Difference between revisions of "Machine/bender"

Latest revision as of 17:12, 1 July 2024

Info

Bender and Papert are two twin blade servers donated to Sugar Labs in 2021.

Bender is our primary KVM host, while Papert is a hot standby and backup machine.

Hostnames

bender.sugarlabs.org
papert.sugarlabs.org

Hardware

HPE ProLiant DL360 Gen10 1RU server
- Dual socket, current configuration has one CPU
Xeon Gold 5218R
- 20 cores/40 threads
- 2.1GHz base frequency, 4.0GHz max turbo frequency
- 27.5MB of cache
64GB RAM
2TB NVMe SSD

Admins

Alex Perez, MrBIOS on #sugar Libera.chat IRC network
Bernie Innocenti, @bernie:matrix.org on Sugar Systems

Hosting

Hosted by Sonic in Santa Rosa CA

Network configuration

Network configuration is managed via netplan. To modify, do:

vi /etc/netplan/bender.yaml
netplan generate
netplan apply

IPv4

Bender and Papert are globally accessible through a public, static IPv4 address

Sonic assigned a /28 network to Sugar Labs. IP assignments are managed in our DNS configuration. Search for "Sonic IP pool" in masters/sugarlabs.org.zone.

Usable IPv4 addresses: 192.184.220.210 ~ 192.184.220.222 (13 addresses)
Subnet Mask: 255.255.255.240
Default Gateway: 192.184.220.209
DNS: 8.8.8.8, 8.8.4.4

IPv6

IPv6 configuration is a bit weird.

* Public block: 2001:5a8:601:f::/64
* Sonic gateway: 2001:5a8:5:3a::15:0/127
* Transport IP: 2001:5a8:5:3a::15:1/127

The gateway is configured to route all traffic for our netblock to the transport IP, which is currently assigned to bender.

Bridges

The br0 bridge is created at startup and shared with the virtual machines hosted on Bender. It gives the VMs unfiltered access to the external network. There's no DHCP, all machines must define a static IP configuration, taking care not to stomp onto each other.

There is also a virbr0 bridge created by libvirt on startup from /etc/libvirt/qemu/networks/default.xml. This is a NAT interface and is not meant for VMs directly serving on the Internet.

Guests simply need to be configured to accept IPv6 routing advertisements. The DNS must be assigned manually.

Virtualization

Virtual machines are KVM guests managed with libvirt.

Storage

All virtual disks come from a pool backed by the main LVM VG:

 virsh # pool-define-as nvme-pool logical --source-name nvme-pool 
 Pool nvme-pool defined
 
 virsh # pool-start nvme-pool
 Pool nvme-pool started
 
 virsh # pool-autostart nvme-pool
 Pool nvme-pool marked as autostarted
 
 virsh # pool-info nvme-pool
 Name:           nvme-pool
 UUID:           5812819f-b8bf-484e-98fb-2e100fe83df2
 State:          running
 Persistent:     yes
 Autostart:      yes
 Capacity:       1.64 TiB
 Allocation:     250.00 GiB
 Available:      1.40 TiB

Disks assigned to VMs will appear here:

virsh # vol-list nvme-pool
 Name         Path
-----------------------------------------
 aslo1-root   /dev/nvme-pool/aslo1-root
 aslo1-srv    /dev/nvme-pool/aslo1-srv
 backup       /dev/nvme-pool/backup

Disks can be created, listed and deleted using the vol-* commands:

 virsh # vol-create-as nvme-pool testvm-root 20G
 Vol testvm-root created
 
 virsh # vol-info testvm-root --pool nvme-pool
 Name:           testvm-root
 Type:           block
 Capacity:       20.00 GiB
 Allocation:     20.00 GiB
 
 virsh # vol-delete testvm-root --pool nvme-pool
 Vol testvm-root deleted

NOTE: avoid allocating large VM volumes as image file within the host's root filesystem, as they're slow and hard to manage. It's ok to use images for test VMs.

