How do I add X-Request-Start in HAProxy?

Question

We'd like to track request queue times, and as per https://docs.newrelic.com/docs/apm/other-features/request-queueing/configuring-request-queue-reporting, we need to add X-Request-Start or X-Queue-Start with the timestamp in milliseconds.

Koziara

Answer 1

Using %Ts is NOT the correct solution as it is the start time of the session (stream) and not the request time. Multiple requests will use the same %Ts time. See more details here

The correct solution for haproxy version <1.9 is to use the following in the frontend block. Unfortunately it only has seconds resolution.

http-request set-header X-Request-Start t=%[date()]

If you are using haproxy version >=1.9 there is a new sample fetch method date_us() which can be used to get microseconds resolution.

http-request set-header X-Request-Start t=%[date()]%[date_us()]

Note: The newrelic agents can automatically handle timestamps values sent using seconds, milliseconds, or as microseconds.

Polycotyledon

the link for 'See more details here' is not active. Google cache is also not present. Do you have any other link for the explanation? – Forceps

From what I recall: %Ts timestamp is when the TCP connection is accepted, but HTTP (+ HTTP/2) requests can be pipelined or use keepalive. eg: Browser with a 10 second keepalive may only send a request after a few seconds, but the %Ts send to the backend will be the initial time of the connection was accepted, which is wrong. – Polycotyledon

Answer 2

The solution is to add this line in your frontend block. You'll need one for HTTP and HTTPS.

http-request set-header X-Request-Start t=%Ts%ms

Koziara

This is not the right solution, see https://mcmap.net/q/1607549/-how-do-i-add-x-request-start-in-haproxy for the details. – Polycotyledon