yahns Ruby server user/dev discussion
 help / color / mirror / code / Atom feed
From: Eric Wong <e@80x24.org>
To: yahns-public@yhbt.net
Subject: [PATCH 1/7] proxy_pass: simplify writing request bodies upstream
Date: Mon, 16 May 2016 01:43:34 +0000	[thread overview]
Message-ID: <20160516014340.8258-2-e@80x24.org> (raw)
In-Reply-To: <20160516014340.8258-1-e@80x24.org>

The cost of extra branches inside a loop is negligible compared
to the cost of all the other method calls we make.  Favor
smaller code instead and inline some (now) single-use methods.

Furthermore, this allows us to reuse the request header buffer
instead of relying on thread-local storage and potentially
having to to swap buffers.
---
 lib/yahns/proxy_pass.rb | 112 +++++++++++++++++++++---------------------------
 1 file changed, 49 insertions(+), 63 deletions(-)

diff --git a/lib/yahns/proxy_pass.rb b/lib/yahns/proxy_pass.rb
index 148957b..a2d7d81 100644
--- a/lib/yahns/proxy_pass.rb
+++ b/lib/yahns/proxy_pass.rb
@@ -22,12 +22,6 @@ def req_start(c, req, input, chunked)
       Thread.current[:yahns_queue].queue_add(self, Yahns::Queue::QEV_WR)
     end
 
-    # we must reinitialize the thread-local rbuf if it may get beyond the
-    # current thread
-    def detach_rbuf!
-      Thread.current[:yahns_rbuf] = ''.dup
-    end
-
     def yahns_step # yahns event loop entry point
       c = @yahns_client
       case req = @rrstate
@@ -42,7 +36,9 @@ def yahns_step # yahns event loop entry point
             if res = req.headers(@hdr = [], rv)
               return c.proxy_response_start(res, rv, req, self)
             else # ugh, big headers or tricked response
-              buf = detach_rbuf!
+              # we must reinitialize the thread-local rbuf if it may
+              # live beyond the current thread
+              buf = Thread.current[:yahns_rbuf] = ''.dup
               @resbuf = rv
             end
             # continue looping in middle "case @resbuf" loop
@@ -83,64 +79,63 @@ def yahns_step # yahns event loop entry point
       c.proxy_err_response(502, self, e, wbuf)
     end
 
-    # returns :wait_readable if complete, :wait_writable if not
-    def send_req_body(req)
-      buf, input, chunked = req
-
-      # get the first buffered chunk or vector
+    def send_req_body_chunk(buf)
       case rv = String === buf ? kgio_trywrite(buf) : kgio_trywritev(buf)
       when String, Array
-        buf = rv # retry inner loop
-      when :wait_writable
-        req[0] = buf
-        return :wait_writable
-      when nil
-        break # onto writing body
+        buf.replace(rv) # retry loop on partial write
+      when :wait_writable, nil
+        # :wait_writable = upstream is reading slowly and making us wait
+        return rv
+      else
+        abort "BUG: #{rv.inspect} from kgio_trywrite*"
       end while true
+    end
 
-      buf = Thread.current[:yahns_rbuf]
+    # returns :wait_readable if complete, :wait_writable if not
+    def send_req_body(req) # @rrstate == [ (str|vec), rack.input, chunked? ]
+      buf, input, chunked = req
 
-      # Note: input (env['rack.input']) is fully-buffered by default so
-      # we should not be waiting on a slow network resource when reading
-      # input.  However, some weird configs may disable this on LANs
+      # send the first buffered chunk or vector
+      rv = send_req_body_chunk(buf) and return rv # :wait_writable
 
