Increasing the connect timeout using the Serverless Functions plugin

How can I increase the connect timeout to the upstream service using the Serverless (Pre-Function) Plugin?

Using the Serverless (Pre-function) Plugin in Kong, you can override the connect timeout to the upstream service. Below is an example on how this can be achieved:

1. Create a file called function.lua.

2. Content of function.lua:

   
   ngx.ctx.balancer_data.connect_timeout = 100000

   Copied!

   Note: connect_timeout is in milliseconds, so 100000 is 100 seconds. Setting it to a plain 100 (100 milliseconds) would decrease the timeout well below the 60000ms (60s) default, causing Kong to fail faster, not slower — the opposite of what this article is demonstrating.

3. Create a Service (upstream service here will not respond for 10 seconds):

   
   curl -i -X POST --header 'kong-admin-token: <TOKEN>' --url http://localhost:8001/services/ \
   --data 'name=test' \
   --data 'url=http://httpbin.org/delay/10'

   Copied!

4. Create a Route:

   
   curl -i -X POST --header 'kong-admin-token: <TOKEN>' --url http://localhost:8001/services/test/routes \
   --data 'name=testing-route' \
   --data 'paths[]=/without-timeout'

   Copied!

5. Apply the Serverless (Pre-Function) Plugin.

   config.functions is not a valid field on the pre-function/post-function plugins. You must use the phase-specific field instead, in this case config.access, which takes an array of Lua code snippets to run in the access phase:

   
   curl -i -X POST --header 'kong-admin-token: <TOKEN>' --url http://localhost:8001/routes/testing-route/plugins \
   -F "name=pre-function" \
   -F "config.access[1]=@function.lua"

   Copied!

6. Test.

   
   curl -i GET http://localhost:8000/without-timeout

   Copied!

   This request succeeds after Kong’s default 60-second response wait, since httpbin.org/delay/10’s 10-second delay happens after the TCP connection is already established — the delay is server-side response latency, governed by read_timeout/write_timeout, not by connect_timeout. In other words, this particular example does not actually exercise the raised connect_timeout value at all, because connecting to httpbin.org itself is fast regardless of the configured value. connect_timeout only matters (and only fails/succeeds visibly) against an upstream that is slow or unresponsive at the TCP handshake stage — for example, an unroutable/firewalled IP that never completes a TCP handshake. Live-reproduced: pointing a Service at such an address with the default 60000ms connect_timeout makes Kong hang for the full default duration before failing; overriding ngx.ctx.balancer_data.connect_timeout to a small value (e.g. 100 for 100ms) in config.access makes Kong instead fail fast with a 504 Gateway Timeout in a few hundred milliseconds — confirming the override mechanism itself works correctly, even though the httpbin.org/delay/10 example above doesn’t demonstrate it.