[Cluster-devel] Change in rgmanager - broke my config

[Cluster-devel] Change in rgmanager - broke my config

[Charles Riley]

Hi,

I am using rhcs on RHEL4.

At some point in the recent past, clurgmgrd was changed so that when it
parses cluster.conf,  it checks to make sure that filesystem mountpoints
are all unique.

For 99% of clusters out there this is probably a good thing.  However,
in this case failover domains and services are configured such that even
though multiple machines are mounting devices on the same mount point,
there is no possible way that any one server could ever attempt to mount
more than one device on the same mountpoint.  Each server runs ONE
service which gets it's own private disk space (no filesystem sharing is
going on, and "run exlclusive" is enabled for the service).  If one of
the hot standby servers in the failover domain for the service isn't
available ( extremely small but non-zero chance of that happening),  the
service simply won't (shouldn't) start.

In short, rgmanager is being overprotective and hosing me up =).

I could just use older rpms, but looking through the release notes there
are several bugs fixed, and I don't really want to lose those fixes.

What I'd like to do is modify rgmanager so it will not enforce that
filesystem mountpoints be unique in a case like mine.

I'm in a bit of a crunch (I need to get a system configured, tested, 
and shipped) so if anyone is familiar with what section of the code I
should be looking at I'd appreciate the headstart =)

Also, it appears that once rgmanager finds a collision, it pretty much
just stops parsing that section of the config.  One side effect of that
is that the system will then let you start that service and report the
service running,  even though it's really not.  Not really a bug, but
probably not how we'd like it to work.

Thanks guys,

Charles


 

[Cluster-devel] Change in rgmanager - broke my config

[Charles Riley]

Hi,

I changed /usr/share/cluster/fs.sh thusly:

--- /usr/share/cluster/fs.sh    2008-05-28 14:55:24.000000000 -0400
+++ /home/image/cluster/fs.sh   2008-09-15 13:43:48.000000000 -0400
@@ -74,7 +74,7 @@
            <content type="string"/>
         </parameter>

-        <parameter name="mountpoint" unique="1" required="1">
+        <parameter name="mountpoint" unique="0" required="1">
            <longdesc lang="en">
                Path in file system heirarchy to mount this file system.
            </longdesc>
@@ -84,7 +84,7 @@
            <content type="string"/>
         </parameter>

-        <parameter name="device" unique="1" required="1">
+        <parameter name="device" unique="0" required="1">
            <longdesc lang="en">
                Block device, file system label, or UUID of file system.
            </longdesc>

I have to waggle my finger at you; changing stuff like this can break
running clusters.. maybe it was in a release note that I didn't see, if
that's the case I retract my waggle and offer sincerest apologies =)

I had to change the device parameter as well in this case, when we're
using iscsi storage our app mounts a local disk for use as tmp space
when it starts.

Charles

Charles Riley wrote:
> Hi,
>
> I am using rhcs on RHEL4.
>
> At some point in the recent past, clurgmgrd was changed so that when it
> parses cluster.conf,  it checks to make sure that filesystem mountpoints
> are all unique.
>
> For 99% of clusters out there this is probably a good thing.  However,
> in this case failover domains and services are configured such that even
> though multiple machines are mounting devices on the same mount point,
> there is no possible way that any one server could ever attempt to mount
> more than one device on the same mountpoint.  Each server runs ONE
> service which gets it's own private disk space (no filesystem sharing is
> going on, and "run exlclusive" is enabled for the service).  If one of
> the hot standby servers in the failover domain for the service isn't
> available ( extremely small but non-zero chance of that happening),  the
> service simply won't (shouldn't) start.
>
> In short, rgmanager is being overprotective and hosing me up =).
>
> I could just use older rpms, but looking through the release notes there
> are several bugs fixed, and I don't really want to lose those fixes.
>
> What I'd like to do is modify rgmanager so it will not enforce that
> filesystem mountpoints be unique in a case like mine.
>
> I'm in a bit of a crunch (I need to get a system configured, tested, 
> and shipped) so if anyone is familiar with what section of the code I
> should be looking at I'd appreciate the headstart =)
>
> Also, it appears that once rgmanager finds a collision, it pretty much
> just stops parsing that section of the config.  One side effect of that
> is that the system will then let you start that service and report the
> service running,  even though it's really not.  Not really a bug, but
> probably not how we'd like it to work.
>
> Thanks guys,
>
> Charles
>
>
>  
>
>