HAProxy reports L4TOUT in 2001ms even if using mode HTTP (no replies)

HAProxy reports backend as down even if they are alive with a status as
L4TOUT in 2001ms (Stats page)

Version 1.5.2-2.el6

The backend is healthy for about 14-15 hours and then goes down stating the
error L4TOUT

Below is the configuration

mode http
option httpchk GET /abc/__status
server APP_ELB internal-ELB.us-east-1.elb.amazonaws.com:80 maxconn 150 check

I can successfuly do a GET on the backend from the HAProxy instance

# wget internal-ELB.us-east-1.elb.amazonaws.com/abc/__status
--2015-08-15 13:29:47--
http://internal-ELB.us-east-1.elb.amazonaws.com/abc/__status
Resolving internal-ELB.us-east-1.elb.amazonaws.com... 10.x.x.x, 10.x.x.x
Connecting to internal-ELB.us-east-1.elb.amazonaws.com|10.x.x.x|:80...
connected.
HTTP request sent, awaiting response... 200 OK
Length: 1239 (1.2K) [text/plain]
Saving to: “__status.2”

100%[====================================================================>]
1,239 --.-K/s in 0s

2015-08-15 13:29:47 (155 MB/s) - “__status.2” saved [1239/1239]

Any known issues or bugs.
HAProxy reload is enough to get it back alive