http2: always delay closing the connection after sending GOAWAY
Currently, we close the connection immediately after sending a GOAWAY
frame if all outstanding responses have been completely sent. However,
the client may have had requests in-flight at that time which have been
queued in the kernel receive buffer. On both Windows and Linux, if the
connection is close()'d when the receive buffer is not empty, the kernel
sends RST. This has the effect of aborting both sides of the connection,
meaning the client may not actually receive all responses that were sent
before the GOAWAY.
Instead, we should delay calling close() until after the receive buffer
has been drained. We don't want to delay indefinitely, which means we
need some kind of timeout. Ideally that timeout should be about 1 RTT +
epsilon, under the assumption that the client will not send any more
frames after receiving the GOAWAY. However, 1 RTT is difficult to
measure. It turns out we were already using a 1 second delay in other
cases, so we reuse that same delay here.
Note that we do not call CloseWrite() to half-close the underlying TLS
connection. This seems unnecessary -- GOAWAY is effectively a half-close
at the HTTP/2 level.
Updates golang/go#18701 (fixes after it's bundled into net/http)
Change-Id: I4d68bada6369ba95e5db02afe6dfad0a393c0334
Reviewed-on: https://go-review.googlesource.com/71372
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
diff --git a/http2/server.go b/http2/server.go
index eae143d..d790c3b 100644
--- a/http2/server.go
+++ b/http2/server.go
@@ -853,8 +853,13 @@
}
}
- if sc.inGoAway && sc.curOpenStreams() == 0 && !sc.needToSendGoAway && !sc.writingFrame {
- return
+ // Start the shutdown timer after sending a GOAWAY. When sending GOAWAY
+ // with no error code (graceful shutdown), don't start the timer until
+ // all open streams have been completed.
+ sentGoAway := sc.inGoAway && !sc.needToSendGoAway && !sc.writingFrame
+ gracefulShutdownComplete := sc.goAwayCode == ErrCodeNo && sc.curOpenStreams() == 0
+ if sentGoAway && sc.shutdownTimer == nil && (sc.goAwayCode != ErrCodeNo || gracefulShutdownComplete) {
+ sc.shutDownIn(goAwayTimeout)
}
}
}
@@ -1218,30 +1223,31 @@
sc.shutdownOnce.Do(func() { sc.sendServeMsg(gracefulShutdownMsg) })
}
+// After sending GOAWAY, the connection will close after goAwayTimeout.
+// If we close the connection immediately after sending GOAWAY, there may
+// be unsent data in our kernel receive buffer, which will cause the kernel
+// to send a TCP RST on close() instead of a FIN. This RST will abort the
+// connection immediately, whether or not the client had received the GOAWAY.
+//
+// Ideally we should delay for at least 1 RTT + epsilon so the client has
+// a chance to read the GOAWAY and stop sending messages. Measuring RTT
+// is hard, so we approximate with 1 second. See golang.org/issue/18701.
+//
+// This is a var so it can be shorter in tests, where all requests uses the
+// loopback interface making the expected RTT very small.
+//
+// TODO: configurable?
+var goAwayTimeout = 1 * time.Second
+
func (sc *serverConn) startGracefulShutdownInternal() {
- sc.goAwayIn(ErrCodeNo, 0)
+ sc.goAway(ErrCodeNo)
}
func (sc *serverConn) goAway(code ErrCode) {
sc.serveG.check()
- var forceCloseIn time.Duration
- if code != ErrCodeNo {
- forceCloseIn = 250 * time.Millisecond
- } else {
- // TODO: configurable
- forceCloseIn = 1 * time.Second
- }
- sc.goAwayIn(code, forceCloseIn)
-}
-
-func (sc *serverConn) goAwayIn(code ErrCode, forceCloseIn time.Duration) {
- sc.serveG.check()
if sc.inGoAway {
return
}
- if forceCloseIn != 0 {
- sc.shutDownIn(forceCloseIn)
- }
sc.inGoAway = true
sc.needToSendGoAway = true
sc.goAwayCode = code
