Bug? Closing upstream fastcgi requests on 0.8.15/0.7.62

Server NGINX
1

Hi all,

I’m seeing a massive spike of nginx connections in read state, which
appears
to be related to fastcgi upstream connections [1]. Usually I’ll have 0.6
avg
and ~25 max connections in read state, but with 0.8.15/0.7.62 I’ve seen
34.0
avg and 52.4 max… and it looks like my fastcgi upstream doesn’t like
that
very much. :wink:

Thoughts?

Thanks,

  • Jeff

[1] I reverted to proxy_pass and the read state connections disappeared.

2

On Tue, Sep 15, 2009 at 08:35:41PM +1000, Jeff W. wrote:

Hi all,

I’m seeing a massive spike of nginx connections in read state, which appears
to be related to fastcgi upstream connections [1]. Usually I’ll have 0.6 avg
and ~25 max connections in read state, but with 0.8.15/0.7.62 I’ve seen 34.0
avg and 52.4 max… and it looks like my fastcgi upstream doesn’t like that
very much. :wink:

Thoughts?

The read state is just reading request (not including body).
This state is before any fastcgi interaction.

3

I’m seeing a massive spike of nginx connections in read state, which appears
to be related to fastcgi upstream connections [1]. Usually I’ll have 0.6 avg
and ~25 max connections in read state, but with 0.8.15/0.7.62 I’ve seen 34.0
avg and 52.4 max… and it looks like my fastcgi upstream doesn’t like that
very much. :wink:

Ouch, looks like I spoke too soon – it’s happening again without
fastcgi in
active use. Here’s what the connections in read state chart looks like:

http://www.gnome.org/~jdub/2009/nginx-0.7.62-connread.png

There could be a slight lagging correlation with the active
connections
chart, showing spikes of connections left in the read state about a
minute
after similar spikes in active connections… but it’s definitely not a
one-to-one relationship, even excluding the static files being served.

  • Jeff
4

On Tue, Sep 15, 2009 at 09:44:41PM +1000, Jeff W. wrote:

http://www.gnome.org/~jdub/2009/nginx-0.7.62-connread.png

There could be a slight lagging correlation with the active connections
chart, showing spikes of connections left in the read state about a minute
after similar spikes in active connections… but it’s definitely not a
one-to-one relationship, even excluding the static files being served.

Many reading states may indicate some network issues.