Project

General

Profile

Actions

Bug #1721

closed

Requests are not processed, fcgi load increases, but processes are idle and server load minimum

Added by c2h5oh over 16 years ago. Updated about 16 years ago.

Status:
Invalid
Priority:
High
Category:
mod_fastcgi
Target version:
ASK QUESTIONS IN Forums:

Description

I am using lighty with php 5.2.6 run via fcgi.

From time to time lighty stops responding to request handled by fcgi. Static content is served ok.

All php-cgi processes are idle, as if request were not sent to be processed, connection status remains at handle-request. FCGI backend load increases as new requests come. Every 5-10 seconds a single request is processed or partially processed.

By partially processed I mean that sometimes I have 2-5 request hanging at read getting ~200 bytes every couple of seconds.

Killing lighty does not help. Killing lighty AND all php-cgi processes (and I do check they have been successfully killed) does not help - once restarted lighty processes <50 fcgi requests and returns to the previous state.

Sometimes problem resolves by itself, but no sooner then after 600seconds (might not be accurate - I have observed carefully only 4 incidents and that was true in 3 cases out of 4, one of which I am sure it was 600 +/-5s and other 2 more like 600-ish s).

The problem seems to be caused by something other than initial load spike: if I kill both lighty and all php-cgi processes and start it back again it is not solved. If I restart the whole server it is immediately fine and stays fine for another day or so.
I failed to find any time pattern matching failures - there seems to be little or no connection with any processes run via cron or server load.
It is not a simple server overload - load haven't exceeded 50%, and remained well below 10% during failure.

System info:
Debian 3.1 (32-bit)
Lighty 1.4.19 with this patch http://trac.lighttpd.net/trac/ticket/897
PHP 5.2.6 +xcache 1.2.2

Actions #1

Updated by c2h5oh over 16 years ago

M'kay, the 600s theory is no more - it was pure coincidence. Another thing: I have split websites across 2 fcgi sockets (on same server) and observed one working fine, and the other not (at the same time).
Another couple divisions later I narrowed it down to a single website, which has not been modified for last 9 months, and was working fine.

Actions #2

Updated by stbuehler over 16 years ago

  • Status changed from New to Fixed
  • Resolution set to invalid

You are using an unoffical (imho broken) patch, so perhaps you should complain about the patch in the other ticket.

Actions #3

Updated by stbuehler about 16 years ago

  • Status changed from Fixed to Invalid
Actions

Also available in: Atom