MMM 1.2.6 on Redhat ENT 5 64bit, something special

[Tomas]

Hello experts,

I want to ask you what shall I pay much attention while deploy MMM on
Redhat ENT 64bit ?

We will prefer a Redhat 5 ENT 64bit version.

I install MMM 1.2 with successfully, it runs, send_arp and fping all
recompiled.
I use libnet 1.1.4.

But , MMM seems very slow for everythings, like watching the "slow
slices", not like the Ubuntu 32bit version in my test environment.

while i switch a role, it takes at least 2 minutes, it just work as
the 32 bit version. but more like slow the speed , everything response
very slow, while a db is down, it will take almost 5 minutes to
response what happens, and first response is not DB down, but the
replication error from another node.

I want to know, what special requirement for perl on MMM setting up
for Redhat 64 bit ?
IS there some production DB successfully using MMM in redhat ENT 5
64bit ?
Any special package need I install in advance ? I really need your
advice.

Thanks a lot.

Best regards,
Yours Tomas

[Nico Kadel-Garcia]

On Thu, Oct 22, 2009 at 8:55 PM, Tomas <hous...@gmail.com> wrote:
>
> Hello experts,
>
> I want to ask you what shall I pay much attention while deploy MMM on
> Redhat ENT 64bit ?

Do you mean "RHEL", RedHat Enterprise Linux, which is RedHat's current
commercial OS?

[Walter Heck]

We are running MMM 1 and 2 on CentOS 5.3 in production boxes with no
slowness. Is it just MMM that is slow? Or other things as well?

Walter

[Tomas]

Yes, it is RedHat Enterprise Linux 64bit. and very strange, it shows
it response every change of DB status, Server availablility, slow,
just like show you a "slow slice", change a writer role to another
master will take about 2 minutes at least, if you shutdown one db, the
floating ip will be reallocated at least 5 minutes, during these 5
minutes, both floating ip can not be pinged,
that means, writer ip will lost 5 minutes before it is reallocated to
another server, which is bad because we are considering it as a HA
solution.

I am not sure if it is because MMM or something else, I compile fping
and send_arp locally, So I want to know what need I pay more
attention of others, like the fundamental thing, perl, specail package
for 64bit RHEL ?

Thanks a lot.
Yours Tomas

On Oct 22, 6:08 pm, Walter Heck <walterh...@gmail.com> wrote:
> We are running MMM 1 and 2 on CentOS 5.3 in production boxes with no
> slowness. Is it just MMM that is slow? Or other things as well?
>
> Walter
>
> On Fri, Oct 23, 2009 at 08:58, Nico Kadel-Garcia <nka...@gmail.com> wrote:
>

[Walter Heck]

I understand you are running in a cloud, am I right? Also, you didn't
answer my previous question: Is it just MMM or is everything slow?

Walter

[Tomas]

Hi Walter,

It is strange, We finally prove it is not a network issue . because we
try MMM 2.0 later, it is ok for us.

But MMM 1.26 is very slow,network is very normal, we have a private
network to provide floating IPs.
it is not a network problem finally, but MMM 1.2 is very slow for
every response, that is quite strange,
I am wondering if some of the software package I installed maybe not
the right one.

Please share your expertize idea ,

Thanks a lot.

[Walter Heck]

You can try turning on debug mode in the configuration files, to see
if that gives you more information. The ultimate thing to do would be
to use strace to see what's going on..

kind regards,

Walter

[The O.G.]

can you put on a sniffer on both server and client and check layer 2
communications while monitoring the arp tables? your network may be
fully functional, but sometimes certain reverse ARP packets can be
filtered out by default configurations for security reasons. this is
especially suspect if your "slowness or delay" correlates with the
client host ARP cache timeout settings. just a thought...

[Tim Soderstrom]

Hi Tomas!

I apologize if it has already been suggested, but have you added the
hostnames you are using with MMM to your /etc/hosts files? You should
do so on every server. We found that MMM can have weird latency issues
if we do not do this (plus it is easier to debug issues if you have
hostnames setup). Note that these should be in /etc/hosts, NOT on your
DNS server. Doing this avoids MMM from querying any external DNS
servers. Again, we have found this can really help network latency.

Tim

[Tomas]

Dear Walter,

We open the debug, which is the trace log of MMM 1.2.6's agent and
MMM monitor , They work, but response with big latency.

Dear Tim,
I would try your method later, but I am still confuse by that, because
we do not set anything related to hostname , why add /etc/hosts will
make it quicker ?
And why MMM 2.0 works fine ? is that MMM 1.X 's code so bad ??? I do
not think so, maybe I miss something like libnet package or ?

Yours Tomas

[Tim Soderstrom]

Our experience has been such that the hosts you define (usually in
mmm_common) are what you need to add to your /etc/hosts, ie:

host db1
ip XXXXXX
port 3306
user rep_monitor
password XXXXXXX
mode master
peer db2

In this example, you want to make sure 'db1' is in your /etc/hosts and
the IP is the same as the IP you used in the 'ip' section above. We
add these host entries to any server that is part of the MMM config.

[Tomas]

Thanks Tim, I will try it . Will it improve latency dramatically ?

On Oct 26, 12:06 pm, Tim Soderstrom <t...@moocowproductions.org>
wrote:

[Pascal Hofmann]

Hi list,

> Our experience has been such that the hosts you define (usually in
> mmm_common) are what you need to add to your /etc/hosts, ie:
>

> [...]

>
> In this example, you want to make sure 'db1' is in your /etc/hosts and
> the IP is the same as the IP you used in the 'ip' section above. We
> add these host entries to any server that is part of the MMM config.

AFAIK these names are never used outside of MMM.

Enable skip_name_resolve in my.cnf:
http://www.mysqlperformanceblog.com/2008/05/31/dns-achilles-heel-mysql-installation/
http://jcole.us/blog/archives/2006/04/26/on-ips-hostnames-and-mysql/

Regards

Pascal