Compare commits

...

3 Commits

Author SHA1 Message Date
Devin Carr 9b8a533435 Release 2023.7.1 2023-07-13 12:31:33 -07:00
Devin Carr 5abb90b539 TUN-7582: Correct changelog wording for --management-diagnostics 2023-07-13 09:47:21 -07:00
João Oliveirinha 0c8bc56930 TUN-7575: Add option to disable PTMU discovery over QUIC
This commit implements the option to disable PTMU discovery for QUIC
connections.
QUIC finds the PMTU during startup by increasing Ping packet frames
until Ping responses are not received anymore, and it seems to stick
with that PMTU forever.

This is no problem if the PTMU doesn't change over time, but if it does
it may case packet drops.
We add this hidden flag for debugging purposes in such situations as a
quick way to validate if problems that are being seen can be solved by
reducing the packet size to the edge.

Note however, that this option may impact UDP proxying since we expect
being able to send UDP packets of 1280 bytes over QUIC.
So, this option should not be used when tunnel is being used for UDP
proxying.
2023-07-13 10:24:24 +01:00
5 changed files with 29 additions and 9 deletions
+1 -1
View File
@@ -1,6 +1,6 @@
## 2023.7.0
### New Features
- You can now enable additional diagnostics over the management.argotunnel.com service for your active cloudflared connectors via a new runtime flag `--management-diagnostics` (or env `TUNNEL_MANAGEMENT_DIAGNOSTICS`). This feature is provided as opt-out and requires the flag to enable. Endpoints such as /metrics provides your prometheus metrics endpoint another mechanism to be reached. Additionally /debug/pprof/(goroutine|heap) are also introduced to allow for remotely retrieving active pprof information from a running cloudflared connector.
- You can now enable additional diagnostics over the management.argotunnel.com service for your active cloudflared connectors via a new runtime flag `--management-diagnostics` (or env `TUNNEL_MANAGEMENT_DIAGNOSTICS`). This feature is provided as opt-in and requires the flag to enable. Endpoints such as /metrics provides your prometheus metrics endpoint another mechanism to be reached. Additionally /debug/pprof/(goroutine|heap) are also introduced to allow for remotely retrieving active pprof information from a running cloudflared connector.
## 2023.4.1
### New Features
+4
View File
@@ -1,3 +1,7 @@
2023.7.1
- 2023-07-13 TUN-7582: Correct changelog wording for --management-diagnostics
- 2023-07-12 TUN-7575: Add option to disable PTMU discovery over QUIC
2023.7.0
- 2023-07-06 TUN-7558: Flush on Writes for StreamBasedOriginProxy
- 2023-07-05 TUN-7553: Add flag to enable management diagnostic services
+12
View File
@@ -82,6 +82,11 @@ const (
// udpUnregisterSessionTimeout is how long we wait before we stop trying to unregister a UDP session from the edge
udpUnregisterSessionTimeoutFlag = "udp-unregister-session-timeout"
// quicDisablePathMTUDiscovery sets if QUIC should not perform PTMU discovery and use a smaller (safe) packet size.
// Packets will then be at most 1252 (IPv4) / 1232 (IPv6) bytes in size.
// Note that this may result in packet drops for UDP proxying, since we expect being able to send at least 1280 bytes of inner packets.
quicDisablePathMTUDiscovery = "quic-disable-pmtu-discovery"
// uiFlag is to enable launching cloudflared in interactive UI mode
uiFlag = "ui"
@@ -692,6 +697,13 @@ func tunnelFlags(shouldHide bool) []cli.Flag {
Value: 5 * time.Second,
Hidden: true,
}),
altsrc.NewBoolFlag(&cli.BoolFlag{
Name: quicDisablePathMTUDiscovery,
EnvVars: []string{"TUNNEL_DISABLE_QUIC_PMTU"},
Usage: "Use this option to disable PTMU discovery for QUIC connections. This will result in lower packet sizes. Not however, that this may cause instability for UDP proxying.",
Value: false,
Hidden: true,
}),
altsrc.NewStringFlag(&cli.StringFlag{
Name: connectorLabelFlag,
Usage: "Use this option to give a meaningful label to a specific connector. When a tunnel starts up, a connector id unique to the tunnel is generated. This is a uuid. To make it easier to identify a connector, we will use the hostname of the machine the tunnel is running on along with the connector ID. This option exists if one wants to have more control over what their individual connectors are called.",
+1
View File
@@ -240,6 +240,7 @@ func prepareTunnelConfig(
PQKexIdx: pqKexIdx,
MaxEdgeAddrRetries: uint8(c.Int("max-edge-addr-retries")),
UDPUnregisterSessionTimeout: c.Duration(udpUnregisterSessionTimeoutFlag),
DisableQUICPathMTUDiscovery: c.Bool(quicDisablePathMTUDiscovery),
}
packetConfig, err := newPacketConfig(c, log)
if err != nil {
+11 -8
View File
@@ -70,6 +70,8 @@ type TunnelConfig struct {
PacketConfig *ingress.GlobalRouterConfig
UDPUnregisterSessionTimeout time.Duration
DisableQUICPathMTUDiscovery bool
}
func (c *TunnelConfig) registrationOptions(connectionID uint8, OriginLocalIP string, uuid uuid.UUID) *tunnelpogs.RegistrationOptions {
@@ -596,14 +598,15 @@ func (e *EdgeTunnelServer) serveQUIC(
}
quicConfig := &quic.Config{
HandshakeIdleTimeout: quicpogs.HandshakeIdleTimeout,
MaxIdleTimeout: quicpogs.MaxIdleTimeout,
KeepAlivePeriod: quicpogs.MaxIdlePingPeriod,
MaxIncomingStreams: quicpogs.MaxIncomingStreams,
MaxIncomingUniStreams: quicpogs.MaxIncomingStreams,
EnableDatagrams: true,
MaxDatagramFrameSize: quicpogs.MaxDatagramFrameSize,
Tracer: quicpogs.NewClientTracer(connLogger.Logger(), connIndex),
HandshakeIdleTimeout: quicpogs.HandshakeIdleTimeout,
MaxIdleTimeout: quicpogs.MaxIdleTimeout,
KeepAlivePeriod: quicpogs.MaxIdlePingPeriod,
MaxIncomingStreams: quicpogs.MaxIncomingStreams,
MaxIncomingUniStreams: quicpogs.MaxIncomingStreams,
EnableDatagrams: true,
MaxDatagramFrameSize: quicpogs.MaxDatagramFrameSize,
Tracer: quicpogs.NewClientTracer(connLogger.Logger(), connIndex),
DisablePathMTUDiscovery: e.config.DisableQUICPathMTUDiscovery,
}
quicConn, err := connection.NewQUICConnection(