Bug #353
mm_sched crashes with segmentation fault
| Status: | Closed | Start date: | 09/22/2010 | |
|---|---|---|---|---|
| Priority: | Normal | Due date: | ||
| Assignee: |  Ruben S. Montero | % Done: | 0% | |
| Category: | Scheduler | |||
| Target version: | - | |||
| Resolution: | fixed | Pull request: | ||
| Affected Versions: |
Description
Dear all,
I had to re-install a lot of VM im my system and noticed that the mm_sched stop assigning the VMs to physical nodes after some time. Doing a "ps aux" revealed that the mm_sched process was not running.
To handle with this issue I decided to run the mm_sched manually and supplied the parameters that usually would have been set by the "one" script (that starts the ONE daemon and the scheduler). I also tried to tweak some values and ran:
oneadmin@one:/opt/one/bin> mm_sched -p 2633 -t 30 -m 600 -d 3 -h 1
after a few seconds/ minutes I got this output:
Segmentation fault
I attached also a strace run for the mm_sched showing the same result.
Could you please have a look at this?
(I am running a ONE 2.0beta git snapshot from 29.07.2010)
Cheers,
Stefan
debug.txt 
- strace for mm_sched
(76.1 KB)
Associated revisions
Merge pull request #353 from gladhorn/master
Fix authentication for users with non-ascii chars in their names
History
#1
Updated by Stefan Freitag almost 11 years ago
Hello Ruben,
just an update to the situation. After some investigation the problem was imo caused by a "bad" template file that contained the line
REQUIREMENTS=HOSTNAME = ""
Do you know if the could lead to the seg fault in the scheduler?
Cheers
Stefan
#2
Updated by Carlos Martín almost 11 years ago
Hi Stefan,
I've tried to reproduce the error submitting this template:
NAME = test REQUIREMENTS = HOSTNAME = ""
But all I get is this error from the Request Manager:
$ onevm create template.one Error: [VirtualMachineAllocate] Error trying to PARSE VM TEMPLATE Returned error code [1].. Reason: syntax error, unexpected EQUAL, expecting $end or VARIABLE at line 3, columns 37:40
What means that the malformed template is never inserted in the DB, and the scheduler doesn't have a chance to read it.
Could you send us the complete template you used?
Regards,
Carlos.
#3
Updated by Stefan Freitag almost 11 years ago
Hello Carlos,
I think you missed some quotes in the requirement section.
Here is the template I used to create the VM
=== snipp ===
NAME = udo-wn117
VCPU = 8
MEMORY = 13312
OS = [
bootloader = "/root/bin/domUloader.py"
]
RAW = [ type="xen", data="bootargs=\"--verbose --entry=xvda1\"" ]
DISK = [
source = "/mnt/gridconfig/images/workernode/wn_sl54_x86_64_2.img",
target = "xvda",
readonly = "no" ]
DISK = [
type = swap,
size = 1024,
target = "xvdb",
readonly = "no" ]
DISK = [
type="block",
clone="yes",
source="/dev/cciss/c0d0p4",
target = "xvdc",
readonly = "no" ]
NIC = [NETWORK="DGRZRWorkernodes", IP = X.XXX.XXX.XXX ]
REQUIREMENTS = "HOSTNAME = \"\""
RANK = FREECPU
=== ===
Doing- onevm create udo-wn117.template
results in
- onevm show 346
VIRTUAL MACHINE 346 INFORMATION
ID : 346
NAME : udo-wn117
STATE : PENDING
LCM_STATE : LCM_INIT
START TIME : 09/24 15:25:58
END TIME : -
DEPLOY ID: :
VIRTUAL MACHINE TEMPLATE
DISK=[
DISK_ID=0,
READONLY=no,
SOURCE=/mnt/gridconfig/images/workernode/wn_sl54_x86_64_2.img,
TARGET=xvda ]
DISK=[
DISK_ID=1,
READONLY=no,
SIZE=1024,
TARGET=xvdb,
TYPE=swap ]
DISK=[
CLONE=yes,
DISK_ID=2,
READONLY=no,
SOURCE=/dev/cciss/c0d0p4,
TARGET=xvdc,
TYPE=block ]
MEMORY=13312
NAME=udo-wn117
NIC=[
BRIDGE=eth0,
IP=XXX.XXX.XXX.XXX,
MAC=YY:YY:YY:YY:YY:YY,
NETWORK=DGRZRWorkernodes,
NETWORK_ID=0 ]
OS=[
BOOTLOADER=/root/bin/domUloader.py ]
RANK=FREECPU
RAW=[
DATA=bootargs="--verbose --entry=xvda1",
TYPE=xen ]
REQUIREMENTS=HOSTNAME = ""
VCPU=8
VMID=346
Cheers
Stefan
#4
Updated by Carlos Martín almost 11 years ago
- Status changed from New to Closed
- Resolution set to fixed
Hi Stefan,
You were right, there was a bug in the scheduler parser that led to a segmentation fault when it tried to compare an empty string. It should be solved by this commit:
Thank you for your feedback!!