Lighttpd

An strace of the lighttpd process when this happens would be useful.

Have you tried checking to see if this is related to browser behavior, or, more precisely, javascript in the browser misbehaving? Please try using curl instead of the browser. Do you see the same behavior?

Hi gstrauss, thanks for your answer.

I am confident this is not browser related, this happens in multiple browser tested, and always disappears disabling deflate.
I was able to reproduce it with curl too.
This is the command output with timings: https://pastebin.com/5CTQ6M2y

The line "* no chunk, no close, no size. Assume close to signal end" seems related, maybe.
It's like lighttpd used 6 second to gzip a 991 bytes file, before start sending html payload, which is a bit strange (of course there is absolutely no load on the machine, that is a quad core Intel with 32gb ram). I'm just guessing by the way.

Just to provide more information, even if not very low level, here is a firefox console screenshot showing the process dowloading some (compressed) svg files, stuck for 6 seconds before continung on. I don't think there is any javascript here.

I've never used strace in my whole life, so I am a bit embarassed, but i'll try if there is no other way.
Any idea so far?

Thank you so much :)

ick. That's a bug. Thanks for reporting it. This should fix it:

--- a/src/http_header.c
+++ b/src/http_header.c
@@ -69,7 +69,7 @@ void http_header_response_set(connection *con, enum http_header_e id, const char
      * (note: if 0 == vlen, header is still inserted with blank value,
      *  which is used to indicate a "removed" header)
      */
-    con->response.htags |= id;
+    (vlen) ? (con->response.htags |= id) : (con->response.htags &= ~id);
     array_set_key_value(con->response.headers, k, klen, v, vlen);
 }

@@ -121,7 +121,7 @@ void http_header_request_set(connection *con, enum http_header_e id, const char
      * (note: if 0 == vlen, header is still inserted with blank value,
      *  which is used to indicate a "removed" header)
      */
-    con->request.htags |= id;
+    (vlen) ? (con->request.htags |= id) : (con->request.htags &= ~id);
     array_set_key_value(con->request.headers, k, klen, v, vlen);
 }

A more extensive modification is needed for a more complete generic fix

Glad to be helpful, and thank you so much for your work! :)

Fixed in #2919