@@ Line 1: / Line 1: @@
-<noinclude>{{TOCright}}</noinclude>
+== Info ==
+Bender and [[Machine/papert | Papert]] are two twin blade servers donated to Sugar Labs in 2021.
+Bender is our primary KVM host, while Papert is a hot standby and [[Service/backup|backup]] machine.
 == Hostnames ==
-* bender.codewiz.org
 * bender.sugarlabs.org
+* papert.sugarlabs.org
 == Hardware ==
-* Crappy minitower
+* HPE ProLiant DL360 Gen10 1RU server
-* Core 2 Duo 1.8GHz
+** Dual socket, current configuration has one CPU
-* 6GB RAM
+* [https://ark.intel.com/content/www/us/en/ark/products/199342/intel-xeon-gold-5218r-processor-27-5m-cache-2-10-ghz.html Xeon Gold 5218R]
-* 2x1TB RAID1, 2x320GB RAID1
+** 20 cores/40 threads
-* Fedora 13 x86_64
+** 2.1GHz base frequency, 4.0GHz max turbo frequency
+** 27.5MB of cache
+* 64GB RAM
+* 2TB NVMe SSD
-== Location ==
+[[Image:BenderRacked.jpg|thumb|320px]]
-Hosted by [http://www.develer.com/ Develer]
 == Admins ==
-* [[User:Bernie|Bernie Innocenti]], _bernie on #sugar Freenode
+* [[User:MrBIOS|Alex Perez]], MrBIOS on #sugar Libera.chat IRC network
-* [[User:dnarvaez|Daniel Narvaez]], dnarvaez on #sugar Freenode
+* [[User:Bernie|Bernie Innocenti]], @bernie:matrix.org on Sugar Systems
-* Stefano Fedrigo <aleph AT develer.com>, sometimes _aleph on Freenode (local access, office hours CET)
-* Develer Infrastructure <it AT lists.develer.com> (local access, office hours CET)
+== Hosting ==
+[[Image:SonicColo.jpg|thumb|320px]]
+Hosted by Sonic in Santa Rosa CA
 == Network configuration ==
-Develer has a 10mbit up/downlink and asks to limit bandwidth usage from hosted services.
+Network configuration is managed via [https://netplan.io/ netplan]. To modify, do:
-There is a traffic shaper, but it may have trouble shaping 6to4 traffic.
+ vi /etc/netplan/bender.yaml
+ netplan generate
+ netplan apply
+=== IPv4 ===
+Bender and Papert are globally accessible through a public, static IPv4 address
-Bender is globally accessible through public, static IPv4 as well as the 6to4 subnet associated
+Sonic assigned a /28 network to Sugar Labs. IP assignments are managed in our DNS configuration. Search for "Sonic IP pool" in <code>masters/sugarlabs.org.zone</code>.
-to it: 2002:5395:9edc::/48.
-The tun6to4 interface on bender is assigned the globally visible address 2002:5395:9edc::1.
+* Usable IPv4 addresses: 192.184.220.210 ~ 192.184.220.222 (13 addresses)
-The subnet 2002:5395:9edc:1::/64 of our 6to4 net is assigned to the bridge virbr0, which binds together
+* Subnet Mask:     255.255.255.240
-several virtual interfaces connected to the libvirt guests. With this network setup, IPv6 routing works
+* Default Gateway: 192.184.220.209
-naturally, without the need to add any special routing rules on bender.
+* DNS:             8.8.8.8, 8.8.4.4
-The virbr0 bridge is created by libvirt on startup with <code>/etc/libvirt/qemu/networks/default.xml</code>.
+=== IPv6 ===
-Libvirt does not yet support assigning IPv6 addresses to bridges, therefore we do this in <code>/etc/rc.local</code>:
+IPv6 configuration is a bit weird.
-  ip addr add 2002:5395:9edc:1::1/64 dev virbr0
+  * Public block: 2001:5a8:601:f::/64
+ * Sonic gateway: 2001:5a8:5:3a::15:0/127
+ * Transport IP: 2001:5a8:5:3a::15:1/127
-To automatically configure network and on the gursts, Bender also runs radvd, the IPv6 Routing Advertisement
