|
On Tuesday 20 December 2005 Joseph Mack NA3T wrote:
> On Tue, 20 Dec 2005, Jan Abraham wrote:
> > This setup works perfectly most of the time, but sometimes
> > the database servers drops tcp connection requests
> > originating from the web servers.
>
> How do you get the director to load balance requests from
> the realservers to the database machines? A different VIP on
> the director?
Okay, lets go into the details, i'll explain it on a stripped down version of
our setup:
DB realservers uses IP 10.0.5.1-.5
DBVIPs are 10.0.4.1 and .2
Directors are 10.0.1.51 and .52
dblb1 ~ # uname -srip
Linux 2.6.12-gentoo-r6 Intel(R) Pentium(R) 4 CPU 2.40GHz GenuineIntel
dblb1 ~ # ip addr
1: ethInt: <BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast qlen 1000
link/ether do:nt:te:ll:it:ha brd ff:ff:ff:ff:ff:ff
inet 10.0.1.51/20 brd 10.255.255.255 scope global ethInt
inet 10.0.4.1/32 scope global ethInt
inet 10.0.4.2/32 scope global ethInt
2: lo: <LOOPBACK,UP> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 brd 127.255.255.255 scope host lo
dblb1 ~ # ipvsadm -L --sort
IP Virtual Server version 1.2.1 (size=1048576)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP dbv1:mysql wrr
-> db2:mysql Route 100 0 0
-> db3:mysql Route 100 0 0
TCP dbv2:mysql wrr
-> db4:mysql Route 100 0 0
-> db5:mysql Route 100 0 0
(db1 is mysql replication master and is accessed directly via its rip,
replicating tables to db2-5; db2+3 serves for longlasting selects, db4+5
serves fast selects , thus two different vips are needed)
db2 ~ # uname -srip
Linux 2.6.14-gentoo-r2 Intel(R) Xeon(TM) CPU 3.06GHz GenuineIntel
db2 ~ # ip addr
1: ethInt: <BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast qlen 1000
link/ether do:nt:te:ll:it:ha brd ff:ff:ff:ff:ff:ff
inet 10.0.5.2/20 brd 10.0.15.255 scope global ethInt
2: lo: <LOOPBACK,UP> mtu 16436 qdisc noqueue
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 brd 127.255.255.255 scope host lo
inet 10.0.4.1/32 brd 10.0.4.1 scope global lo
inet 10.0.4.2/32 brd 10.0.4.2 scope global lo
db2 ~ # arptables -L -n
Chain INPUT (policy ACCEPT)
Chain OUTPUT (policy ACCEPT)
-j DROP -s 10.0.4.1
-j DROP -s 10.0.4.2
Chain FORWARD (policy ACCEPT)
(You may notice that both virtual IPs are set up on all realservers. This is
to be easily able to switch a realserver from dbv1 to dbv2; i don't think it
can be a cause for the strange behaviour).
>
> > The tcp syn packets went from the webserver through the
> > ipvs machine, the mac adresses are replaced there and the
> > packet arrived at one of the database servers (it's
> > visible in tcpdump at the database server and looks
> > correct).
>
> ie has the correct IP etc?
Everything is setup the way it should and works for 99.9% of the connects. I
think it's more likely a problem on the realserver, but I have absolutely no
idea...
> this is a new one.
Umph. Any clues?
Jan
|
