Selector¶

In GOST, the selection of nodes in a node group is done through node selector. The selector is responsible for selecting zero or one nodes in a node group using the node selection strategy. Selector can be applied to forwarding chains, hops, and forwarders. Node selectors are used in GOST for load balancing.

strategy (string, default=round)

Node selection strategy:

round - round robin
rand - random
fifo - top-down
hash - Based on a specific hash value (client IP or destination address)
parallel - race mode, dials all nodes concurrently and uses the first successful connection

maxFails (int, default=1)

The maximum number of failed connections for a specified node, When the number of failed connections with a node exceeds this set value, the node will be marked as a dead node, dead node will not be selected to use.

failTimeout (duration, default=10s)

Specify the dead node's timeout period. When a node is marked as a dead node, it will not be selected within this set time interval. After this set time interval, it will participate in node selection again.

Forwarding Chain¶

Selector can be set on each hop level of chain. The default selector uses the round strategy for node selection.

CLIFile (YAML)

gost -L http://:8080 -F socks5://192.168.1.1:1080,192.168.1.2:1080?strategy=rand&maxFails=3&failTimeout=60s

services:
- name: service-0
  addr: ":8080"
  handler:
    type: http
    chain: chain-0
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    selector:
      strategy: rand
      maxFails: 3
      failTimeout: 60s
    nodes:
    - name: node-0
      addr: 192.168.1.1:1080
      connector:
        type: socks5
      dialer:
        type: tcp
    - name: node-1
      addr: 192.168.1.2:1080
      connector:
        type: socks5
      dialer:
        type: tcp

Forwarder¶

Forwarder is used for port forwarding, it consists of a node group and a node selector. When forwarding is performed, zero or one node is selected from the node group for the forwarding destination address through the selector. The forwarder is now similar to a single-hop forwarding chain.

CLIFile (YAML)

gost -L "tcp://:8080/:8081,:8082?strategy=round&maxFails=1&failTimeout=30s

services:
- name: service-0
  addr: :8080
  handler:
    type: tcp
  listener:
    type: tcp
  forwarder:
    nodes:
    - name: target-0
      addr: :8081
    - name: target-1
      addr: :8082
    selector:
      strategy: round
      maxFails: 1
      failTimeout: 30s

Chain Group¶

services:
- name: service-0
  addr: ":8080"
  handler:
    type: http
    chainGroup:
      chains:
      - chain-0
      - chain-1
      selector:
        strategy: round
        maxFails: 1
        failTimeout: 10s
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8081
      connector:
        type: http
      dialer:
        type: tcp
- name: chain-1
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8082
      connector:
        type: http
      dialer:
        type: tcp

Backup Node and Chain¶

By marking one or more nodes or chains as backup, all backup nodes or chains will only participate in selection when all non-backup nodes or chains are marked as failed.

Backup Nodes¶

services:
- name: service-0
  addr: :8080
  handler:
    type: http
    chain: chain-0
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    selector:
      strategy: round
      maxFails: 1
      failTimeout: 10s
    nodes:
    - name: node-0
      addr: :8081
      metadata:
        maxFails: 3
        failTimeout: 30s
      connector:
        type: http
      dialer:
        type: tcp
    - name: node-1
      addr: :8082
      connector:
        type: http
      dialer:
        type: tcp
    - name: node-2
      addr: :8083
      metadata:
        backup: true
      connector:
        type: http
      dialer:
        type: tcp
    - name: node-3
      addr: :8084
      metadata:
        backup: true
      connector:
        type: http
      dialer:
        type: tcp

Mark nodes as backup via the metadata.backup option.

Under normal circumstances, only two non-backup nodes node-0 and node-1 participate in node selection. When both node-0 and node-1 are marked as failed, node-2 and node-3 will participate in node selection. When any one of node-0 and node-1 is recovered, node-2 and node-3 exit node selection.

Node-level Failure State Control

Pay attention to the node-0 node here, through the metadata.maxFails and metadata.failTimeout options, you can control the failure status of this node separately, and the corresponding parameters in the selector are used by default.

