feat: support dns parse in stream subsystem

[NJLOVER]

Current Behavior

i add a stream_route with host gatorcloud-pg.postgres-system.svc.cluster.local( first i use gatorcloud-pg.postgres-system, it doesn't work too )
{
"server_port": 9102,
"upstream": {
"type": "roundrobin",
"nodes": {
"gatorcloud-pg.postgres-system.svc.cluster.local:5432": 1
}
}
}

the error log is:

2022/08/18 07:46:16 [error] 45#45: *13383 stream [lua] resolver.lua:47: parse_domain(): failed to parse domain: gatorcloud-pg.postgres-system.svc.cluster.local, error: failed to query the DNS server: dns client error: 101 empty record received while prereading client data, client: 10.233.102.128, server: 0.0.0.0:9103
2022/08/18 07:46:16 [error] 45#45: *13383 stream [lua] upstream.lua:79: parse_domain_for_nodes(): dns resolver domain: gatorcloud-pg.postgres-system.svc.cluster.local error: failed to query the DNS server: dns client error: 101 empty record received while prereading client data, client: 10.233.102.128, server: 0.0.0.0:9103
2022/08/18 07:46:16 [error] 45#45: *13383 stream [lua] init.lua:942: stream_preread_phase(): failed to set upstream: no valid upstream node while prereading client data, client: 10.233.102.128, server: 0.0.0.0:9103

the same dns for route, it is ok.

{
"uri": "/test",
"methods": ["GET","POST","PUT","DELETE"],
"priority": 10000112,
"upstream": {
"type": "roundrobin",
"scheme": "http",
"nodes": {
"gatorcloud-pg.postgres-system.svc.cluster.local:5432": 1
}
}
}

2022/08/18 07:53:31 [error] 46#46: *67040 upstream prematurely closed connection while reading response header from upstream, client: 10.233.102.128, server: _, request: "GET /test HTTP/1.1", upstream: "http://10.233.19.207:5432/test", host: "172.30.3.230"

this log mean "gatorcloud-pg.postgres-system.svc.cluster.local:5432" resolve to 10.233.19.207:5432 in lv4.

Expected Behavior

dns resolve error

Error Logs

No response

Steps to Reproduce

use stream_route with a host upstream, then dns resolve error

Environment

apache/apisix:2.14.1-alpine

[NJLOVER]

it is in k8s env. the host is a serviceName.

[tzssangglass]

pls use text not images.

[tzssangglass]

First, show the conclusion: the lib lua-resty-dns-client we used now is not supported to run in the stream subsystem.

https://github.com/Kong/lua-resty-dns-client/blob/master/src/resty/dns/client.lua#L829-L860

  local supported_semaphore_wait_phases = {
    rewrite = true,
    access = true,
    content = true,
    timer = true,
    ssl_cert = true,
    ssl_session_fetch = true,
  }


  local ngx_phase = get_phase()


  if not supported_semaphore_wait_phases[ngx_phase] then
    -- phase not supported by `semaphore:wait`
    -- return existing query (item)
    --
    -- this will avoid:
    -- "dns lookup pool exceeded retries" (second try and subsequent retries)
    -- "API disabled in the context of init_worker_by_lua" (first try)
    return item, nil, try_list
  end


  -- block and wait for the async query to complete
  local ok, err = item.semaphore:wait(poolMaxWait)
  if ok and item.result then
    -- we were released, and have a query result from the
    -- other thread, so all is well, return it
    --[[
    log(DEBUG, PREFIX, "Query sync result: ", key, " ", fquery(item),
           " result: ", json({ result = item.result, err = item.err}))
    --]]
    return item.result, item.err, try_list
  end

When called in the http subsystem, ngx_phase is access, in supported_semaphore_wait_phases, and when called in the stream subsystem, ngx_phase is preread.

This causes the code to go through different branches of logic when run into if not supported_semaphore_wait_phases[ngx_phase] then, and the final dns parsing result, e.g.

This difference causes the lookup function in the stream subsystem to return records such as:

2022/08/19 23:17:31 [warn] 918740#918740: *163 stream [lua] client.lua:1247: resolve(): records : {
  key = "mix.sd.test.local:1",
  qname = "mix.sd.test.local",
  r_opts = {
    additional_section = true,
    qtype = 1
  },
  semaphore = {
    sem = cdata<struct ngx_stream_lua_sema_s *>: 0x7fc9373260b8,
    <metatable> = {
      __index = {
        count = <function 1>,
        new = <function 2>,
        post = <function 3>,
        version = "0.1.23",
        wait = <function 4>
      }
    }
  },
  try_list = { -- ["(short)mix.sd.test.local:(na) - cache-miss","mix.sd.test.local:1 - cache-miss/scheduled"]
     {
      msg = { -- cache-miss
         "cache-miss",
        <metatable> = <1>{
          __tostring = <function 5>
        }
      },
      qname = "(short)mix.sd.test.local"
    }, {
      msg = { -- cache-miss/scheduled
         "cache-miss", "scheduled",
        <metatable> = <table 1>
      },
      qname = "mix.sd.test.local",
      qtype = 1
    },
    ["(short)mix.sd.test.local:nil"] = 1,
    ["mix.sd.test.local:1"] = 2,
    <metatable> = {
      __tostring = <function 6>
    }
  }
} while prereading client data, client: 127.0.0.1, server: 0.0.0.0:9100

and in the http subsystem to return records such as:

2022/08/19 23:17:08 [warn] 918434#918434: *115 [lua] client.lua:1247: resolve(): records : { {
    address = "127.0.0.1",
    class = 1,
    name = "mix.sd.test.local",
    section = 1,
    ttl = 3600,
    type = 1
  },

[tzssangglass]

In my test, I hacked supported_semaphore_wait_phases like:

  local supported_semaphore_wait_phases = {
    rewrite = true,
    access = true,
    content = true,
    timer = true,
    ssl_cert = true,
    ssl_session_fetch = true,
    preread = true,
  }

and it works well.

Have ask about this plan: Kong/lua-resty-dns-client#144