[Linuxha-users] Caught in a loop

Michael Mansour mic at npgx.com.au
Tue Feb 1 02:18:56 GMT 2005


Hi Simon,

I made the change to the PIDFILE setting below, but I also had to make the 
modification to httpd.conf since that is the file which produces the pidfile 
that apachectl reads.

Once I did the above, I was able to stop/start httpd with both local and 
clustered setups independent of one another. I tried all combinations of stop/
start for local and clustered httpd and all worked fine.

The lems process monitor was also working as expected, and I ran this against 
the /apache/admin/scripts/shutdown script (which didn't interfere with my 
local httpd).

The only thing I could suggest as an improvement to this process would be 
maybe adding a "--restart" application option to either clstartapp or 
clhaltapp (or maybe a clrestartapp command?) ie. something that will perform 
a:

# clhaltapp -A apache -V
# clstartapp -A apache -V

in one go.

The other thing which concerned me was when I had to make a modification to a 
file within the /apache clustered filesystem, but I couldn't start the apache 
app so couldn't get access to the filesystem. What I had to do was use 
drbd_tool to mount the drbd0 device on both nodes, mount the /apache 
filesystem, make the change on the primary node, unmount the /apache 
filesystem, shutdown the drbd0 device and then test the clstartapp again to 
see if the app would start. Is there a better/easier way to perform this 
trouble-shooting process?

Thanks.

Michael.