+The gateway is configured to route all traffic for our netblock to the transport IP, which is currently assigned to bender.
-daemon. The contents of <code>/etc/radvd.conf</code> are:
- interface virbr0
+=== Bridges ===
- {
+The br0 bridge is created at startup and shared with the virtual machines hosted on Bender. It gives the VMs unfiltered access to the external network. There's no DHCP, all machines must define a static IP configuration, taking care not to stomp onto each other.
-        IgnoreIfMissing on;
-        AdvSendAdvert on;
+There is also a virbr0 bridge created by libvirt on startup from <code>/etc/libvirt/qemu/networks/default.xml</code>. This is a NAT interface and is not meant for VMs directly serving on the Internet.
-        MinRtrAdvInterval 30;
-        MaxRtrAdvInterval 100;
-        AdvDefaultPreference low;
-        AdvHomeAgentFlag off;
-        #bernie: subnet 1 of our /48 6to4 on Develer Consiagnet
-        prefix 2002:5395:9edc:1::1/64
-        {
-                AdvOnLink on;
-                AdvAutonomous on;
-        };
- };
 Guests simply need to be configured to accept IPv6 routing advertisements. The DNS must be assigned manually.
+== Virtualization ==
+Virtual machines are KVM guests managed with libvirt.
+=== Storage ===
+All virtual disks come from a pool backed by the main LVM VG:
+  virsh # pool-define-as nvme-pool logical --source-name nvme-pool
+  Pool nvme-pool defined
+  virsh # pool-start nvme-pool
+  Pool nvme-pool started
+  virsh # pool-autostart nvme-pool
+  Pool nvme-pool marked as autostarted
+  virsh # pool-info nvme-pool
+  Name:           nvme-pool
+  UUID:           5812819f-b8bf-484e-98fb-2e100fe83df2
+  State:          running
+  Persistent:     yes
+  Autostart:      yes
+  Capacity:       1.64 TiB
+  Allocation:     250.00 GiB
+  Available:      1.40 TiB
-== Hosted VMs ==
+Disks assigned to VMs will appear here:
-All buildslaves currently run off bender as KVM virtual machines managed by libvirtd.
-The machines are not globally addressable on IPv4 and are only reachable by IPv6.  The
+  virsh # vol-list nvme-pool
-IPv6 address is dynamically configured by radvd to a subnet of the public 6to4 net.
+  Name         Path
+ -----------------------------------------
+  aslo1-root   /dev/nvme-pool/aslo1-root
+  aslo1-srv    /dev/nvme-pool/aslo1-srv
+  backup       /dev/nvme-pool/backup
-All the VMs are in /srv/images. See the README in the same directory.
+Disks can be created, listed and deleted using the vol-* commands:
-They have been created with
+  virsh # vol-create-as nvme-pool testvm-root 20G
+  Vol testvm-root created
+  virsh # vol-info testvm-root --pool nvme-pool
+  Name:           testvm-root
+  Type:           block
+  Capacity:       20.00 GiB
+  Allocation:     20.00 GiB
+  virsh # vol-delete testvm-root --pool nvme-pool
+  Vol testvm-root deleted
- virt-install --name buildslave-name --ram 512 \
+'''NOTE''': avoid allocating large VM volumes as image file within the host's root filesystem, as they're slow and hard to manage. It's ok to use images for test VMs.
- --disk path=/srv/images/buildslave-name.img,size=10
- --network network:default \
- --location "mirror-http-address" \
- --extra-args="console=tty0 console=ttyS0,115200n8 serial"