Difference between revisions of "Testbed Update Plan"

From PDP/Grid Wiki
Jump to navigationJump to search
 
(18 intermediate revisions by the same user not shown)
Line 1: Line 1:
== Planning the update of the middleware/development test bed ==
+
The upgrade of the P4CTB has taken place; all information of this page is merged into [[Agile testbed]].
 
 
 
 
There is a number of tasks involved in bringing the testbed to where we like to be. We also need to agree on a timeframe in which we like to see these things accomplished.
 
 
 
=== inventory of current services ===
 
 
 
This section should list the '''current''' services we run and use on the testbeds. For each service, we should explain what we should like to do with it (keep, drop, change?).
 
 
 
{| class="wikitable" border="0" cellpadding="8"
 
|-style="background-color: #f81;"
 
! Service
 
! System
 
! keep/move/lose
 
! Comments
 
|-
 
| LDAP
 
| span
 
| lose
 
| to be discontinued after migration to central LDAP
 
|-
 
| DHCP
 
| span
 
| move
 
| by dnsmasq, /etc/hosts and /etc/ethers. Should migrate elsewhere
 
|-
 
| DHCP
 
| bleek
 
| keep
 
| bleek is going to be the main home directory and DHCP server, after it's been upgraded.
 
|-
 
| Cruisecontrol
 
| bleek
 
| move
 
| build system for VL-e and BiG Grid; may move to Hudson. Currently transferred to cruisecontrol.testbed
 
|-
 
| Hudson
 
| kudde
 
| move
 
| continuous integration, currently for jGridstart but could serve others
 
|-
 
| Home directories
 
| everywhere
 
| move
 
| should be merged onto single NFS server
 
|-
 
| X509 host keys and pre-generated SSH keys
 
| span
 
| move
 
| all in /var/local/hostkeys
 
|-
 
| Robot certificate (etoken)
 
| kudde
 
| keep/move
 
| hardware token is plugged into kudde to generate vlemed robot certificates using cronjob + software in <tt>/root/etoken</tt>. Token can be moved to another machine but it should remain within the P4CTB
 
|-
 
| VMWare Server 1.0
 
| bleek
 
| move
 
| If bleek is destined to be home directory server, VMWare should be put on other hardware.
 
|}
 
 
 
==== VMWare on bleek.nikhef.nl ====
 
 
 
The following machines are all candidate for being scrapped, unless serious objections arise.
 
 
 
drwxr-xr-x  2 dennisvd users 4096 Feb  5  2007 ./poc_test_centos3
 
drwxr-xr-x  2 dennisvd users 4096 Mar 30  2007 ./CentOS 4
 
drwxr-xr-x  2 dennisvd users 4096 Apr 10  2007 ./centos3-server
 
drwxr-xr-x  2 dennisvd users 4096 Jun  5  2008 ./Debian Sid
 
drwxrwxr-x  2 dennisvd users 4096 Oct 21  2008 ./poc1
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./Debian Etch 4.0r0
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./debian 4 minimal
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./scientific linux 4
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./CentOS4-Server-i386
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./CentOS3-i386
 
drwxr-xr-x  2 dennisvd users 4096 Oct 21  2008 ./tinyCentOS3
 
drwxr-xr-x  2 msalle  users 4096 Jun 10 10:33 ./debian4-64
 
drwxrwxr-x  2 dennisvd users 4096 Jun 10 10:33 ./vle-poc-r3
 
drwxr-xr-x  2 janjust  users 4096 Sep 28 16:20 ./debian5-64                      (saved for Mischa)
 
drwxr-xr-x  2 dennisvd users 4096 Oct 27 18:20 ./cruisecontrol_centos4_i386      (CC no longer uses this)
 
drwxr-xr-x  2 dennisvd users 8192 Oct 28 15:07 ./cruisecontrol_centos4_x86_64    (CC no longer uses this)
 
drwxrwxr-x  2 dennisvd users 4096 Nov 11 11:49 ./ren.nikhef.nl                  (used by WicoM, can be scrapped)
 
 
 
Besides these, a couple of zip files exists which are packed versions of virtual machine images. These will also disappear unless someone wants to take them home for nostalgic reasons.
 