> Hi Michael,
> 	On the lems front simply change the
> "<process_string>httpd</process_string>" line in the "httpd.xml" 
> file to tie it down to matching the correct processes, for example:
> 
> <process_string>httpd -f /apache/admin/conf/httpd.conf</process_string>
> 
> That should take care of that one.
> 
> As for the /apache/admin/scripts/apachectl script possibly worth
> changing the PIDFILE setting rather than setting RUNNING to 0, ie;
> 
> PIDFILE=/var/run/httpd-cluster.pid
> 
> Hope this works for you! Let me know if not.
> 
> Regards,
> Simon.
> 
> On Mon, 2005-01-31 at 14:50 +1000, Michael Mansour wrote:
> > Hi Simon,
> > 
> > After thinking about this alot and running through some of my own setups 
for 
> > it, I couldn't get it to work. What I tried was to have a separate /etc/
httpd/
> > conf.d.cluster directory together with the normal /etc/httpd/conf 
directory.
> > 
> > So I decided to test your suggestion below, but found this didn't work 
either, 
> > with the problem being that the report shown in /var/log/cluster/apache.
start.
> > log is:
> > 
> > /apache/admin/scripts/apachectl start: httpd (pid 9605) already running
> > 
> > so it seemed no attempt at running another port 80 listener was made.
> > 
> > I'll explain this test setup for you. I set this up as:
> > 
> > 1. /etc/httpd_local be the ServerRoot for the local httpd (local:80)
> > 
> > 2. /etc/httpd be the ServerRoot for the cluster (clusternode1:80)
> > 
> > For 1, it starts up fine using the normal OS start scripts "service httpd 
> > start":
> > 
> >  9605 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9608 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9609 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9610 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9611 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9612 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9613 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> >  9614 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 
> > and:
> > 
> > tcp        0      0 local:80              0.0.0.0:*               LISTEN
> > 
> > But when I then execute 2, the apache app starts up fine with 
"clstartapp":
> > 
> > [root at node1 conf]# clstartapp -A apache -V
> > INFO  31/01/2005 04:10:43 Validated checksum for cluster configuration
> > INFO  31/01/2005 04:10:43 Checked that node names resolve to IP addresses
> > INFO  31/01/2005 04:10:43 Validated Build run has completed against this 
> > configuration.
> > INFO  31/01/2005 04:10:43 drbd kernel module loaded already on xxx.xxx.
xxx.xxx
> > INFO  31/01/2005 04:10:43 Checking heartbeats for any sign of life...
> > WARN  31/01/2005 04:10:43 Attemping ICMP ping of xxx.xxx.xxx.xxx...
> > INFO  31/01/2005 04:10:44 drbd kernel module loaded already on xxxxxxxxx
> > INFO  31/01/2005 04:10:44 Local DRBD devices started successfully.
> > INFO  31/01/2005 04:10:44 DRBD: Skipping ENBD decisioning and relying on 
meta 
> > data...
> > INFO  31/01/2005 04:10:44 Attempting to start DRBD services on xxxxxxxxx
> > INFO  31/01/2005 04:10:45 DRBD devices started successfully on xxxxxxxxx
> > INFO  31/01/2005 04:10:45 Validated consistency of available data for 
DRBD.
> > INFO  31/01/2005 04:10:45 Both data copies believed good.
> > WARN  31/01/2005 04:10:45 Locking services not available.
> > INFO  31/01/2005 04:10:45 Attempting to register application apache as 
> > starting...
> > INFO  31/01/2005 04:10:46 Application registered successfully as starting.
> > INFO  31/01/2005 04:10:46 Checking IP address for application is not in 
use...
> > INFO  31/01/2005 04:10:48 Required application IP address is not pingable 
- 
> > continuing.
> > INFO  31/01/2005 04:10:48 Attempting to make local DRBD devices primary...
> > INFO  31/01/2005 04:10:48 All local DRBD now primary.
> > INFO  31/01/2005 04:10:48 Running "/sbin/fsck -t ext3 -a /dev/drbd0"...
> > INFO  31/01/2005 04:10:48 Running "PATH=$PATH:/sbin:/bin:/usr/sbin; mount 
-t 
> > ext3 -o rw /dev/drbd0 /apache"...
> > INFO  31/01/2005 04:10:48 File systems mounted on DRBD devices.
> > INFO  31/01/2005 04:10:48 choose_interface: Link beat ok on interface eth0
> > INFO  31/01/2005 04:10:48 choose_interface: Assigning IP address xxx.xxx.
xxx.
> > xxx to interface eth0...
> > INFO  31/01/2005 04:10:48 cmd=/sbin/ifconfig eth0:1 inet xxx.xxx.xxx.xxx 
> > netmask 255.255.255.0 2>&1
> > INFO  31/01/2005 04:10:48 Running /sbin/cluster/tools/send_arp xxx.xxx.
xxx.xxx 
> > xx:xx:xx:xx:xx:xx xxx.xxx.xxx.xxx FF:FF:FF:FF:FF:FF eth0 to send 
gratuitous 
> > arp request
> > INFO  31/01/2005 04:10:48 choose_interface: Successfully assigned IP 
address 
> > xxx.xxx.xxx.xxx to eth0:1
> > INFO  31/01/2005 04:10:48 choose_interface: Running IP level testing for 
> > interface eth0:1
> > INFO  31/01/2005 04:10:48 choose_interface: Test for IP xxx.xxx.xxx.
xxx(tcp) 
> > was OK.
> > INFO  31/01/2005 04:10:48 choose_interface: IP level testing for interface 
> > eth0:1 succeeded
> > INFO  31/01/2005 04:10:48 Applications start completed successfully
> > 
> > but the output of the apache.start.log being:
> > 
> > /apache/admin/scripts/apachectl start: httpd (pid 12088) already running
> > 
> > so there's no listen on port 80 of the cluster IP (clusternode1:80). 
> > 
> > So you know also, if I start 2 by itself (without first starting 1), the 
> > listen on clusternode1:80 works.
> > 
> > So what I decided to do was a test, for the following bit:
> > 
> >     # check for pidfile
> >     if [ -f $PIDFILE ] ; then
> >         PID=`cat $PIDFILE`
> >         if [ "x$PID" != "x" ] && kill -0 $PID 2>/dev/null ; then
> >             STATUS="httpd (pid $PID) running"
> >             RUNNING=1
> > 
