Topic
2 replies Latest Post - ‏2010-10-22T06:27:05Z by Chuck.Spacey
ddn_mukesh
ddn_mukesh
2 Posts
ACCEPTED ANSWER

Pinned topic Mounting cnfs exported GPFS files

‏2010-07-30T13:35:48Z |
We have created cnfs cluster as follows:

GPFS cluster information
========================
GPFS cluster name: abcd.efg.com
GPFS cluster id: xxxxxxxx

Cluster NFS global parameters

Shared root directory: /stage2
Virtual IP address: cggnfs
rpc.mountd port number: 9999
nfsd threads: 256
Reboot on failure enabled: yes
CNFS monitor enabled: yes

Node Daemon node name IP address CNFS state group CNFS IP address list

6 frttap2.xxxxx.com 192.168.4.106 enabled 6 192.168.6.106
7 frttap3.xxxxx.com 192.168.4.107 enabled 7 192.168.6.107

  1. cat /etc/exports

/stage2 *(rw,async,no_root_squash,fsid=777)


Now i have some questions?
1. We have created virtual IP address (using system-config-network) before creating --cnfs cluster and mentioned those IP addresses in the /etc/hosts file also. While creating --cnfs cluster, the system generated error message. Then we deleted the virtual IP addresses using system-config-network but mistakenly left the virtual IP address entries in /etc/hosts file. When we created --cnfs then the cluster has taken the IP addresses as mentioned in the /etc/hosts directory. IS it a requirement to make entries in the /etc/hosts table before adding the Virtual IPs for virtual nfs hosts in the --cnfs cluster.

2. There is a mention in the manual that rpc.mountd shall be using same port in the cnfs participating nodes. We have checked that rpc.mountd is running on different ports in the participating nodes. We have defined the --cnfs rpc.mountd port as 9999. But how to mount these exported file system on client using cnfs port number. What is the significance of rpc.mountd port number defined in the --cnfs cluster.

3. We are exporting --cnfs from network using LDAP authorization and trying to mount it on other connected network with NIS authorisation. Then how DNS will come into picture. I have seen in another question to mount the --cnfs exported filsystem as node:/directory. But in this case there are 2 nodes whose host names are not resolvable in the client network where we want to mount the exported --cnfs. Kindly mention the complete syntax for mounting -cnfs on client.

4. Earlier with simple nfs export, we faced problem with rpc.lockd. Out application has worked correctly when we mounted the nfs exported file system with -nolock option. Do we require to mount the --cnfs exported file also as --nolock.

5. What is the significance of Virtual Ip Address of the cluster: cggnfs in our case. Is it resolvable in the connected network.
Thanks in anticipation
Updated on 2010-10-22T06:27:05Z at 2010-10-22T06:27:05Z by Chuck.Spacey
  • LuoChen
    LuoChen
    14 Posts
    ACCEPTED ANSWER

    Re: Mounting cnfs exported GPFS files

    ‏2010-08-02T13:44:36Z  in response to ddn_mukesh
    1. It is not required to add the IP addresses to the /etc/hosts file, BUT reverse DNS lookup must resolve the IP addresses to names. CNFS enforces this - if these IPs are not in DNS, then they MUST be in /etc/hosts (or somehow resolvable with gethostbyname). This is because some legacy client/server daemons give errors during reverse lookup.

    2. The client does not need to specify server mountd port (it internally uses portmap to figure it out), but there are know failover issues if the mountd binds to different ports on the participating CNFS nodes.

    3. You can set up LDAP authorization with NIS as you wish, but ultimately the client needs to mount the server using the server:path notation. If the server host names are not resolvable in the client, you can use IP addresses. If this doesn't resolve your issue, could you give an illustration of what you would use in this environment.

    4. CNFS supports lockd. Please indicate what problems you faced. --nolock should not be required.

    5. The virtual name (not IP) is required for some legacy (SLES9) clients only. There are some known client issues when NLM (locking) failover will not work correctly if the client uses a virtual name (instead of node name or IP) to mount. This should not be required for most environments.
  • Chuck.Spacey
    Chuck.Spacey
    1 Post
    ACCEPTED ANSWER

    Re: Mounting cnfs exported GPFS files

    ‏2010-10-22T06:27:05Z  in response to ddn_mukesh
    LuoChen wrote:
    1. It is not required to add the IP addresses to the /etc/hosts file, BUT reverse DNS lookup must resolve the IP addresses to names. CNFS enforces this - if these IPs are not in DNS, then they MUST be in /etc/hosts (or somehow resolvable with gethostbyname). This is because some legacy client/server daemons give errors during reverse lookup.

    2. The client does not need to specify server mountd port (it internally uses portmap to figure it out), but there are know failover issues if the mountd binds to different ports on the participating CNFS nodes.

    3. You can set up LDAP authorization with NIS as you wish, but ultimately the client needs to mount the server using the server:path notation. If the server host names are not resolvable in the client, you can use IP addresses. If this doesn't resolve your issue, could you give an illustration of what you would use in this environment.

    4. CNFS supports lockd. Please indicate what problems you faced. --nolock should not be required.

    5. The virtual name (not IP) is required for some legacy (SLES9) clients only. There are some known client issues when NLM (locking) failover will not work correctly if the client uses a virtual name (instead of node name or IP) to mount. This should not be required for most environments.

    It's very detailed, It is just the solution for my problem, Many thanks to your description!