+      # yay, sent the first chunk, now read the body!
+      rbuf = buf
       if chunked
-        while input.read(0x2000, buf)
-          vec = [ "#{buf.size.to_s(16)}\r\n", buf, "\r\n".freeze ]
-          case rv = kgio_trywritev(vec)
-          when Array
-            vec = rv # partial write, retry in case loop
-          when :wait_writable
-            detach_rbuf!
-            req[0] = vec
-            return :wait_writable
-          when nil
-            break # continue onto reading next chunk
-          end while true
+        if String === buf # initial body
+          req[0] = buf = []
+        else
+          # try to reuse the biggest non-frozen buffer we just wrote;
+          rbuf = buf.max_by(&:size)
+          rbuf = ''.dup if rbuf.frozen? # unlikely...
         end
-        close_req_body(input)
-
-        # note: we do not send any trailer, they are folded into the header
-        # because this relies on full request buffering
-        send_req_buf("0\r\n\r\n".freeze)
-        # prepare_wait_readable already called by send_req_buf
-      else # identity request, easy:
-        while input.read(0x2000, buf)
-          case rv = kgio_trywrite(buf)
-          when String
-            buf = rv # partial write, retry in case loop
-          when :wait_writable
-            detach_rbuf!
-            req[0] = buf
-            return :wait_writable
-          when nil
-            break # continue onto reading next block
-          end while true
+      end
+
+      # Note: input (env['rack.input']) is fully-buffered by default so
+      # we should not be waiting on a slow network resource when reading
+      # input.  However, some weird configs may disable this on LANs
+      # and we may wait indefinitely on input.read here...
+      while input.read(0x2000, rbuf)
+        if chunked
+          buf[0] = "#{rbuf.size.to_s(16)}\r\n".freeze
+          buf[1] = rbuf
+          buf[2] = "\r\n".freeze
         end
+        rv = send_req_body_chunk(buf) and return rv # :wait_writable
+      end
+
+      rbuf.clear # all done, clear the big buffer
 
-        close_req_body(input)
-        prepare_wait_readable
+      # we cannot use respond_to?(:close) here since Rack::Lint::InputWrapper
+      # tries to prevent that (and hijack means all Rack specs go out the door)
+      case input
+      when Yahns::TeeInput, IO
+        input.close
       end
+
+      # note: we do not send any trailer, they are folded into the header
+      # because this relies on full request buffering
+      # prepare_wait_readable is called by send_req_buf
+      chunked ? send_req_buf("0\r\n\r\n".freeze) : prepare_wait_readable
     rescue Errno::EPIPE, Errno::ECONNRESET, Errno::ENOTCONN
       # no more reading off the client socket, just prepare to forward
       # the rejection response from the upstream (if any)
@@ -153,15 +148,6 @@ def prepare_wait_readable
       :wait_readable # all done sending the request, wait for response
     end
 
-    def close_req_body(input)
-      # we cannot use respond_to?(:close) here since Rack::Lint::InputWrapper
-      # tries to prevent that (and hijack means all Rack specs go out the door)
-      case input
-      when Yahns::TeeInput, IO
-        input.close
-      end
-    end
-
     # n.b. buf must be a detached string not shared with
     # Thread.current[:yahns_rbuf] of any thread
     def send_req_buf(buf)

  reply	other threads:[~2016-05-16  1:43 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-05-16  1:43 [PATCH 0/7] proxy_pass cleanups Eric Wong
2016-05-16  1:43 ` Eric Wong [this message]
2016-05-16  1:43 ` [PATCH 2/7] proxy_pass: hoist out proxy_res_headers method Eric Wong
2016-05-16  1:43 ` [PATCH 3/7] proxy_pass: simplify proxy_http_response Eric Wong
2016-05-16  1:43 ` [PATCH 4/7] proxy_pass: split out body and trailer reading in response Eric Wong
2016-05-16  1:43 ` [PATCH 5/7] proxy_pass: trim down proxy_response_finish, too Eric Wong
2016-05-16  1:43 ` [PATCH 6/7] proxy_pass: split out req_res into a separate file Eric Wong
2016-05-16  1:43 ` [PATCH 7/7] proxy_pass: fix resumes after complete buffering is unblocked Eric Wong
2016-05-16  2:05 ` false-positive spam [Re: [PATCH 0/7] proxy_pass cleanups] Eric Wong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: https://yhbt.net/yahns/README

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20160516014340.8258-2-e@80x24.org \
    --to=e@80x24.org \
    --cc=yahns-public@yhbt.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

	https://yhbt.net/yahns.git/

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).