> > in the /apache/admin/scripts/apachectl file, I modified it to RUNNING=0 to 
see 
> > what would happen, basically forcing the startup of apache on 
clusternode1. I 
> > then halted the app and stopped the local httpd.
> > 
> > I then restarted the local httpd (which establishes a local:80) and then 
> > clstartapp of the apache app, which successfully loaded the clusternode1:
80 as 
> > we can see:
> > 
> > tcp        0      0 clusternode1:80        0.0.0.0:*               LISTEN
> > tcp        0      0 local:80        0.0.0.0:*               LISTEN
> > 
> > and:
> > 
> > 16683 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16686 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16687 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16688 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16689 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16690 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16691 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16692 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16693 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16839 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16840 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16841 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16842 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16843 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16844 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16845 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16846 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 16848 ?        S      0:00 /usr/sbin/httpd -f /apache/admin/conf/httpd.
conf
> > 
> > as we can see modification produced the result I wanted.
> > 
> > I tested the websites and they all worked fine.
> > 
> > Halting the apache app also stopped just clusternode1:80 listen, again 
what I 
> > wanted.
> > 
> > I'm not sure what impact the modification above will make to the overall 
usage 
> > of linuxha.net, but it's the only way I was able to get this to work.
> > 
> > One other thing of note, when I ran:
> > 
> > # service httpd stop
> > 
> > while both local:80 and clusternode1:80 were up, the clusternode1:80 httpd 
> > processes were killed instead of local:80, not something I wanted. Also, 
while 
> > this was the case clstat showed the apache application was still running 
on 
> > clusternode1:
> > 
> >  Application       Node      State  Started  Monitor  Stale  Fail-over?
> >       apache      local    STARTED  0:00:00  Running      0         Yes
> > 
> > even though this isn't the case since we see:
> > 
> > 16683 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16686 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16687 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16688 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16689 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16690 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16691 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16692 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 16693 ?        S      0:00 /usr/sbin/httpd -d /etc/httpd_local
> > 
> > which is the local:80 instance of httpd and not clusternode1:80 - I 
suspect 
> > the lems monitor is only assessing whether httpd daemons are running, and 
not 
> > specifically whether the httpd running is local:80 or clusternode1:80.
> > 
> > I'd really like to get this working properly, so if you have any 
suggestions 
> > as to how I can achieve this I'd really appreciate it.
> > 
> > Michael.
> > 
> > > Hello Michael,
> > > 	Having two Server roots (i.e. /etc/httpd and /etc/httpd_local) is 
> > > the way to go I would guess. Initially copy all the files form one 
> > > to the other and then customise each by changing the Listen entry in 
> > > each to the local and clustered IP addresses as appropriate.
> > > 
> > > 	The -d option on httpd startup can then be using to specify the
> > > required server root (/etc/httpd or /etc/httpd_local) as required.
> > > 
> > > Regards,
> > > Simon.
> > > 
> > > On Fri, 2005-01-28 at 14:10 +1000, Michael Mansour wrote:
> > > > Hi Simon,
> > > > 
> > > > Just so you know, I have only one apache instance which runs as the 
> > "apache" 
> > > > application starts up. I didn't realise I could have had two.
> > > > 
> > > > The way I have it setup is to not have any apache server started on 
system 
> > > > boot. I then form the cluster and start the apache app, which reads 
the 
> > other 
> > > > conf files from /etc/httpd/conf.d/*conf and starts up the virtual 
servers.
> > > > 
> > > > I'm going to think about how I can now have two apache instances 
running, 
> > the 
> > > > local one and the clustered one. Do I need to have two sets of /etc/
httpd/
> > > > conf.d directories in this case? one for the local and one for the 
> > clustered?
> > > > 
> > > > Thanks.
> > > > 
> > > > Michael.
> > > > 
> > > > > Hello Michael,
> > > > > 	From what you're saying when the use clhaltapp on node1 both Apache
> > > > > daemons stop? If that is the case you will need to modify the script
> > > > > that is in place for the" stopscript" command for the application in
> > > > > the /etc/cluster/apache/appconf.xml. First thing to do is to test 
it-
> > > > > with the apache application running on node1 as well as your local
> > > > > Apache, run whatever the "stopscript" is. It should only stop the
> > > > > clustered Apache instance. If no httpd processes are running you now
> > > > > know that this script needs to be modified in some way.
> > > > > 
> > > > > 	Possibly both use the same pid file (/var/run/httpd.pid)?
> > > > > 
> > > > > Regards,
> > > > > Simon




More information about the Linuxha-users mailing list