Query regarding "http first read error: EOF" error in varnish

[Jayanth Kalyanasundaram]

On Wed, Jan 13, 2016 at 4:30 PM, Geoff Simmons <geoff at uplex.de> wrote:
>
>
> - -   Timestamp      Bereq: 1452675822.047840 0.015508 0.015508
> - -   FetchError     http first read error: EOF
> - -   BackendClose   232 reload_2016-01-12T07:28:50.cp_12
> - -   Timestamp      Beresp: 1452675876.038544 54.006212 53.990704
>
> This says that about 54 seconds elapsed after Varnish sent the backend
> request until it gave up waiting for the response. That could very
> well be a first byte timeout, especially since the waiting time is so
> close to the round number 54 (although 54 seconds is an unusual value
> for first_byte_timeout).
>
> > The issue doesn't seem to be a connection problem with the backend
> > because a simple curl call of the same URL from the varnish box
> > seems to work fine. So we'd like to know what could be the reason
> > behind the error mentioned above and how we can fix it.
>
> Does Varnish get the FetchError every time, or only sometimes? It
> could have been that your curl tests were "just lucky".
>
 Hey, like you said the issue only happens sometimes. But the
first_byte_timeout for all the backends is set to 300 seconds.

[root at varnish ~]# sudo varnishadm "param.show first_byte_timeout"
first_byte_timeout
        Value is: 300.000 [seconds]
        Default is: 60.000
        Minimum is: 0.000

Also the time difference between the backend request and backend response
seems to be different each time this issue occurs. For instance, in this
case http://pastebin.com/eS2Ldnbv the difference is just 0.6 seconds. So I
don't think it can be an issue with timeouts.

Would this also happen if the backend server gives an empty response to the
request ?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20160114/59561f78/attachment.html>