Backup Chains¶

services:
- name: service-0
  addr: :8080
  handler:
    type: http
    chainGroup:
      chains:
      - chain-0
      - chain-1
      - chain-2
      - chain-3
      selector:
        strategy: round
        maxFails: 1
        failTimeout: 10s
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8081
      connector:
        type: http
      dialer:
        type: tcp
- name: chain-1
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8082
      connector:
        type: http
      dialer:
        type: tcp
- name: chain-2
  metadata:
    backup: true
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8083
      connector:
        type: http
      dialer:
        type: tcp
- name: chain-3
  metadata:
    backup: true
  hops:
  - name: hop-0
    nodes:
    - name: node-0
      addr: :8084
      connector:
        type: http
      dialer:
        type: tcp

Similar to backup nodes, chains chain-2 and chain-3 are marked as backup via the metadata.backup option.

Weighted Random Selection Strategy¶

The selector supports setting weights for nodes and chains based on the random selection strategy. The default weight is 1.

services:
- name: service-0
  addr: :8080
  handler:
    type: auto
    chain: chain-0
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    selector:
      strategy: rand
      maxFails: 1
      failTimeout: 10s
    nodes:
    - name: node-0
      addr: :8081
      metadata:
        weight: 20 
      connector:
        type: http
      dialer:
        type: tcp
    - name: node-1
      addr: :8082
      metadata: 
        weight: 10
      connector:
        type: http
      dialer:
        type: tcp

Set weights on nodes (or chains) via the metadata.weight option. The weight ratio of node-0 to node-1 is 2:1, so node-0 is twice as likely to be selected as node-1.

Hash Strategy¶

Hash strategy is based on the Hash value of a specific data to select. The current Hash type supports the client IP and the request target host address, and the client IP is used by default.

gost -L http://:8080 -F "socks5://192.168.1.1:1080,192.168.1.2:1080?strategy=hash"

Target Address Hash Value¶

Each service can set the hash type individually.

CLIFile (YAML)

gost -L http://:8080?hash=host -F "socks5://192.168.1.1:1080,192.168.1.2:1080?strategy=hash&maxFails=1&failTimeout=10s"

services:
- name: service-0
  addr: ":8080"
  handler:
    type: http
    chain: chain-0
    metadata:
      hash: host
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    selector:
      strategy: hash
      maxFails: 1
      failTimeout: 10s
    nodes:
    - name: node-0
      addr: 192.168.1.1:1080
      connector:
        type: socks5
      dialer:
        type: tcp
    - name: node-1
      addr: 192.168.1.2:1080
      connector:
        type: socks5
      dialer:
        type: tcp

Specify the hash type as host with the hash option.

Race Strategy¶

3.3.0

The race strategy (parallel) concurrently dials all nodes in the node group and uses the first successfully established connection, closing the remaining ones. This strategy is ideal for latency-sensitive scenarios, automatically selecting the fastest-responding node.

services:
- name: service-0
  addr: :8080
  handler:
    type: auto
    chain: chain-0
  listener:
    type: tcp
chains:
- name: chain-0
  hops:
  - name: hop-0
    selector:
      strategy: parallel
      maxFails: 1
      failTimeout: 10s
    nodes:
    - name: node-0
      addr: 192.168.1.1:1080
      connector:
        type: socks5
      dialer:
        type: tcp
    - name: node-1
      addr: 192.168.1.2:1080
      connector:
        type: socks5
      dialer:
        type: tcp

Difference from load balancing

Unlike traditional load balancing, the race strategy does not distribute requests across nodes. Instead, it attempts to connect to all nodes simultaneously and uses the first one that succeeds. Therefore, it works best with a small number of nodes (2-3) in latency-sensitive scenarios. For larger node pools or traffic distribution needs, use round robin (round) or random (rand) strategies instead.

Multiplexing disabled

The race strategy's Multiplex() method is hardcoded to return false, meaning multiplexing is forcibly disabled. This is because each request may go through a different node, making it impossible to guarantee that subsequent requests reuse the same established connection.

Selector¶

Forwarding Chain¶

Forwarder¶

Chain Group¶

Backup Node and Chain¶

Backup Nodes¶

Backup Chains¶

Weighted Random Selection Strategy¶

Hash Strategy¶

Target Address Hash Value¶

Race Strategy¶

Comments