Hi.
This issue is already partially mentioned in here.
Behavior
Certain requests from time to time and up with timeout error.
This code exits with 'timeout' in err variable to be specific.
local prog = exec.new('/tmp/exec.sock')
prog.argv = argv
local res, err = prog()
Environment
I am using openresty with lua-resty-exec and sockexec to expose outputs of some system calls as web service. Openresty is running on small devices like Raspberry Pi.
Cause
At first, it seemed that the error appeares randomly but now it seems that it only appears when there are multiple shell commands executed in a single request.
To be more specific, openresty receives one request, but to handle this request multiple shell commands have to be executed with lua-resty-exec. For each command new instance of exec is created and therefore new socket is created to handle each command also.
Then the function grab_ns is called for each socket to receive data from the socket. The function is called in a loop until all data are read or an error occurs.
For the first time, the function is called with the argument timeout=60000.
In the subsequent calls the function is called with the argument timeout=0 that inside the function transforms into timeout=1here.
This works fine if only one socket is active.
What I think is happening here is, that if multiple sockets are active the worker might not handle the call self.sock:receive() immediately and might switch to different context with a different socket. When it then returns to continue the timeout (1 ms) is already expired.
(Please note that in the documentation of tcpsock:receivehere is set timeout=1000.)
When the socket timeouts in the nginx logs appears similar log like this:
Suggested solution
If I edit the timeout of subsequent calls to self.sock:receive() to for example timeout=100 the issue seems to be solved.
(see patch.diff)
wanrui
commented
5 years ago
Hi. This issue is already partially mentioned in here.
Behavior Certain requests from time to time and up with timeout error.
This code exits with
'timeout'inerrvariable to be specific.Environment I am using openresty with lua-resty-exec and sockexec to expose outputs of some system calls as web service. Openresty is running on small devices like Raspberry Pi.
Cause At first, it seemed that the error appeares randomly but now it seems that it only appears when there are multiple shell commands executed in a single request. To be more specific, openresty receives one request, but to handle this request multiple shell commands have to be executed with lua-resty-exec. For each command new instance of
execis created and therefore new socket is created to handle each command also. Then the function grab_ns is called for each socket to receive data from the socket. The function is called in a loop until all data are read or an error occurs. For the first time, the function is called with the argumenttimeout=60000. In the subsequent calls the function is called with the argumenttimeout=0that inside the function transforms intotimeout=1here.This works fine if only one socket is active. What I think is happening here is, that if multiple sockets are active the worker might not handle the call
self.sock:receive()immediately and might switch to different context with a different socket. When it then returns to continue the timeout (1 ms) is already expired. (Please note that in the documentation oftcpsock:receivehere is settimeout=1000.)When the socket timeouts in the nginx logs appears similar log like this:
Suggested solution If I edit the timeout of subsequent calls to
self.sock:receive()to for exampletimeout=100the issue seems to be solved. (see patch.diff)