lua-resty-http library - async behavior questions

smadhavan · 2018-06-07T10:01:42+00:00

Forwarding to the group:> Interesting, when I update the script to print res.body and pipe it through jq, I get the response below. Is that what you mean...

lua-resty-http library - async behavior questions

smadhavan

Hello !

We use balancer_by_lua which invokes lua code to select an upstream server based on chash. ngx_timer_at is used by one worker to send requests to etcd server in every 10 secs interval.

Lua specific code for sending http request is as follows.

local function fetch(url)

local client = http:new()

client:set_timeout(10000)

client:connect(_M.conf.etcd_host, _M.conf.etcd_port)

local res, err = client:request({path=url, method="GET"})

if err then

return nil, err

end

local body, err = res:read_body()

if err then

return nil, err

end

-- testloginfo("fetch response body: " .. body)

local ok, data = "" body)

if not ok then

testlogerr("In fetch, pcall or json decode failed, body: " .. body)

return nil, data

end

data.etcdIndex = res.headers["x-etcd-index"]

-- testloginfo("fetch end of function, data.etcdIndex: " .. data.etcdIndex)

return data, nil

end

We are seeing high response time in load testing and was trying to understand whether lua-http-resty API calls are blocking the nginx workers.

specific questions are:

1. Is the client:connect and client:request API's blocking calls ?

2. How does res:read_body reads the response from server without blocking ?

Thanks,

Sreekanth

hamish.forbes

You should probably do some profiling and generate some flamegraphs to figure out what's actually causing slowdowns, guessing is gonna be hard... http://openresty.org/en/profiling.html

1) resty-http uses the ngx.socket.tcp API https://github.com/openresty/lua-nginx-module#ngxsockettcp

So at a minimum every network operation will yield the worker allowing another request to do stuff

2) same

How large is your JSON response? AFAIK the json decode call will be blocking so if its huge that could cause a problem.

Again you should profile to find out though.

On Thursday, 7 June 2018 03:01:42 UTC+1, Sreekanth Madhavan wrote:

Hello !

We use balancer_by_lua which invokes lua code to select an upstream server based on chash. ngx_timer_at is used by one worker to send requests to etcd server in every 10 secs interval.

Lua specific code for sending http request is as follows.

local function fetch(url)
local client = http:new()
client:set_timeout(10000)
client:connect(_M.conf.etcd_host, _M.conf.etcd_port)

local res, err = client:request({path=url, method="GET"})
if err then
return nil, err
end

local body, err = res:read_body()
if err then
return nil, err
end
-- testloginfo("fetch response body: " .. body)

local ok, data = "" body)
if not ok then
testlogerr("In fetch, pcall or json decode failed, body: " .. body)
return nil, data
end

data.etcdIndex = res.headers["x-etcd-index"]
-- testloginfo("fetch end of function, data.etcdIndex: " .. data.etcdIndex)
return data, nil
end

We are seeing high response time in load testing and was trying to understand whether lua-http-resty API calls are blocking the nginx workers.

specific questions are:

1. Is the client:connect and client:request API's blocking calls ?

2. How does res:read_body reads the response from server without blocking ?

Thanks,
Sreekanth

smadhavan

Hi Hamish,

Thanks for the response. max json body response is 1KB. Please note that in the load tests the read_body API returns nil every time as etcd server sends Content Length 0 in the response. (as there is no data change in etcd server)

So we end up in a case where read_body returns nil or timeout after 10 secs. Can this cause any issue ?

I will do profiling and publish the flamegraghs.

Thanks,

-Sreekanth

On Jun 7, 2018, at 6:08 AM, Hamish Forbes <hamish...@gmail.com> wrote:

You should probably do some profiling and generate some flamegraphs to figure out what's actually causing slowdowns, guessing is gonna be hard... http://openresty.org/en/profiling.html

1) resty-http uses the ngx.socket.tcp API https://github.com/openresty/lua-nginx-module#ngxsockettcp
So at a minimum every network operation will yield the worker allowing another request to do stuff

2) same

How large is your JSON response? AFAIK the json decode call will be blocking so if its huge that could cause a problem.
Again you should profile to find out though.

On Thursday, 7 June 2018 03:01:42 UTC+1, Sreekanth Madhavan wrote:
Hello !

We use balancer_by_lua which invokes lua code to select an upstream server based on chash. ngx_timer_at is used by one worker to send requests to etcd server in every 10 secs interval.

Lua specific code for sending http request is as follows.

local function fetch(url)
local client = http:new()
client:set_timeout(10000)
client:connect(_M.conf.etcd_host, _M.conf.etcd_port)

local res, err = client:request({path=url, method="GET"})
if err then
return nil, err
end

local body, err = res:read_body()
if err then
return nil, err
end
-- testloginfo("fetch response body: " .. body)

local ok, data = "" body)
if not ok then
testlogerr("In fetch, pcall or json decode failed, body: " .. body)
return nil, data
end

data.etcdIndex = res.headers["x-etcd-index"]
-- testloginfo("fetch end of function, data.etcdIndex: " .. data.etcdIndex)
return data, nil
end

We are seeing high response time in load testing and was trying to understand whether lua-http-resty API calls are blocking the nginx workers.

specific questions are:

1. Is the client:connect and client:request API's blocking calls ?

2. How does res:read_body reads the response from server without blocking ?

Thanks,
Sreekanth

--
.