Re: Random “http first read error: EOF” errors

[Guillaume Quintard]

Can you try something: add 'set bereq.http.connection = "Close"; ' at the
beginning of vcl_backend_fetch and see if that helps?

-- 
Guillaume Quintard

On Thu, Mar 30, 2017 at 1:04 PM, Hazar Güney <hazarguney at gmail.com> wrote:

> MaxKeepAliveRequests 20
> KeepAliveTimeout 2
>
> Version is "4.1.3 revision 5e3b6d2". We have also seen "straight
> insufficient bytes" error with POST requests to a specific php script
> hosted by another backend and fixed it by using "pipe" instead of "pass"
> but this specific backend gives "http first read error: EOF" error. Another
> example from today:
>
> *   << BeReq    >> 126635444
> -   Begin          bereq 126635443 fetch
> -   Timestamp      Start: 1490870598.921499 0.000000 0.000000
> -   BereqMethod    GET
> -   BereqURL       XXXX
> -   BereqProtocol  HTTP/1.1
> -   BereqHeader    Host: XXXX
> -   BereqHeader    User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64)
> AppleWebKit/537.36 (KHTML, like Gecko) Chrome/56.0.2924.87 Safari/537.36
> -   BereqHeader    Accept: image/webp,image/*,*/*;q=0.8
> -   BereqHeader    Referer: XXXX
> -   BereqHeader    Accept-Language: tr-TR,tr;q=0.8,en-US;q=0.6,en;q=0.4
> -   BereqHeader    RIP: XXXX
> -   BereqHeader    X-Forwarded-For: XXXX
> -   BereqHeader    Accept-Encoding: gzip
> -   BereqHeader    X-Varnish: 126635444
> -   VCL_call       BACKEND_FETCH
> -   VCL_return     fetch
> -   BackendOpen    35 reload_2017-03-20T11:32:44.st2 10.35.78.11 80
> 172.17.0.2 48896
> -   BackendStart   10.35.78.11 80
> -   Timestamp      Bereq: 1490870598.922050 0.000552 0.000552
> *-   FetchError     http first read error: EOF*
> -   BackendClose   35 reload_2017-03-20T11:32:44.st2
> -   Timestamp      Beresp: 1490870598.922622 0.001124 0.000572
> -   Timestamp      Error: 1490870598.922627 0.001129 0.000005
> -   BerespProtocol HTTP/1.1
> -   BerespStatus   503
> -   BerespReason   Service Unavailable
> -   BerespReason   Backend fetch failed
> -   BerespHeader   Date: Thu, 30 Mar 2017 10:43:18 GMT
> -   BerespHeader   Server: Varnish
> -   VCL_call       BACKEND_ERROR
> -   BereqHeader    X-Varnish-Backend-5xx: 1
> -   VCL_return     retry
> -   Timestamp      Retry: 1490870598.922657 0.001159 0.000030
> -   Link           bereq 126832283 retry
> -   End
>
> On Wed, Mar 29, 2017 at 12:03 PM, Mattias Geniar <mattias at nucleus.be>
> wrote:
>
>> > Backend is Apache.
>>
>> In older Varnish versions, you could sometimes see a similar error;
>>
>> >   11 FetchError   c straight insufficient bytes
>>
>> The error message you’re seeing might be related, as it mentions the EOF.
>>
>> This happens when the backend sends a Content-Length header that doesn’t
>> match the _actual_ content length it’s sending. In Apache, this was
>> commonly caused by a mod_deflate misconfiguration.
>>
>> For testing, could you try disabling Gzip either in your backend or strip
>> the Accept-Encoding header in Varnish to force a plain text response?
>>
>> Mattias
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20170330/d42768f7/attachment-0001.html>