 
 
-rw-r--r--  1 dennisvd users  232361772 Aug  1  2007 debian-4-minimal.zip              (scrap)
 
-rw-r--r--  1 janjust  users  430972968 Aug  1  2007 gforge-4.5.15-fc3.zip            (scrap)
 
-rw-r--r--  1 janjust  users  654212858 Aug  1  2007 lcg27ui.zip                      (scrap)
 
-rw-r--r--  1 janjust  users  203298842 Aug  1  2007 tinyCentOS-3.8.zip                (scrap)
 
-rw-r--r--  1 janjust  users  197921825 Aug  1  2007 tinyCentOS.zip                    (scrap)
 
-rw-r--r--  1 janjust  users 1249781095 Aug  1  2007 vle-poc-r1-build003.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1259094081 Aug  1  2007 vle-poc-r1-build004.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1310361793 Aug  1  2007 vle-poc-r1-build005.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1341440768 Aug  1  2007 vle-poc-r2-build001.zip          (scrap)
 
-rw-r--r--  1 dennisvd users  647782099 Aug 30  2007 minimal-sl4-apt.zip              (scrap)
 
-rw-r--r--  1 janjust  users  803171882 Sep 18  2007 CentOS-3.9-glite-3.0.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1386416706 Sep 18  2007 vle-poc-r2-build002.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1394177060 Sep 21  2007 tutorial07.zip                    (scrap)
 
-rw-r--r--  1 janjust  users  659597940 Oct 21  2008 centos4.7-i386.zip                (scrap)
 
-rw-r--r--  1 janjust  users 1536737886 Oct 22  2008 vle-poc-r3-build001.zip          (scrap)
 
-rw-r--r--  1 janjust  users 1541972232 Oct 22  2008 tutorial08.zip                    (save, but I already have backup)
 
-rw-r--r--  1 janjust  users  941597293 Nov 11  2008 centos4.7-glite-3.1-wn-i386.zip  (scrap)
 
 
 
=== Data plan for precious data ===
 
 
 
Precious means anything that took effort to put together, but nothing that lives in version control elsewhere. Think home directories, system configurations, pre-generated ssh host keys, X509 host certs, etc.
 
 
 
One idea is to put all of this on a box that is not involved in regular experimentation and messing about, and have backup arranged from this box to Sara with ADSM (which is the current service running on bleek). ''After'' this is arranged we begin to migrate precious data from all the other machines here, leaving the boxen in a state that we don't get sweaty palms over scratching and reinstalling them.
 
 
 
=== Hardware inventory ===
 
 
 
Perhaps this should be done first. Knowing what hardware we have is prerequisite to making sensible choices about what we try to run where.
 
 
 
Changes here should probably also go to [[NDPF System Functions]].
 
 
 
{| class="wikitable" border="0" cellpadding="8"
 
|-style="background-color: #ccc;"
 
! name
 
! ipmi name*
 
! type
 
! chipset
 
! #cores
 
! mem
 
! OS
 
! disk
 
! service tag
 
! remarks
 
|-
 
| bleek
 
| bleek
 
| PE1950
 
| Intel 5150 @2.66GHz
 
| 4
 
| 8GB
 
| CentOS4-64
 
| software raid1 2&times;500GB disks
 
| CQ9NK2J
 
| High Availability, dual power supply; precious data; backed up
 
|-
 
| toom.testbed
 
| toom
 
| PE1950
 
| Intel E5440 @2.83GHz
 
| 8
 
| 16GB
 
| CentOS5-64
 
| Hardware raid1 2&times;715GB disks
 
| DC8QG3J
 
| current Xen 3 hypervisor with mktestbed scripts
 
|-
 
| kudde
 
| kudde
 
| PE1950
 
| Intel E5440 @2.83GHz
 
| 8
 
| 16GB
 
| CentOS5-64
 
| Hardware raid1 2&times;715GB disks
 
| CC8QG3J
 
| Contains hardware token/robot proxy for vlemed; current Xen 3 hypervisor with mktestbed scripts
 
|-
 
| span
 
| span
 
| PE2950
 
| Intel E5440 @2.83GHz
 
| 8
 
| 24GB
 
| CentOS5-64
 
| Hardware raid10 on 4&times;470GB disks (950GB net)
 
| FP1BL3J
 
| DHCP,DNS,NFS,LDAP; home dirs must be moved to bleek; current Xen 3 hypervisor with mktestbed scripts
 
|-
 
| melkbus
 
| hals
 
| PEM600
 
| Intel E5450 @3.00GHz
 
| 8
 
| 32GB
 
| CentOS5-64
 
| 2x 320GB SAS disks
 
| 76T974J
 
| to be renamed to hals; Oscar's private machine?
 
