That does seem quite mysterious…

Maybe add an IO.inspect(stripe_event) after this or wrap the assignment like so

conn
|> IO.inspect(label: "conn before when signature OK")
|> Plug.Conn.assign(:stripe_event, stripe_event)
|> IO.inspect(label: "conn after when signature OK")

to determine if there’s anything unexpected going on with the conn that do not reach the function in the controller.

The request could also potentially be reaching the controller and just not matching on webhooks_payment if there’s no catchall function head e.g.
def webhooks_payment(conn, params), do: IO.inspect(conn, label: "conn from catchall webhooks_payments")

If you followed Stripe Webhooks in Phoenix with Elixir Pattern Matching | Connor Fritz (which it kind of looks like from your function signatures) , I might know the fix . I ran into the same problem when referencing said article .

In Connor’s code , I believe the offending lines are :

{:ok, body, _} = Plug.Conn.read_body(conn)
conn

The conn returned is ‘stale’ in a sense . So I changed this to

{:ok, body, conn} = Plug.Conn.read_body(conn)
conn

and the problem seems to have gone away . I haven’t done any further digging into exactly why this is the case (the exact difference between the conn passed into Plug.Conn.read_body and the one that comes out , and how this affects the pipeline downstream . If anyone has any insight on it , am curious to hear)

While I can’t comment on the exact reasons I’d always advise to use the Plug.Conn that’s returned to you. If you have such a return value then the underlying library has changed it and gives you the modified copy. So just use that.