https://github.com/google/seesaw

------------------------

在分布式系统中，负载均衡是非常重要的环节，通过负载均衡将请求派发到网络中的一个或多个节点上进行处理。通常来说，负载均衡分为硬件负载均衡及软件负载均衡。硬件负载均衡，顾名思义，在服务器节点之间安装专门的硬件进行负载均衡的工作，F5便为其中的佼佼者。软件负载均衡则是通过在服务器上安装的特定的负载均衡软件或是自带负载均衡模块完成对请求的分配派发。

一般而言，有以下几种常见的负载均衡策略：

一.轮询。作为非常经典的负载均衡策略，早期该策略应用地非常广泛。其原理很简单，给每个请求标记一个序号，然后将请求依次派发到服务器节点中，适用于集群中各个节点提供服务能力等同且无状态的场景。其缺点也非常明显，该策略将节点视为等同，与实际中复杂的环境不符。加权轮询为轮询的一个改进策略，每个节点会有权重属性，但是因为权重的设置难以做到随实际情况变化，仍有一定的不足。

二.随机。与轮询相似，只是不需要对每个请求进行编号，每次随机取一个。同样地，该策略也将后端的每个节点是为等同的。另外同样也有改进的加权随机的算法，不再赘述。

三.最小响应时间。通过记录每次请求所需的时间，得出平均的响应时间，然后根据响应时间选择最小的响应时间。该策略能较好地反应服务器的状态，但是由于是平均响应时间的关系，时间上有些滞后，无法满足快速响应的要求。因此在此基础之上，会有一些改进版本的策略，如只计算最近若干次的平均时间的策略等。

四. 最小并发数。客户端的每一次请求服务在服务器停留的时间可能会有较大的差异,随着工作时间加长,如果采用简单的轮循或随机均衡算法,每一台服务器上的连接进程可能会产生较大的不同,并没有达到真正的负载均衡｡最小并发数的策略则是记录了当前时刻，每个备选节点正在处理的事务数，然后选择并发数最小的节点。该策略能够快速地反应服务器的当前状况，较为合理地将负责分配均匀，适用于对当前系统负载较为敏感的场景。

五.哈希。在后端节点有状态的情况下，需要使用哈希的方法进行负载均衡，此种情况下情况比较复杂，本文对此不做探讨。

另外还有其他的负载均衡策略不再一一列举，有兴趣的同学可以自己去查阅相关资料。

分布式系统面临着远比单机系统更加复杂的环境，包括不同的网络环境、运行平台、机器配置等等。在如此复杂的环境中，发生错误是不可避免的，然后如何能够做到容错性，将发生错误的代价降低到最低是在分布式系统中必须要考虑的问题。选择不同的负载均衡策略将会有非常大的不同。

------------------------

Seesaw v2

Note: This is not an official Google product.

About

Seesaw v2 is a Linux Virtual Server (LVS) based load balancing platform.

It is capable of providing basic load balancing for servers that are on the same network, through to advanced load balancing functionality such as anycast, Direct Server Return (DSR), support for multiple VLANs and centralised configuration.

Above all, it is designed to be reliable and easy to maintain.

Requirements

A Seesaw v2 load balancing cluster requires two Seesaw nodes - these can be physical machines or virtual instances. Each node must have two network interfaces - one for the host itself and the other for the cluster VIP. All four interfaces should be connected to the same layer 2 network.

Building

Seesaw v2 is developed in Go and depends on several Go packages:

Additionally, there is a compile and runtime dependency on libnl and a compile time dependency on the Go protobuf compiler.

On a Debian/Ubuntu style system, you should be able to prepare for building by running:

apt-get install golang
apt-get install libnl-3-dev libnl-genl-3-dev

If your distro has a go version before 1.5, you may need to fetch a newer release from https://golang.org/dl/.

