Expect header #2343
Expect header #2343
Conversation
const-t
force-pushed
the
kt-533-expect
branch
from
9dd9a0e to
a58000b
Compare
const-t
force-pushed
the
kt-533-expect
branch
from
a58000b to
3939451
Compare
| * request into @seq_queue, to proccess responses in correct order. | ||
| */ | ||
| static int | ||
| tfw_http_send_continuation(TfwCliConn *cli_conn, TfwHttpReq *req) |
There was a problem hiding this comment.
Why don't we use tfw_h1_send_resp()? I didn't get which specifics we need for 100-continue, but maybe it's better to make tfw_h1_send_resp() API more flexible rather than keeping an alternate workflow?
There was a problem hiding this comment.
tfw_h1_send_resp() uses tfw_http_resp_fwd() that can't be used in this case without modifications, when tfw_http_resp_fwd() called it expects that request already in seq_queue, however tfw_http_send_continuation() sends requests immediately if seq_queue is empty and tfw_http_send_continuation() doesn't free request paired with response, but tfw_http_resp_fwd() does.
Modifications of tfw_http_resp_fwd() will overcomplicate the function, because of this I created a new one, that do only what we need in the case of 100-continue.
However, I think we can call tfw_http_send_continuation() from tfw_h1_send_resp() if extend API of tfw_h1_send_resp(). But it may have some overhead when error occurred during response allocation. Is it worth to do that?
There was a problem hiding this comment.
No, I think we're good with the current code. Could you please add the 1st paragraph of your answer as a comment for tfw_http_send_continuation()
Add new special header `Expect` with only one allowed value `101-continue`. See RFC 9110 Section 10.1.1.
Tempesta FW processes `Expect` itself and doesn't forward this header to upstream to avoid unnecessary traffic.
Once `Expect: 100-continue` received and no content received, Tempesta FW return `100-continue` response. When Tempesta received complete request, `Expect` header will be removed from the request during forwarding request to upstream. Tempesta never sends `Expect` to upstream, this header supported only when received from client. In case of pipelined requests, when first request dosn't contain `Expect` header, but second does, Tempesta responds in the same order as requests were received. The first will be sent response to the first request, then `100-continue` will be sent and after final response to the second request will be sent. This is achieved by using connection `seq_queue`, once `Expect` header received, `100-continue` response will be paired with corresponding request in the `seq_queue`, however if for some reason `100-continue` has not been sent and part of the body or whole body was received, 100-continue will be unpaired from request and freed, the request will be deleted from the `seq_queue` until it's not be processed completely. Therefore in this case only final response will be sent.
const-t
force-pushed
the
kt-533-expect
branch
from
3939451 to
738ceca
Compare
| * request into @seq_queue, to proccess responses in correct order. | ||
| */ | ||
| static int | ||
| tfw_http_send_continuation(TfwCliConn *cli_conn, TfwHttpReq *req) |
There was a problem hiding this comment.
No, I think we're good with the current code. Could you please add the 1st paragraph of your answer as a comment for tfw_http_send_continuation()
No description provided.