|-
 
| put.testbed
 
| put
 
| PE2950
 
|
 
|
 
|
 
|
 
|
 
| HMXP93J
 
| former garitxako;FreeNAS
 
|-
 
| blade14.testbed
 
| bl0-14
 
| PEM610
 
| Intel E5504 @2.00GHz
 
| 2x4
 
| 16GB
 
| OVMS 3.0.3
 
|
 
| 4NZWF4J
 
| former autana
 
|-
 
| blade13.testbed
 
| bl0-13
 
| PEM610
 
|
 
|
 
|
 
|
 
|
 
| 5NZWF4J
 
| former arauca; OVM 3.0.3
 
|-
 
| voor
 
| arrone
 
| PE1950
 
|
 
|
 
|
 
|
 
|
 
| 982MY2J
 
| former arrone; status unclear; Jan Just?
 
|-
 
| wiers
 
| aulnes
 
| PE1950
 
|
 
|
 
|
 
|
 
|
 
| B82MY2J
 
| former aulnes; status unknown
 
|-
 
| ent
 
| (no ipmi)
 
| Mac Mini
 
| Intel Core Duo  @1.66GHz
 
| 2
 
| 2GB
 
| OS X 10.6
 
| SATA 80GB
 
|
 
| OS X box (no virtualisation)
 
|}
 
 
 
* *ipmi name is used for IPMI access; use <code><name>.ipmi.nikhef.nl</code>.
 
* System details such as serial numbers can be retrieved from the command line with <code>dmidecode -t 1</code>.
 
* The service-tags can be retrieved through IPMI, but unless you want to send raw commands with ipmitool first you need freeipmi-tools. This contains ipmi-oem that can be called thus:
 
ipmi-oem -h host.ipmi.nikhef.nl -u username -p password dell get-system-info service-tag
 
 
 
==== IPMI serial-over-LAN ====
 
 
 
* For details, see [[Serial Consoles]].
 
* can be done by <code>ipmitool -I lanplus -H name.ipmi.nikhef.nl -U user sol activate</code>.
 
* SOL access needs to be activated in the BIOS ''once'', by setting console redirection through COM2.
 
 
 
For older systems that do not have a web interface for IPMI, the command-line version can be used. Install the OpenIPMI service so root can use ipmitool. Here is a sample of commands to add a user and give SOL access.
 
 
 
ipmitool user enable 5
 
ipmitool user set name 5 ctb
 
ipmitool user set password 5 '<blah>'
 
ipmitool channel setaccess 1 5 ipmi=on
 
# make the user administrator (4) on channel 1.
 
ipmitool user priv 5 4 1
 
ipmitool channel setaccess 1 5 callin=on ipmi=on link=on
 
ipmitool sol payload enable 1 5
 
 
 
=== Data Migration ===
 
 
 
[[Image:Testbed cloud hosts.png|thumb|250px|Function of machines in the testbed]]
 
 
 
Bleek.nikhef.nl is designated to become the home directory server, DHCP server and OpenNebula server. It will be the only persistent machine in the entire testbed, the rest should be considered volatile. It will be the only machine where backups are done. But before all this can be arranged, it needs to be reinstalled with CentOS 5 (currently CentOS 4). All important data and configurations are going to be migrated to span.nikhef.nl as an intermediate step, and after the upgrade this will be moved back and merged on bleek.
 
 
 
Disk space usage on bleek (in kB):
 
 
 
50760      etc
 
317336    lib
 
432640    opt
 
877020    export
 
1035964    root
 
2353720    var
 
3258076    usr
 
357844380  srv
 
 
 
 
 
There is a script in place on span.nikhef.nl to do the backup from bleek, where /etc/rsyncd.conf is already set up.
 
rsync -a --password-file /etc/rsync-bleek-password  --exclude /sys** --exclude /proc** --delete --delete-excluded bleek::export-perm  /srv/backup/bleek/
 
 
 
It's not run automatically, so it should be run manually at the very latest right before reinstalling bleek.
 
