ESI request forced HTTPS connection even if config.upstream_use_ssl is set to false

[kzolnowski]

I'm trying to use offloaded SSL traffic with ESI. Handler config:

require("ledge").set_handler_defaults({
    upstream_host = "127.0.0.2",
    upstream_port = 80,
    advertise_ledge = false,
    upstream_use_ssl = false,
})

While trying to use ESI I've got error:

[error] 4219#0: *69822327 [lua] processor_1_0.lua:461: esi_fetch_include(): ssl handshake failed: handshake failed, client: 127.0.0.3, server: test, request: "GET / HTTP/1.1", host: "test"
[error] 4219#0: *69822327 SSL_do_handshake() failed (SSL: error:140770FC:SSL routines:SSL23_GET_SERVER_HELLO:unknown protocol), client: 127.0.0.3, server: test, request: "GET / HTTP/1.1", host: "test"

I found that ESI proccessor is trying to use ssl handshake even if config.upstream_use_ssl is set to false, so I already fixed that in code (see attachment) ;)

esi_ssl_offload.diff.txt

[hamishforbes]

Hey Krzys!

Hmmm, It's actually little more complicated than that.

ESIs could be to an external server which does require HTTPS, so just disabling the handshake when our current upstream does not use SSL isn't correct.

We loopback ESIs with the same domain as the current request to the same IP and port as the current request was received on, the assumption being if we are handling a request for a domain then we can handle the ESI for that domain too.
Perhaps this optimisation should be configurable, allowing it to be disabled or the behaviour controlled to allow scenarios like yours to work

Also I think there's probably another bug here whereby a page on HTTP or HTTPS that has an ESI for the same domain but a different scheme is going to attempt to loopback to the wrong port.

[pintsized]

Hey, I realise this is ancient but I think this is now addressed here: b6bd6e1