This procedure will add a liquid-cooled blade to an HPE Cray EX system.
The Cray command line interface (CLI) tool is initialized and configured on the system. See Configure the Cray Command Line Interface.
Knowledge of whether DVS is operating over the Node Management Network (NMN) or the High Speed Network (HSN).
Blade is being added to an existing liquid-cooled cabinet in the system.
The Slingshot fabric must be configured with the desired topology for desired state of the blades in the system.
The System Layout Service (SLS) must have the desired HSN configuration.
Check the status of the high-speed network (HSN) and record link status before the procedure.
The System Admin Toolkit (SAT) is installed and configured on the system.
The blades must have the coolant drained and filled during the swap to minimize cross-contamination of cooling systems.
(ncn-mw#) Begin discovery for the blade.
Use the sat swap command to map the nodes’ Ethernet interface MAC addresses to the appropriate IP addresses and component names (xnames), and begin discovery for the blade.
The --src-mapping and --dst-mapping arguments may be used to pass in the Ethernet interface mapping files containing the IP
addresses, MAC addresses, and component xnames for the nodes on the blade. If the slot was previously populated, the file passed into the
--src-mapping argument should be the mapping file saved during the
Removing a Liquid-cooled Blade from a System Using SAT procedure, and the
--dst-mapping argument should be a mappings file of the MAC addresses, IP addresses, and component xnames from the new blade. If the new
blade was removed from another system, the mappings file was saved while performing the
Removing a Liquid-cooled Blade from a System Using SAT procedure on the other system.
If the blade is a new blade, (that is, not from another running system), do not use either --src-mapping or --dst-mapping.
sat swap blade --src-mapping <SRC_MAPPING> --dst-mapping <DST_MAPPING> --action enable <SLOT_XNAME>
If the slot was not previously populated, the --src-mapping and --dst-mapping arguments should be omitted.
sat swap blade --action enable <SLOT_XNAME>
(ncn-mw#) Clear out the existing Redfish event subscriptions from the BMCs on the blade.
Set the environment variable SLOT to the blade’s location.
SLOT=<SLOT_XNAME>
Clear the Redfish event subscriptions.
export TOKEN=$(curl -s -S -d grant_type=client_credentials \
-d client_id=admin-client \
-d client_secret=`kubectl get secrets admin-client-auth -o jsonpath='{.data.client-secret}' | base64 -d` \
https://api-gw-service-nmn.local/keycloak/realms/shasta/protocol/openid-connect/token | jq -r '.access_token')
for BMC in $(cray hsm inventory redfishEndpoints list --type NodeBMC --format json | jq .RedfishEndpoints[].ID -r | grep ${SLOT}); do
/usr/share/doc/csm/scripts/operations/node_management/delete_bmc_subscriptions.py $BMC
done
Each BMC on the blade will have output like the following:
Clearing subscriptions from NodeBMC x3000c0s9b0
Retrieving BMC credentials from SCSD
Retrieving Redfish Event subscriptions from the BMC: https://x3000c0s9b0/redfish/v1/EventService/Subscriptions
Deleting event subscription: https://x3000c0s9b0/redfish/v1/EventService/Subscriptions/1
Successfully deleted https://x3000c0s9b0/redfish/v1/EventService/Subscriptions/1
(ncn-mw#) Determine which Boot Orchestration Service (BOS) templates to use to shut down nodes on the target blade.
There will be separate session templates for UANs and computes nodes.
List all the session templates.
If it is unclear which session template is in use, proceed to the next substep.
cray bos v1 sessiontemplate list
Find the node xnames with sat status. In this example, the target blade is in slot x9000c3s0.
sat status --filter 'xname=x9000c3s0*'
Example output:
+---------------+------+----------+-------+------+---------+------+-------+-------------+----------+
| xname | Type | NID | State | Flag | Enabled | Arch | Class | Role | Net |
+---------------+------+----------+-------+------+---------+------+-------+-------------+----------+
| x9000c3s0b1n0 | Node | 1 | Off | OK | True | X86 | River | Compute | Sling |
| x9000c3s0b2n0 | Node | 2 | Off | OK | True | X86 | River | Compute | Sling |
| x9000c3s0b3n0 | Node | 3 | Off | OK | True | X86 | River | Compute | Sling |
| x9000c3s0b4n0 | Node | 4 | Off | OK | True | X86 | River | Compute | Sling |
+---------------+------+----------+-------+------+---------+------+-------+-------------+----------+
Find the bos_session value for each node via the Configuration Framework Service (CFS).