 
 
==== Cruisecontrol migration ====
 
 
 
The former cruisecontrol instance on bleek has been stopped. The service has ben transferred to cruisecontrol.testbed(toom.nikhef.nl), while the data in <code>/srv/project/rpmbuild</code> has been transferred to span.nikhef.nl and is exported from there with NFS.
 
 
 
=== Network plan ===
 
 
 
There are three VLANs in use. All the physical machines (i.e. the hypervisors, Dom0 in Xen terminology) should configure bridges for all three; virtual machines then get interfaces for any combination depending on their role.
 
 
 
{|
 
! vlan
 
! description
 
! network
 
! gateway
 
! ACL
 
|-
 
| 2
 
| [[NDPF_System_Functions#P4CTB|P4CTB]]
 
| 194.171.96.16/28
 
| 194.171.96.30
 
| No inbound traffic on privileged ports
 
|-
 
| 8
 
| [[NDPF_System_Functions#Nordic (Open_Experimental)|Open/Experimental]]
 
| 194.171.96.32/27
 
| 194.171.96.62
 
| Open
 
|-
 
| untagged
 
| local
 
| 10.198.0.0/16
 
|
 
| testbed only
 
|}
 
 
 
Since there is limited public IP available, we should put machines in the local network as much as possible. For outbound connectivity NATting is arranged via their Dom0.
 
 
 
The machines that run Xen 3.0 on CentOS 5 use the following configuration for networking:
 
 
 
/etc/sysconfig/network:
 
NETWORKING=yes
 
NETWORKING_IPV6=yes
 
HOSTNAME=span
 
FORWARD_IPV4=yes
 
NOZEROCONF=true
 
GATEWAY=194.171.96.30
 
GATEWAYDEV=eth0.2
 
 
 
/etc/sysconfig/network-scripts/ifcfg-eth0:
 
TYPE=Ethernet
 
DEVICE=eth0
 
HWADDR=00:1e:4f:xx:xx:xx
 
BOOTPROTO=none
 
ONBOOT=yes
 
USERCTL=no
 
IPV6INIT=no
 
IPV4INIT=yes
 
NETMASK=255.255.0.0
 
IPADDR=10.198.x.y
 
 
 
(Fill in the mac address of the actual hardware, and the 10.198.0.0 network config for the machine.)
 
 
 
/etc/sysconfig/network-scripts/ifcfg-eth0.2:
 
VLAN=yes
 
DEVICE=eth0.2
 
BOOTPROTO=static
 
ONBOOT=yes
 
TYPE=Ethernet
 
NETMASK=255.255.255.240
 
IPADDR=194.171.96.x
 
USERCTL=no
 
IPV6INIT=no
 
IPV4INIT=yes
 
 
 
This should result in a working network configuration before the start of the Xen daemon.
 
 
 
The Xen network scripts to use require an [http://renial.net/repository/perma/xen/network-bridge-vlan additional script] to configure a VLAN bridge without any virtual network devices on it. The file /etc/xen/xend-config.sxp should have:
 
(network-script 'network-multi-vlan')
 
(vif-script vif-bridge bridge=xenbr2)
 
 
 
And the script /etc/xen/scripts/network-multi-vlan contains:
 
dir=$(dirname "$0")
 
"$dir/network-bridge" "$@" vifnum=0
 
"$dir/network-bridge" "$@" vifnum=2 netdev=eth0.2
 
"$dir/network-bridge-vlan" "$@" vlan=8 netdev=peth0
 
 
 
The network-bridge-vlan is non-standard; it creates a vlan interface, binds it to a bridge but leaves it unconfigured otherwise. This way the Dom0 doesn't have a Open/Experimental interface itself, but DomUs can connect their vifs to the bridge. The xen scripts are fairly complicated in the way they rename interfaces and transfer addresses and routes to the virtual counterparts; in case things go wrong, diagnostic information may be obtained from:
 
brctl show
 
cat /proc/net/vlan/*
 
ip link show
 
netstat -rn
 
but be awary that the Xen scripts actually ''rename'' interfaces to make them look like normal ones.
 
 
 
All systems have at least 1GB interface, but put has two which may be trunked. This could be useful for serving machine images. The blade systems have extra interfaces and '''may''' be capable of doing iSCSI offloading to the NIC.
 
