Mailing List Archive: Crazy Values in backhand-status and question on redirecting

Crazy Values in backhand-status and question on redirecting

Jul 24, 2003, 10:16 AM

Post #1 of 4 (2228 views)

Hi.

1) Are there any known problems with multicasted stats?

Server with Entry 1 below (w/ mod_ackhand 1.2.1) has 4 GB RAM but
the values for memory and req (seen on server 0 w/ mod_backhand
from current CVS) below are rather unbelievable.

Entry | Hostname | Age | Address | Total Mem | Avail Mem |
# ready servers/# total servers | ~ms/req [#req] | Arriba |
# CPUs | Load/HWM | CPU Idle

0 | rhea.hbz-nrw.de | 0 | 193.30.112.104 | 1976 MB | 1704 MB |
10/10 | 0 [0] | 535356 | 2 | 0.003/7 | 0.000

1 | kronos.hbz-nrw.de | 0 | 193.30.112.118 | 4 MB | 4174864384 MB |
263714/17 | 0 [-1282416640] | 0 | 0 | 0.017/3 | 8.000

Stats from server 0 are not sent to server 1 so status there
consists only of its own status (here he gets #CPUs and memory
right but Age, Load and CPU Idle are curious):

0 | kronos.hbz-nrw.de | 1059066619 | 193.30.112.118 | 3981 MB |
2966 MB | 0/20 | 0 [0] | 263714 | 4 | /.////1 | /.///

2) We are thinking of using mod_backhand in an rather unusal way:
in our scenario there will one "frontend" server with
apache/backhand on port 80.
This server should redirect requests to port 80 on other servers;
the apache/backhand on those servers however won=B4t be on port 80
but somewhere else (port 80 will be used by a different proprietary
server) and balancing should take place primarily based on CPU load
of the "backend" servers.
Is this possible?
Or will the redirector in any case try to redirect to the port of
the apache from which the stats originate?
Or can I influence this behaviour by adding a port to the <ip addr>
part of the MulticastStats config of the backend servers like this:
MulticastStats 193.30.112.456:80 193.30.112.104:4445,1

Any hints are welcome.
Thanx in advance

heiko

Crazy Values in backhand-status and question on redirecting [ In reply to ]

jesus at omniti

Jul 24, 2003, 11:18 AM

Post #2 of 4 (2146 views)

Permalink

Heiko Jansen wrote:
> Hi.
>
> 1) Are there any known problems with multicasted stats?
>
> Server with Entry 1 below (w/ mod_ackhand 1.2.1) has 4 GB RAM but
> the values for memory and req (seen on server 0 w/ mod_backhand
> from current CVS) below are rather unbelievable.

(1) Make sure all servers are running the exact same version of mod_backhand.
(2) When submitting problems like this it is essential to provide the
operating system name and version as well as the hardware architecture that is
hosting each node.

> Entry | Hostname | Age | Address | Total Mem | Avail Mem |
> # ready servers/# total servers | ~ms/req [#req] | Arriba |
> # CPUs | Load/HWM | CPU Idle
>
> 0 | rhea.hbz-nrw.de | 0 | 193.30.112.104 | 1976 MB | 1704 MB |
> 10/10 | 0 [0] | 535356 | 2 | 0.003/7 | 0.000
>
> 1 | kronos.hbz-nrw.de | 0 | 193.30.112.118 | 4 MB | 4174864384 MB |
> 263714/17 | 0 [-1282416640] | 0 | 0 | 0.017/3 | 8.000
>
> Stats from server 0 are not sent to server 1 so status there
> consists only of its own status (here he gets #CPUs and memory
> right but Age, Load and CPU Idle are curious):
>
> 0 | kronos.hbz-nrw.de | 1059066619 | 193.30.112.118 | 3981 MB |
> 2966 MB | 0/20 | 0 [0] | 263714 | 4 | /.////1 | /.///
>
> 2) We are thinking of using mod_backhand in an rather unusal way:
> in our scenario there will one "frontend" server with
> apache/backhand on port 80.
> This server should redirect requests to port 80 on other servers;
> the apache/backhand on those servers however won´t be on port 80
> but somewhere else (port 80 will be used by a different proprietary
> server) and balancing should take place primarily based on CPU load
> of the "backend" servers.
> Is this possible?
> Or will the redirector in any case try to redirect to the port of
> the apache from which the stats originate?
> Or can I influence this behaviour by adding a port to the <ip addr>
> part of the MulticastStats config of the backend servers like this:
> MulticastStats 193.30.112.456:80 193.30.112.104:4445,1

Yes. you can influence it like that.

But you are talling it to MulticastStats to 193.30.112.104. I am not sure
what you are trying to accomplish with that. It is neither a multicast
address nor a broadcast address -- you'll need to fix that first.

--
Theo Schlossnagle
Principal Consultant
OmniTI Computer Consulting, Inc. -- http://www.omniti.com/
Phone: +1 410 872 4910 x201 Fax: +1 410 872 4911
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7

Crazy Values in backhand-status and question on redirecting [ In reply to ]

JANSEN at hbz-nrw

Jul 29, 2003, 7:29 AM

Post #3 of 4 (2124 views)

Permalink

Am 24 Jul 2003 um 14:18 schrieb Theo E. Schlossnagle:
[Re: [m_b_users] Crazy Values in bac]

> (1) Make sure all servers are running the exact same version of mod_back=
hand.

I changed the other server to use the mod_backhand from CVS and now
everything works and provides sensible values - except for the CPU
idle value: it is always 0.000 on all servers (while the load value
changes).

> (2) When submitting problems like this it is essential to provide the
> operating system name and version as well as the hardware architecture t=
hat is
> hosting each node.

You=B4re right. Should have done this. E420R and Fire 280R; both with
Solaris 2.9.

> But you are talling it to MulticastStats to 193.30.112.104. I am not su=
re
> what you are trying to accomplish with that. It is neither a multicast
> address nor a broadcast address -- you'll need to fix that first.

I was trying to see if it would send stats only to the one denoted
server. Indeed it worked, but it=B4s not intended to run this way.

Another question: how is the load of the servers calculated? Is it
just the load in the moment when the stats are send or is there a
way to get the average load of the last x minutes?

thanks in advance
heiko

Crazy Values in backhand-status and question on redirecting [ In reply to ]

jesus at omniti

Jul 29, 2003, 8:33 AM

Post #4 of 4 (2138 views)

Permalink

Heiko Jansen wrote:
> I was trying to see if it would send stats only to the one denoted
> server. Indeed it worked, but it´s not intended to run this way.
>
> Another question: how is the load of the servers calculated? Is it
> just the load in the moment when the stats are send or is there a
> way to get the average load of the last x minutes?

Last 1 minute average.

You can use byBusyChildren for a rough estimate of the current run-queue
length. If you Apache instances are in heavy use, then that will accurately
represent how busy the machine is.

Note, you will likely need to perform a graceful restart for those numbers to
be available.

--
Theo Schlossnagle
Principal Consultant
OmniTI Computer Consulting, Inc. -- http://www.omniti.com/
Phone: +1 410 872 4910 x201 Fax: +1 410 872 4911
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7