cray cfs components describe x9000c3s0b1n0 --format toml | grep bos_session
Example output:
bos_session = "e98cdc5d-3f2d-4fc8-a6e4-1d301d37f52f"
Find the required templateName value with BOS.
cray bos v1 session describe BOS_SESSION --format toml | grep templateName
Example output:
templateName = "compute-nid1-4-sessiontemplate"
Determine the list of xnames associated with the desired session template.
cray bos v1 sessiontemplate describe SESSION_TEMPLATE_NAME --format toml | grep node_list
Example output:
node_list = [ "x9000c3s0b1n0", "x9000c3s0b2n0", "x9000c3s0b3n0", "x9000c3s0b4n0",]
(ncn-mw#) Power on and boot the nodes.
Use sat bootsys to power on and boot the nodes. Specify the appropriate component name (xname) for the slot, and
a comma-separated list of the BOS session templates determined in the previous step.
BOS_TEMPLATES=cos-2.0.30-slurm-healthy-compute
sat bootsys boot --stage bos-operations --bos-limit x9000c3s0 --recursive --bos-templates $BOS_TEMPLATES
Validate the firmware.
Verify that the correct firmware versions are present for the node BIOS, node controller (nC), NIC mezzanine card (NMC), GPUs, and so on.
Review FAS Admin Procedures to perform a dry run using FAS to verify firmware versions.
If necessary, update firmware with FAS. See Update Firmware with FAS for more information.
There should be a cray-cps pod (the broker), three cray-cps-etcd pods and their waiter, and at least one cray-cps-cm-pm pod.
Usually there are two cray-cps-cm-pm pods: one on ncn-w002 and one on another worker node.
(ncn-mw#) Check the cray-cps pods on worker nodes and verify they are Running.
kubectl get pods -Ao wide | grep cps
Example output:
services cray-cps-75cffc4b94-j9qzf 2/2 Running 0 42h 10.40.0.57 ncn-w001
services cray-cps-cm-pm-g6tjx 5/5 Running 21 41h 10.42.0.77 ncn-w003
services cray-cps-cm-pm-kss5k 5/5 Running 21 41h 10.39.0.80 ncn-w002
services cray-cps-etcd-knt45b8sjf 1/1 Running 0 42h 10.42.0.67 ncn-w003
services cray-cps-etcd-n76pmpbl5h 1/1 Running 0 42h 10.39.0.49 ncn-w002
services cray-cps-etcd-qwdn74rxmp 1/1 Running 0 42h 10.40.0.42 ncn-w001
services cray-cps-wait-for-etcd-jb95m 0/1 Completed
(ncn-w#) SSH to each worker node running CPS/DVS and run dmesg -T.
Ensure that there are no recurring "DVS: merge_one" error messages shown. These error messages indicate that DVS is detecting an IP address change for one of the client nodes.
dmesg -T | grep "DVS: merge_one"
Example output:
[Tue Jul 21 13:09:54 2020] DVS: merge_one#351: New node map entry does not match the existing entry
[Tue Jul 21 13:09:54 2020] DVS: merge_one#353: nid: 8 -> 8
[Tue Jul 21 13:09:54 2020] DVS: merge_one#355: name: 'x3000c0s19b1n0' -> 'x3000c0s19b1n0'
[Tue Jul 21 13:09:54 2020] DVS: merge_one#357: address: '10.252.0.26@tcp99' -> '10.252.0.33@tcp99'
[Tue Jul 21 13:09:54 2020] DVS: merge_one#358: Ignoring.
(nid#) SSH to the client node and check each DVS mount.
mount | grep dvs | head -1
Example output:
/var/lib/cps-local/0dbb42538e05485de6f433a28c19e200 on /var/opt/cray/gpu/nvidia-squashfs-21.3 type dvs (ro,relatime,blksize=524288,statsfile=/sys/kernel/debug/dvs/mounts/1/stats,attrcache_timeout=14400,cache,nodatasync,noclosesync,retry,failover,userenv,noclusterfs,killprocess,noatomic,nodeferopens,no_distribute_create_ops,no_ro_cache,loadbalance,maxnodes=1,nnodes=6,nomagic,hash_on_nid,hash=modulo,nodefile=/sys/kernel/debug/dvs/mounts/1/nodenames,nodename=x3000c0s6b0n0:x3000c0s5b0n0:x3000c0s4b0n0:x3000c0s9b0n0:x3000c0s8b0n0:x3000c0s7b0n0)
(ncn-mw#) Determine the pod name for the Slingshot fabric manager pod and check the status of the fabric.
kubectl exec -it -n services $(kubectl get pods --all-namespaces |grep slingshot | awk '{print $2}') -- fmn_status
(ncn-mw#) Check for duplicate IP address entries in the Hardware State Management Database (HSM). Duplicate entries will cause DNS operations to fail.
