There is a big performance difference between APISIX 3.10 and 2.13 when no plug-ins are enabled.
After comparing the flame graphs of APISIX 3.10 and 2.13 versions, it can be clearly seen that the biggest difference is in the pick_server function. After comparing the function codes of the two versions, it is found that the code of version 3.10 has added logic to traverse all upstream nodes, and when there are more upstream nodes during performance testing, so the difference is more obvious.

After backtracking the changes, it was found that the changes traversing all upstream nodes were introduced by this pr #7594.

change point

1. Remove

   apisix/apisix/balancer.lua

   Lines 198 to 202 in 73618d4

   	for _, node in ipairs(up_conf.nodes) do
   	if core.utils.parse_ipv6(node.host) and str_byte(node.host, 1) ~= str_byte("[") then
   	node.host = '[' .. node.host .. ']'
   	end
   	end

   from the request path, and escape it to upstream's check_schema function to implement this logic
2. Illegal ipv6 addresses are directly rejected in the admin api

• I have explained the need for this PR and the problem it solves
• I have explained the changes or the new features added to this PR
• I have added tests corresponding to this change
• I have updated the documentation to reflect this change
• I have verified that this change is backward compatible (If not, please discuss on the APISIX mailing list first)

Note: This change is backwards compatible. When detecting data from etcd, it modifies the host for compatibility

2025/10/30 16:17:45 [warn] 2277252#2277252: *1 [lua] upstream.lua:408: checker(): PREV CONF.NODES : { {
    host = "::1",
    port = 80,
    priority = 0,
    weight = 1
  },
  <metatable> = {}
}, context: init_worker_by_lua*
2025/10/30 16:17:45 [warn] 2277252#2277252: *1 [lua] upstream.lua:414: checker(): AFTER CONF.NODES : { {
    host = "[::1]",
    port = 80,
    priority = 0,
    weight = 1
  },
  <metatable> = {}