 
 
TODO: draw a network lay-out.
 
 
 
=== LDAP migration ===
 
 
 
We're going to ditch our own directory service (it served us well, may it rest in peace) in favour of the central Nikhef service. This means changing user ids in some (all?) cases which should be done preferable in a single swell foop.
 
 
 
We should request to add a testbed 'service' to LDAP with ourselves as managers, so we can automatically populate /root/.ssh/authorized_keys.
 
 
 
Here's a simple example of an ldapsearch call to find a certain user.
 
 
 
ldapsearch -x -H ldaps://hooimijt.nikhef.nl/ -b dc=farmnet,dc=nikhef,dc=nl uid=dennisvd
 
 
 
And here is the ldap.conf file to use for ldap authentication.
 
 
 
base dc=farmnet,dc=nikhef,dc=nl
 
timelimit 120
 
bind_timelimit 120
 
idle_timelimit 3600
 
nss_initgroups_ignoreusers root,ldap,named,avahi,haldaemon,dbus,radvd,tomcat,radiusd,news,mailman
 
uri ldaps://gierput.nikhef.nl/ ldaps://hooimijt.nikhef.nl/ ldaps://stalkaars-01.farm.nikhef.nl/ ldaps://stalkaars-03.farm.nikhef.nl/ ldaps://vlaai.nikhef.nl/
 
ssl on
 
tls_cacertdir /etc/openldap/cacerts
 
 
 
=== Migration to a cloud infrastructure ===
 
 
 
Previous testbed cloud experiences are [[Agile testbed/Cloud|reported here]].
 
 
 
Currently, using plain libvirt seems to fit most of our needs.
 
 
 
=== Installing Debian on blade 13 and 14 with Fiber Channel ===
 
 
 
This is a quick note to record a recent quirk. Although FC support on Debian works fine, using the multipath-tools-boot package is a bit tricky. It will update the initrd to include the multipath libraries and tools, to make it available at boot time.
 
 
 
This happened on blade-13; on reboot it was unable to mount the root partition (The message was 'device or resource busy') because the device mapper had somehow taken hold of the SCSI disk. By changing the root=UUID=xxxx stanza in the GRUB menu to root=/dev/dm-2 (this was guess-work) I managed to boot the system. There were probably several remedies to resolve the issue:
 
# rerun update-grub. This should replace the UUID= with a link to /dev/mapper/xxxx-part1
 
# blacklist the disk in the device mapper (and running mkinitramfs)
 
# remove the multipath-tools-boot package altogether.
 
 
 
I opted for blacklisting; this is what's in /etc/multipath.conf:
 
blacklist {
 
  wwid 3600508e000000000d6c6de44c0416105
 
}
 
 
 
 
 
==== A bonnie test of a VM with disk on local disk vs. a VM with disk on FC: ====
 
 
 
* lofarwn.testbed had its disk locally
 
Version 1.03e      ------Sequential Output------ --Sequential Input- --Random-
 
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
 
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
 
lofarwn          4G 20869  35 30172  5 23198  6 45957  85 510971  84 +++++ +++
 
                    ------Sequential Create------ --------Random Create--------
 
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
 
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
 
                  16 16759  99 +++++ +++ +++++ +++ 16810 100 +++++ +++ +++++ +++
 
lofarwn,4G,20869,35,30172,5,23198,6,45957,85,510971,84,+++++,+++,16,16759,99,+++++,+++,+++++,+++,16810,100,+++++,+++,+++++,+++
 
* ige-cert.testbed has disk on LVM via FC.
 
Version 1.03e      ------Sequential Output------ --Sequential Input- --Random-
 
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
 
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
 
ige-cert        4G 53384  96 216611  37 102283  24 51060  95 689474  79 +++++ +++
 
                    ------Sequential Create------ --------Random Create--------
 
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
 
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
 
                  16 12676 100 +++++ +++ +++++ +++ 12761  99 +++++ +++ +++++ +++
 
ige-cert,4G,53384,96,216611,37,102283,24,51060,95,689474,79,+++++,+++,16,12676,100,+++++,+++,+++++,+++,12761,99,+++++,+++,+++++,+++
 

Latest revision as of 15:38, 9 January 2015

The upgrade of the P4CTB has taken place; all information of this page is merged into Agile testbed.