TCP KeepAlive

From NovaOrdis Knowledge Base
Jump to navigation Jump to search

External

Internal

Overview

TCP Keep-Alive is a mechanism that insures small probe packets are periodically sent to the other end of the TCP connection, with the purpose of detecting whether the peer host crashed. Under normal circumstances, if the TCP Keep-Alive is enabled and if the TCP stacks at both ends are up, the keep-alive probe exchange will continue indefinitely if the application layers at both ends stay idle, not sending data - TCP keep-alive does not need application traffic to latch onto. For a discussion on how long a TCP stays up in presence or absence of application traffic, see Duration of a TCP Connection. Note that the keep-alive mechanism has to be explicitly enabled, whether it is enabled by default or not is implementation dependent. The keep-alive probe logic is implemented in the TCP stack: if the keep-alive option is enabled, and no application data has been exchanged across the socket in either direction for keep-alive time, by default 7,200 seconds, TCP automatically sends a keep-alive probe to the peer. This probe is a TCP segment which contains null data. In an Ethernet network, a keepalive frame length is 60 bytes, while the server response to this, also a null data frame, is 54 bytes. If a peer receives a keep-alive probe, then it must respond to it. One of three responses is expected:

  1. The peer responds with the expected ACK. The application is not notified, since everything is OK. TCP will send another probe following another keep-alive time seconds of inactivity.
  2. The peer responds with an RST, which tells the local TCP that the peer host has crashed and rebooted. As result, the socket will be closed.
  3. There is no response from the peer. The OS will close the socket and will release the associated resources. The application listening on that particular socket will receive an error from the OS.

Aside from application notification in case of connection failure, another benefit of enabling TCP Keep-Alive is that it keeps the connection "active" so if the connection goes over a firewall that watches for inactivity, that will prevent the firewall from dropping the connection.

According to RFC 1122 4.2.3.6, responding to and/or relaying TCP Keep-Alive packets is optional:

Implementors MAY include “keep-alives” in their TCP implementations, although this practice is not universally accepted. If keep-alives are included, the application MUST be able to turn them on or off for each TCP connection, and they MUST default to off. [...] It is extremely important to remember that ACK segments that contain no data are not reliably transmitted by TCP.

If a Java application's socket detects a failed keep-alive exchange, it throws:

java.io.IOException: Operation timed out
	at sun.nio.ch.FileDispatcherImpl.read0(Native Method)
	at sun.nio.ch.SocketDispatcher.read(SocketDispatcher.java:39)
	at sun.nio.ch.IOUtil.readIntoNativeBuffer(IOUtil.java:223)
	at sun.nio.ch.IOUtil.read(IOUtil.java:197)
	at sun.nio.ch.SocketChannelImpl.read(SocketChannelImpl.java:380)
        ...

Configuration

Keep-alive can be turned on or off on an individual socket basis. Moreover, either side of the connection can have its keep-alive status configured independently. Java networking API will report whether a certain socket has its keep-alive mechanism turned on or off, and allows configuring it programmatically. For details, see Socket SO KEEPALIVE.

However, the numeric parameters dictating how TCP Keep-Alive behaves can only be changed at the system level, in the kernel:

Keep-Alive Time

The time of connection inactivity after which the first keep-alive request is sent. In other words, is the duration between two keepalive transmissions in idle condition. The default value on Linux is 2 hours (7,200 seconds). On Linux, it is available as "net.ipv4.tcp_keepalive_time" kernel configuration parameter.

Keep-Alive Interval

The duration between two successive keepalive retransmissions, if acknowledgement to the previous keepalive transmission is not received. On Linux, it is available as "net.ipv4.tcp_keepalive_intvl" kernel configuration parameter.

Kee-Alive Retry

The number of retransmissions to be carried out before declaring that remote end is not available. On Linux, it is available as "net.ipv4.tcp_keepalive_probes" kernel configuration parameter.

O/S Specific Details

The fact that TCP KeepAlive is enabled or not, and how it is configured, it is OS-dependent. This is how the configuration values can be obtained and changed on various systems:

Linux

To get the values:

sysctl net.ipv4.tcp_keepalive_time
sysctl net.ipv4.tcp_keepalive_probes
sysctl net.ipv4.tcp_keepalive_intvl

or read them from /proc/sys/net/ipv4/tcp_keepalive_time, /proc/sys/net/ipv4/tcp_keepalive_probes and /proc/sys/net/ipv4/tcp_keepalive_intvl.

The values for tcp_keepalive_time and tcp_keepalive_intvl are expressed in seconds.

To change them to survive reboot:

sysctl -w net.ipv4.tcp_keepalive_time=180 net.ipv4.tcp_keepalive_probes=3 net.ipv4.tcp_keepalive_intvl=10

For more details see Kernel Runtime Configuration.

Mac

To get the values:

sysctl net.inet.tcp | grep -E "keepidle|keepintvl|keepcnt"
net.inet.tcp.keepidle: 7200000
net.inet.tcp.keepintvl: 75000
net.inet.tcp.keepcnt: 8

Note that values of keepidle and keepintvl are specified in milliseconds, as opposed to Linux that uses seconds.

To set the values to survive reboot:

sysctl -w net.inet.tcp.keepidle=180000 net.inet.tcp.keepcnt=3 net.inet.tcp.keepintvl=10000

For more methods to set Mac kernel parameters, see Mac Kernel Parameter Configuration.

Questions and TODO

  • Can keepalive be set per TCP connection, or is a system-wide setting (all TCP/IP connections)? Enable set-keep-alive on|off and experiment with it.
  • So it is true that if I don't have keep alive, my write can block forever if I power off the other end suddenly.