有一个API平台服务，所有接口都通过API平台转发到实际的服务上，然后再把实际服务的结果返回给客户端，API平台的规范是所有实际服务的接口返回都要统一结构为

{

"code": 0,

"msg": "",

"data": {}

}

否则无法处理，现在使用influxdb提供的api，他的返回结构并不是API平台需要的结构，所以需要做一层代理转发请求并修改返回结果为API平台需要的结构。

采用openresty来实现这个需求，只需安装好openresty，然后编写一个配置文件即可实现。

实现配置如下：ngx_lua_restructure_influxdb_response.conf

upstream influxdb{

server 127.0.0.1:8086;

}

server {

listen 18086;

proxy_pass_request_headers off; # handle gziped capture

location / {

proxy_pass http://influxdb/query;

}

location /query {

add_header Content-Type 'application/json; charset=utf-8'; # make sure content-type is json

content_by_lua_block {

local method_map = {

GET = ngx.HTTP_GET,

POST = ngx.HTTP_POST,

}

ngx.req.read_body()

local method = method_map[ngx.req.get_method()]

local data = ngx.req.get_body_data()

local args = ngx.req.get_uri_args()

local res = ngx.location.capture("/", {method=method, data=data, args=args})

local cjson = require "cjson"

local new_body = {}

if res.status ~= ngx.HTTP_OK then

new_body["code"] = res.status

else

new_body["code"] = 0

end

new_body["msg"] = ''

new_body["data"] = cjson.decode(res.body)

ngx.say(cjson.encode(new_body))

}

}

}

influxdb的api在8086端口，所有直接的请求返回结果为：

# curl -Gs '100.113.164.108:8086/query?db=data&q=SELECT+*+FROM+cpu+limit+1&pretty=true'

{

"results": [

{

"statement_id": 0,

"series": [

{

"name": "cpu",

"columns": [

"time",

"cpu",

"host",

"usage_guest",

"usage_guest_nice",

"usage_idle",

"usage_iowait",

"usage_irq",

"usage_nice",

"usage_softirq",

"usage_steal",

"usage_system",

"usage_user"

],

"values": [

[

"2017-12-08T08:30:00Z",

"cpu-total",

"Tencent-SNG",

0,

0,

96.92993315559063,

0.4456548651008583,

0,

0,

0,

0,

1.8321366676828983,

0.7922753157508718

]

]

}

]

}

]

}

经过18086端口的转发后得到的结果为：

# curl -Gs '100.113.164.108:18086/query?db=db_zhiyun&q=SELECT * FROM cpu limit 1&pretty=true' | python -m json.tool

{

"code": 0,

"data": {

"results": [

{

"series": [

{

"columns": [

"time",

"cpu",

"host",

"usage_guest",

"usage_guest_nice",

"usage_idle",

"usage_iowait",

"usage_irq",

"usage_nice",

"usage_softirq",

"usage_steal",

"usage_system",

"usage_user"

],

"name": "cpu",

"values": [

[

"2017-12-08T08:30:00Z",

"cpu-total",

"Tencent-SNG",

0,

0,

96.929933155591,

0.44565486510086,

0,

0,

0,

0,

1.8321366676829,

0.79227531575087

]

]

}

],

"statement_id": 0

}

]

},

"msg": ""

}

实现中遇到的坑：

使用curl请求可以得到正确结果，但是通过API平台过来的请求全部报错，错误信息为在capture后拿到的body是一传乱码，并不是json。

使用ngx.req.raw_header()查看curl和API平台请求的header信息的区别，发现curl没有设置Accept-Encoding，而API平台的请求都带有Accept-Encoding: gzip, deflate，导致capture子请求(即请求真实influxdb接口)的返回结果被gzip压缩后把json变成了乱码，

从而导致在后面cjson.decode的时候出错。

OPTION ONE

Placed:

more_clear_input_headers Accept-Encoding;

in "location /cap2"

OPTION TWO

Placed:

proxy_pass_request_headers off;

in "location /cap2"

OPTION THREE

Downloaded https://github.com/brimworks/lua-zlib :

make linux

cp zlib.so /usr/lib/lua/5.1/zlib.so

added following to conf:

local zlib = require("zlib")

local stream = zlib.inflate()

local inflated_body = stream(res1.body)

local res2 = ngx.location.capture("/cap2?"..inflated_body)

由于外层的nginx配置默认的content-type为application/octet-stream;，所以为了确保返回为json，又在/query中明确的设置了content-type