[one-users] Missing monitoring information from VMware (ESXi 4.1)

[Håkan Iaksson]

 We just upgraded from openebula-3.0.0 to 3.2.0,
 but it looks like monitoring in the VMware-driver is still incomplete,
 or is our opennebula installation not working properly?

 I'm speaking about the CPU and MEM stats when using "onevm list"

 [oneadmin ~]$ onevm list
     ID USER     GROUP    NAME         STAT CPU     MEM        HOSTNAME  
      TIME
    175 oneadmin oneadmin one-175      runn   0      0K dev-clmesx2.ini 
 32 23:33:06
    218 oneadmin oneadmin one-218      runn   5    512M   kvm02.init.se 
 11 19:47:49
    220 oneadmin oneadmin one-220      runn 100   1024M   kvm02.init.se 
 06 01:07:35
    222 oneadmin oneadmin one-222      runn   0      0K dev-clmesx2.ini 
 05 20:23:45
    227 oneadmin oneadmin one-227      runn   0      0K dev-clmesx1.ini 
 00 00:05:28

 It's always 0 CPU and MEM for vm's on ESX, the KVM hosts works fine out 
 of the box.

 The host-monitoring is working though (with the cli):

 [oneadmin ~]$ onehost list
   ID NAME               RVM   TCPU   FCPU   ACPU   TMEM   FMEM   AMEM   
 STAT
   18 dev-clmesx1.ini      1    400    400    300    32G  25.6G  31.5G   
  on
   19 dev-clmesx2.ini      2    200    200      0     4G   3.2G     3G   
  on
   23 kvm01.init.se        3    800    580    500  31.3G  25.3G  29.3G   
  on

 But the "hosts" page in the sunstone webgui always show "0% CPU Use" 
 and "20% Memory Use" for VMware hosts.

 It seems like sunstone does not get the same information as the cli, 
 looking at the
 "Host information" for the first host dev-clmesx1.init.se it says:
 Max Mem                32G
 Used Mem (real)         0K
 Used Mem (allocated)    0K
 Used CPU (real)         0
 Used CPU (allocated)    200
 Running VMs             2

 Shouldn't those values be the same as reported by the cli? Only the Max 
 Mem is correct.

 The funny thing is that the graphs under "Monitoring information" that 
 are updated by oneacctd
 does show accurate changes in CPU and MEM usage over time.

 Isn't this information vital for opennebula to work correctly?
 How can the scheduler determine if a ESX-server can handle more vm's
 when it does not know how much resources each vm consume?

 The obvious way to get this information would be with the dominfo 
 command in virsh.
 Looking at the vmware_driver.rb there is only one place where dominfo 
 is used,
 and it looks like it only checks the status of the virtual machines.
 
 I realise dominfo can not show anything but static information about 
 CPU and MEM usage,
 but it seems better to report that, than not reporting anything?

 Sorry if this has been asked before, but I've searched the mailinglist 
 for information about this,
 have not found much at all about monitoring and VMware.

 -----
 Håkan Isaksson