is no native conditional directive, you can approximate this behavior using the following methods, depending on your build:

• OpenResty / ngx_http_lua_module: If you have the Lua module installed, you can use header_filter_by_lua_block to inspect the status code. While Lua cannot retroactively change the proxy_ignore_client_abort setting for a request already in progress, you can implement complex logic to manage how the response is buffered or passed back to the client.
• Upstream Application Logic: The most reliable way to ensure a process completes regardless of the client's state is to move the long-running task to a background queue (e.g., RabbitMQ, Redis) and return a 202 Accepted status immediately. This decouples the HTTP connection lifecycle from the process execution.
• Proxy Intercept Errors: Using proxy_intercept_errors on; allows Nginx to serve a custom error_page for non-2xx responses. While this changes what the client sees, it does not prevent the upstream connection from being closed if the client aborts and proxy_ignore_client_abort is off.

Verification and Diagnostics

To verify current behavior in your environment, you can use a simple upstream script that logs when it receives a disconnect signal:

1. Set proxy_ignore_client_abort off;
2. Trigger a long-running request and terminate the client (e.g., Ctrl+C in curl).
3. Check upstream logs to confirm the process was terminated.
4. Repeat with proxy_ignore_client_abort on; to confirm the process continues to completion.

Note on Patches: While various community discussions exist regarding adding status-based abort logic, there is no widely adopted or official patch in the mainline Nginx distribution that implements a proxy_ignore_client_abort_status directive.

Missing Diagnostic Detail

To provide a more specific Lua implementation if required, please confirm if you are using a standard Nginx build or an OpenResty distribution.