Verify that each node hostname resolves to a single IP address.
nslookup x9000c3s0b0n1
Example output with only one IP address resolving:
Server: 10.92.100.225
Address: 10.92.100.225#53
Name: x9000c3s0b0n1
Address: 10.100.0.26
Reload the KEA configuration.
curl -s -k -H "Authorization: Bearer ${TOKEN}" -X POST -H "Content-Type: application/json" \
-d '{ "command": "config-reload", "service": [ "dhcp4" ] }' https://api-gw-service-nmn.local/apis/dhcp-kea | jq
If there are no duplicate IP addresses within HSM, the following response is expected:
[
{
"result": 0,
"text": "Configuration successful."
}
]
If there is a duplicate IP address, then an error message similar to the message below is expected. This example message indicates a duplicate IP address (10.100.0.105) in the HSM:
[{'result': 1, 'text': "Config reload failed: configuration error using file '/usr/local/kea/cray-dhcp-kea-dhcp4.conf': failed to add new host using the HW address '00:40:a6:83:50:a4 and DUID '(null)' to the IPv4 subnet id '0' for the address 10.100.0.105: There's already a reservation for this address"}]
(ncn-mw#) Delete the duplicate entries, if there are any.
If there are duplicate entries in the HSM as a result of this procedure, then delete the duplicate entries (10.100.0.105 in this example).
Show the EthernetInterfaces for the duplicate IP address:
cray hsm inventory ethernetInterfaces list --ip-address 10.100.0.105 --format json | jq
Example output for an IP address that is associated with two MAC addresses:
[
{
"ID": "0040a68350a4",
"Description": "Node Maintenance Network",
"MACAddress": "00:40:a6:83:50:a4",
"IPAddresses": [
{
"IPAddress": "10.100.0.105"
}
],
"LastUpdate": "2021-08-24T20:24:23.214023Z",
"ComponentID": "x9000c3s0b0n1",
"Type": "Node"
},
{
"ID": "0040a683639a",
"Description": "Node Maintenance Network",
"MACAddress": "00:40:a6:83:63:9a",
"IPAddresses": [
{
"IPAddress": "10.100.0.105"
}
],
"LastUpdate": "2021-08-27T19:15:53.697459Z",
"ComponentID": "x9000c3s0b0n1",
"Type": "Node"
}
]
Delete the older entry.
cray hsm inventory ethernetInterfaces delete 0040a68350a4
(ncn-mw#) Use the following example curl command to check for active DHCP leases.
If there are zero DHCP leases, then there is a configuration error.
curl -H "Authorization: Bearer ${TOKEN}" -X POST -H "Content-Type: application/json" \
-d '{ "command": "lease4-get-all", "service": [ "dhcp4" ] }' https://api-gw-service-nmn.local/apis/dhcp-kea | jq
Example output with no active DHCP leases:
[
{
"arguments": {
"leases": []
},
"result": 3,
"text": "0 IPv4 lease(s) found."
}
]
(ncn-mw#) Check DNS using nslookup.
nslookup 10.100.0.105
Example output:
105.0.100.10.in-addr.arpa name = nid001032-nmn.
105.0.100.10.in-addr.arpa name = nid001032-nmn.local.
105.0.100.10.in-addr.arpa name = x1005c3s0b0n0.
105.0.100.10.in-addr.arpa name = x1005c3s0b0n0.local.
(ncn-mw#) Verify the ability to connect using SSH.
ssh x9000c3s0b0n1
Example output:
The authenticity of host 'x9000c3s0b0n1 (10.100.0.105)' can't be established.
ECDSA key fingerprint is SHA256:wttHXF5CaJcQGPTIq4zWp0whx3JTwT/tpx1dJNyyXkA.
Are you sure you want to continue connecting (yes/no/[fingerprint])? yes
Warning: Permanently added 'x1005c3s0b0n0' (ECDSA) to the list of known hosts.
Last login: Tue Aug 31 10:45:49 2021 from 10.252.1.9