[Haizea] OpenNebula / Haizea: multiple users issue

[Jonathan Rouzaud-Cornabas]

Hello,

I have modify Haizea to support OpenNebula 3 (I will submitted the 
modified version to github very soon).
I currently try to test OpenNebula+Haizea with multiple users but seems 
that there is an issue.

When I instantiate 40 VMs for one users (onetemplate instantiate 
tmplid), Haizea balances the VMs between the different hosts (4 in my test).

But when I instantiate 20 VMs by user for two users at the same time, 
Haizea balances the VMs between only two nodes and can not instantiate 
all VMs as there is not enough ressources if not all nodes are used.

I have the following error message:
[2011-09-30 14:23:04.25] LSCHED  Scheduling lease #45 (1 nodes) -- Immediate
[2011-09-30 14:23:04.25] LSCHED  Lease request #45 cannot be scheduled: 
Could not find enough resources for this request

With onevm list a:
     23 user_1   users    one-23       runn   1    512M paramount-9.ren 
00 00:12:29
     24 user_0   users    one-24       runn   1    512M paramount-33.re 
00 00:12:27
     25 user_0   users    one-25       runn   0      0K paramount-33.re 
00 00:09:38
     26 user_1   users    one-26       runn   0      0K paramount-33.re 
00 00:09:38
     27 user_0   users    one-27       runn   0      0K paramount-33.re 
00 00:09:37
     28 user_1   users    one-28       runn   0      0K paramount-5.ren 
00 00:09:37
     29 user_0   users    one-29       runn   0      0K paramount-5.ren 
00 00:09:37
     30 user_1   users    one-30       runn   0      0K paramount-5.ren 
00 00:09:37
     31 user_0   users    one-31       runn   0      0K paramount-5.ren 
00 00:09:37
     32 user_1   users    one-32       pend   0      0K 
00 00:09:37
     33 user_0   users    one-33       pend   0      0K 
00 00:09:37
     34 user_1   users    one-34       pend   0      0K 
00 00:09:37
     35 user_0   users    one-35       pend   0      0K 
00 00:09:37
     36 user_1   users    one-36       pend   0      0K 
00 00:09:37
     37 user_0   users    one-37       pend   0      0K 
00 00:09:37
     38 user_1   users    one-38       pend   0      0K 
00 00:09:37
     39 user_0   users    one-39       pend   0      0K 
00 00:09:36
     40 user_1   users    one-40       pend   0      0K 
00 00:09:36
     41 user_0   users    one-41       pend   0      0K 
00 00:09:36
     42 user_1   users    one-42       pend   0      0K 
00 00:09:36
     43 user_0   users    one-43       pend   0      0K 
00 00:09:36
     44 user_1   users    one-44       pend   0      0K 
00 00:09:36

And onehost list:
   ID NAME               RVM   TCPU   FCPU   ACPU   TMEM   FMEM   AMEM 
  STAT
    1 paramount-5.ren      5    400    400    230   7.8G   7.6G   3.3G 
    on
    2 paramount-8.ren      0    400    400    400   7.8G   7.7G   7.8G 
    on
    0 paramount-33.re      5    400    400    260   7.8G   7.5G   3.8G 
    on
    3 paramount-9.ren      1    400    400    390   7.8G   7.6G   7.3G 
    on

The VM Template is:
TEMPLATE CONTENTS
CPU=0.4
DISK=[
   DRIVER=qcow2,
   IMAGE_ID=7,
   TARGET=sda ]
FEATURES=[
   ACPI=yes ]
HAIZEA=[
   DURATION=unlimited,
   PREEMPTIBLE=no,
   START=now ]
MEMORY=1024
NAME=WorkingNODE
NIC=[
   NETWORK_ID=3 ]
OS=[
   ARCH=x86_64,
   BOOT=hd,
   ROOT=sda ]
RAW=[
   DATA="<devices><serial type=\"pty\"><source 
path=\"/dev/pts/5\"/><target port=\"0\"/></serial><console type=\"pty\" 
tty=\"/dev/pts/5\"><source path=\"/dev/pts/5\"/><target 
port=\"0\"/></console><graphics type=\"vnc\" port=\"5901\" 
autoport=\"yes\"/></devices>",
   TYPE=kvm ]
TEMPLATE_ID=6


Anyone have already seen this behavior ? I have tried with another set 
of nodes with different ressources and have the same issue.

With the default OpenNebula scheduler, it is working.

Thanks,

Regards,

Jonathan Rouzaud-Cornabas