hamishforbes / lua-resty-upstream Goto Github PK

Your example returns 500 on the static files:

2014/02/18 10:00:28 [error] 18581#0: *926 lua entry thread aborted: runtime error: /usr/local/openresty/nginx/lua/load-balancer.lua:56: attempt to call local 'reader' (a nil value)
stack traceback:
coroutine 0:
    /usr/local/openresty/nginx/lua/load-balancer.lua: in function </usr/local/openresty/nginx/lua/load-balancer.lua:1>, client: 192.168.0.219, server: balancer, request: "GET /i/new/sprites.png HTTP/1.1", host: "balancer"

in this code:

local reader = res.body_reader
repeat
    local chunk, err = reader(65536)
    if err then
      ngx_log(ngx_ERR, "Read Error: "..(err or ""))
      break
    end

    if chunk then
      print(chunk)
      flush(true)
    end
until not chunk

I'm changed code to this and it's ok, but i don't sure that this change don't makes the hidden bugs.

local reader = res.body_reader
if reader then
    repeat
        local chunk, err = reader(65536)
        if err then
          ngx_log(ngx_ERR, "Read Error: "..(err or ""))
          break
        end

        if chunk then
          print(chunk)
          flush(true)
        end
until not chunk
end

Hey, could we get a new release / tag with those commits?

upstream_socket  = require("resty.upstream.socket")
upstream_api = require("resty.upstream.api")

upstream, configured = socket_upstream:new("my_upstream_dict")
api = upstream_api:new(upstream)

In your example, you use socket_upstream and not upstream_socket

Can this be used with lua-resty-redis to connect to an upstream of redis servers? If yes, provide me some example, if you please. Thanks in advance.

I need to have several different pools (with different hosts) for load balancing on the different locations. How i can specify what pool will handle this request?

Any plans to support additional load balancing schemes like hashing, least conn, etc? Regular old hashing wouldn't seem too hard (same with least_conn); consistent (ketama) hashing has already been done in https://github.com/agentzh/lua-resty-chash; perhaps we could wrap this? I'll happily contribute if you think this is worthwhile (I'd love to be able to wrap this lib using chashing for lua-resty-waf to connect to a cluster of persistent variable storage backends).

Hi,

As of the last commits I'm not able to run my code (largely inspired from the basic example)

nginx: [error] init_by_lua_file error: lib/resty/upstream/socket.lua:229: attempt to index a nil value
stack traceback:
lib/resty/upstream/socket.lua:229: in function 'calc_gcd_weight'
lib/resty/upstream/socket.lua:244: in function 'save_pools'
lib/resty/upstream/api.lua:307: in function 'add_host'
app/init.lua:80: in function <app/init.lua:31>

I was actually in the process of doing the same modification you just made (peer management per worker to get more predictable behavior)
I based the algorithm from Nginx's ngx_http_upstream.c, and it seems to work OK.
Here is the commit : mtourne@3e81df5

Oh and, may I suggest this to get really fast serialization / de-serialization :
https://github.com/cloudflare/lua-capnproto

I configured primary and backup pools. like this:

if not configured then -- Only reconfigure on start, shared mem persists across a HUP
            api:create_pool({id = "primary", timeout = 1000, read_timeout = 30000, keepalive_pool = 256, keepalive_timeout = (120*1000)})

            api:create_pool({id = "backup", timeout = 1000, read_timeout = 30000, keepalive_pool = 256, keepalive_timeout = (120*1000)})

            api:set_priority("primary", 0)
            api:set_priority("backup", 10)

            api:add_host("backup", { host = "192.168.0.2", port = "80", weight = 10, healthcheck = true})
            api:add_host("primary", { host = "192.168.0.3", port = "80",  weight = 5, healthcheck = true})
end

And after several requests i have this errors in log:

2014/03/12 07:27:09 [error] 19483#0: [lua] http.lua:165: check_response(): closed from 1, context: ngx.timer
2014/03/12 07:27:09 [error] 19483#0: [lua] http.lua:165: check_response(): closed from 1, context: ngx.timer
2014/03/12 07:27:09 [error] 19484#0: [lua] http.lua:165: check_response(): closed from 1, context: ngx.timer
2014/03/12 07:27:09 [error] 19484#0: [lua] http.lua:165: check_response(): closed from 1, context: ngx.timer
2014/03/12 07:27:09 [error] 19484#0: [lua] socket.lua:196: post_process(): Host "1" in Pool "primary" is down, context: ngx.timer
2014/03/12 07:27:09 [error] 19484#0: [lua] socket.lua:196: post_process(): Host "1" in Pool "backup" is down, context: ngx.timer

I made this several requests from browser and all was seemed ok.

Can you add the handled request statistics for pool and hosts in pool?
Somethink like this:
add requests counter into default_pool and default_host tables:

local default_pool = {
    up = true,
    method = 'round_robin',
    timeout = 2000, -- socket connect timeout
    priority = 0,
    failed_timeout = 60,
    max_fails = 3,
    requests = 0,
    hosts = {}
}
local default_host = {
    host = '',
    port = 80,
    up = true,
    weight = 0,
    failcount = 0,
    lastfail = 0,
    requests = 0
}

Then store host id, which successfully handled request, in ngx.ctx (like you store failed hosts in ngx.ctx). And increment requests counter on the pool and host in log_by_lua phase.
This is possible?

Dear Developers

I'm sorry for my silly question.

I Have newest OpenResty installed.
in sort, I (when access http://127.0.0.1:8080/pools) already got :

{"primary":{"keepalive_pool":256,"hosts":[{"host":"127.0.0.1","weight":10,"port":81,"id":"1","healthcheck":{"last_check":1447815898.606,"interval":60},"failcount":0,"up":true,"lastfail":0},{"host":"127.0.0.1","weight":10,"lastfail":0,"id":"2","healthcheck":{"last_check":1447815898.606,"interval":60},"failcount":0,"up":true,"port":82}],"up":true,"read_timeout":10000,"method":"round_robin","failed_timeout":60,"timeout":100,"keepalive_timeout":120000,"priority":0,"max_fails":3}}

Note : I striped out all ssl parts.

for port 80, I have :

    server {
        listen 80;
        server_name lua-load-balancer;

        location / {
            content_by_lua_file /usr/local/openresty/nginx/html/lua-load-balancer.lua;
        }
    }

yes I copied lua-load-balancer.lua to that directory and chmod +x.

But when I access http://127.0.0.1 , I got 404

Kindly please give me any enlightenment.

Sincerely
-bino-

After several hours of load balancing with 2 hosts, i have 504 in all requests and this errors:

2014/02/24 13:33:51 [error] 5432#0: *17441 [lua] http.lua:231: request(): Upstream Error: 504, client: X.X.X.X, server: example.com, request: "GET / HTTP/1.1", host: "example.com"

But these 2 hosts is ok. Then i restart nginx (only restart) - all is well.
And after several hours this situation repeats.

When I test this module, I found a error when sending request over HTTP2.0

load-balancer.lua:21: Error getting client body reader: http v2 not supported yet, client: 10.10.10.1, server: , request: "GET /favicon.ico HTTP/2.0", host: "www.example.cc"

This will cause all POST request failed. It seems like the reason is the function ngx_http_lua_ngx_req_raw_header and ngx_http_lua_req_socket in lua-nginx-module do not support SPDY and H2 yet.

So is there any way to fix this ?

BTW, thank u for this module, I learn a lot from this.