HAProxy后端服务器数量过多导致启动/验证时长过长问题求助

Hey there, let's break down why your HAProxy startup time skyrockets to 45s with 50k backends and walk through actionable fixes tailored to your 1.7.5 version:

1. Fully Disable DNS Resolution at Startup #

You’re right to suspect DNS as a key culprit. While you added init-addr none to individual servers, let’s enforce it globally and eliminate any unintended lookup triggers:

• Add init-addr none to your defaults section to apply the setting to all backend servers by default (avoids repeating it for every server line):

  defaults
      init-addr none
      # ... your existing default configurations
  

• Double-check for unused resolvers blocks in your config. If present, HAProxy might still attempt DNS resolution despite init-addr none—remove these blocks or explicitly omit the resolvers parameter from server lines to prevent this.

2. Skip Initial Health Check Validation #

HAProxy runs an initial health check for every server on startup, which adds massive overhead with 50k backends. You can disable or delay these checks:

• Add option init-checks none to your backend (or defaults) section to skip initial health checks entirely:

  backend foo-1234
      option init-checks none
      cookie FOO_1234 insert indirect nocache
      server foo-s-1234-0 server:port check cookie foo-s1234-0 init-addr none
      # ... rest of your server definitions
  

• Alternatively, use slowstart to delay health checks until after startup, paired with longer check intervals to reduce initial load:

  server foo-s-1234-0 server:port check cookie foo-s1234-0 init-addr none slowstart 60s inter 10s
  

3. Add Servers Dynamically via Runtime API #

Instead of loading all 50k servers during startup, start HAProxy with a minimal base config, then add servers dynamically. This eliminates startup latency entirely:

1. Enable the Runtime API in your global config:

   global
       stats socket /var/run/haproxy.sock mode 660 level admin
       # ... your other global settings
   

2. Start HAProxy with just the backend definition (no server lines).
3. Use socat to add servers in bulk (you can script this for efficiency):

   # Example: Add one server via the socket
   echo "add server foo-1234/foo-s-1234-0 server:port cookie foo-s1234-0 init-addr none check" | socat stdio /var/run/haproxy.sock
   

   For 50k servers, generate a list of add server commands in a script and pipe it to the socket—this is far faster than loading all servers at startup.

4. Upgrade HAProxy (Optional) #

HAProxy 1.7.5 is quite outdated (released in 2017). Newer versions (2.0+) include significant optimizations for large backend pools, like faster config parsing and reduced startup overhead. If compatibility allows, upgrading could resolve the issue with minimal config changes (though be sure to test thoroughly first).

内容的提问来源于stack exchange，提问作者westernemergency