After setting GOPATH to an appropriate location (for example ~/go):

go get -u golang.org/x/crypto/ssh
go get -u github.com/dlintw/goconf
go get -u github.com/golang/glog
go get -u github.com/miekg/dns
go get -u github.com/kylelemons/godebug/pretty

Ensure that ${GOPATH}/bin is in your ${PATH} and in the seesaw directory:

make test
make install

If you wish to regenerate the protobuf code, the protobuf compiler and Go protobuf compiler generator are also needed:

apt-get install protobuf-compiler
go get -u github.com/golang/protobuf/{proto,protoc-gen-go}

The protobuf code can then be regenerated with:

make proto

Installing

After make install has run successfully, there should be a number of binaries in ${GOPATH}/bin with a seesaw_ prefix. Install these to the appropriate locations:

SEESAW_BIN="/usr/local/seesaw"
SEESAW_ETC="/etc/seesaw"
SEESAW_LOG="/var/log/seesaw"

INIT=`ps -p 1 -o comm=`

install -d "${SEESAW_BIN}" "${SEESAW_ETC}" "${SEESAW_LOG}"

install "${GOPATH}/bin/seesaw_cli" /usr/bin/seesaw

for component in {ecu,engine,ha,healthcheck,ncc,watchdog}; do
  install "${GOPATH}/bin/seesaw_${component}" "${SEESAW_BIN}"
done

if [ $INIT = "init" ]; then
  install "etc/init/seesaw_watchdog.conf" "/etc/init"
elif [ $INIT = "systemd" ]; then
  install "etc/systemd/system/seesaw_watchdog.service" "/etc/systemd/system"
  systemctl --system daemon-reload
fi
install "etc/seesaw/watchdog.cfg" "${SEESAW_ETC}"

# Enable CAP_NET_RAW for seesaw binaries that require raw sockets.
/sbin/setcap cap_net_raw+ep "${SEESAW_BIN}/seesaw_ha"
/sbin/setcap cap_net_raw+ep "${SEESAW_BIN}/seesaw_healthcheck"

The setcap binary can be found in the libcap2-bin package on Debian/Ubuntu.

Configuring

Each node needs a /etc/seesaw/seesaw.cfg configuration file, which provides information about the node and who its peer is. Additionally, each load balancing cluster needs a cluster configuration, which is in the form of a text-based protobuf - this is stored in /etc/seesaw/cluster.pb.

An example seesaw.cfg file can be found in etc/seesaw/seesaw.cfg.example - a minimal seesaw.cfg provides the following:

anycast_enabled - True if anycast should be enabled for this cluster.
name - The short name of this cluster.
node_ipv4 - The IPv4 address of this Seesaw node.
peer_ipv4 - The IPv4 address of our peer Seesaw node.
vip_ipv4 - The IPv4 address for this cluster VIP.

The VIP floats between the Seesaw nodes and is only active on the current master. This address needs to be allocated within the same netblock as both the node IP address and peer IP address.

An example cluster.pb file can be found in etc/seesaw/cluster.pb.example - a minimal cluster.pb contains a seesaw_vip entry and two node entries. For each service that you want to load balance, a separate vserver entry is needed, with one or more vserver_entry sections (one per port/proto pair), one or more backends and one or more healthchecks. Further information is available in the protobuf definition - see pb/config/config.proto.

On an upstart based system, running restart seesaw_watchdog will start (or restart) the watchdog process, which will in turn start the other components.

Anycast

Seesaw v2 provides full support for anycast VIPs - that is, it will advertise an anycast VIP when it becomes available and will withdraw the anycast VIP if it becomes unavailable. For this to work the Quagga BGP daemon needs to be installed and configured, with the BGP peers accepting host-specific routes that are advertised from the Seesaw nodes within the anycast range (currently hardcoded as 192.168.255.0/24).

Command Line

Once initial configuration has been performed and the Seesaw components are running, the state of the Seesaw can be viewed and controlled via the Seesaw command line interface. Running seesaw (assuming /usr/bin is in your path) will give you an interactive prompt - type ? for a list of top level commands. A quick summary:

config reload - reload the cluster.pb from the current config source.
failover - failover between the Seesaw nodes.
show vservers - list all vservers configured on this cluster.
show vserver <name> - show the current state for the named vserver.

Troubleshooting

A Seesaw should have five components that are running under the watchdog - the process table should show processes for:

seesaw_ecu
seesaw_engine
seesaw_ha
seesaw_healthcheck
seesaw_ncc
seesaw_watchdog

All Seesaw v2 components have their own logs, in addition to the logging provided by the watchdog. If any of the processes are not running, check the corresponding logs in /var/log/seesaw (e.g. seesaw_engine.{log,INFO}).

Google开源软负载seesaw