Hello everybody,

shortly after starting haproxy on one of our servers, haproxy goes to
100% and stays there.

We have multiple instances of haproxy running on various servers, but
only one instance is consistently running at 100%.

We are running

* 3 instances with 1.4.24-1~bpo70+1  (Debian wheezy-backports)
* 2 instances with 1.4.18-0ubuntu1.2 (Ubuntu precise)

However only the one Debian instance is having the problem.

Note that haproxy 1.4.24-1~bpo70+1 should contain the latest CPU related
fixes discussed here.

When I strace that instance, it's showing lots and lots of:

  epoll_wait(0, {}, 200, 0)               = 0

Thus is seems very similar to the problem discussed here [1], which
seems not to have reached any solution (?).

I've seen this problem only lately. And there are two things that
happened lately:

* we upgraded from Ubuntu's 1.4.15-1ubuntu0.1 to Debian's 1.4.24-1~bpo70+1
* we've started to use TPROXY

Here's our configuration (all IPs have been replaced):

#############################################
global
    log             localhost    local0
    maxconn         10000
    user            root
    group           root
    #pidfile        /var/run/haproxy.pid
    stats socket /var/run/haproxy_stats user haproxy group haproxy mode
660 level admin
    daemon

defaults
    mode            http
    balance         roundrobin
    log             global
    #option httplog [ clf ]        format CLF pro analyzeri http logu
    option          dontlognull
    # =  Change log level for non-completely successful connections
    option          log-separate-errors
    option          abortonclose
    option          redispatch
    retries         3
    option          httpchk HEAD /keepalive.html HTTP/1.0
    timeout check            3s
    timeout         connect 30s
    timeout         client  30s
    timeout         server  30s

listen smtp 4.3.2.1:25,4.3.2.1:465 #smtp ssmpt
    mode tcp
    maxconn 100
    source 0.0.0.0 usesrc clientip
    server          serverm 192.168.1.1

listen imap 4.3.2.1:143,4.3.2.1:993,4.3.2.1:4190 #imap imaps esieve
    mode tcp
    maxconn 100
    timeout         client  12h
    timeout         server  12h
    source 0.0.0.0 usesrc clientip
    server          serverm 192.168.1.1

# --- scm.sourcepole.ch ---

listen git-ssh 4.3.2.1:22
    mode tcp
    maxconn 100
    timeout         client  30m
    timeout         server  30m
    server          servers 192.168.1.2

listen git 4.3.2.1:9418
    mode tcp
    maxconn 100
    server          servers 192.168.1.2

backend stats
    stats           enable
    stats           uri /
    stats           auth foo:bar
    stats           realm Lalala
    stats           refresh 2

#############################################

Any tips/ideas on how to proceed from here?
*t

[1] http://www.mail-archive.com/[email protected]/msg08467.html

Reply via email to