ConfigMap

Description

Receives ConfigMap information of public faults to connect to the resumable training process.

If the actual parameter values in ConfigMap are different from the defined value ranges, ClusterD discards the fault information.
When faults are injected through ConfigMap or gRPC interfaces, the maximum number of faults on all nodes is 50,000. If this threshold is exceeded, ClusterD discards any newly injected fault information.
The label of ConfigMap must be mc-consumer-publicfault=true, and the data key must be PublicFault.

When ConfigMap is used to send public faults, the data volume cannot exceed 1 MB. Otherwise, ConfigMap fails to be updated.

Parameters

For details, see the following table.

**Table 1** Fault description
Parameter	Meaning	Value	Type	Mandatory or Not
id	Unique ID of a message	A string of 8 to 128 characters, including uppercase letters, lowercase letters, digits, hyphens (-), underscores (_), and periods (.). The value must be unique.	String	Yes
timestamp	Timestamp for message sending	The value is a 13-digit number in ms and must be later than 2025-01-01T00:00:00Z.	Int64	Yes
version	Message version	The value is 1.0.	String	Yes
resource	Fault sender	The value can be CCAE, fd-online, pingmesh, Netmind, or dpcStorage. NOTE: The fault sender of public faults must exist in publicFaultResource of the fault configuration file. For new fault senders, manually configure them in the fault configuration file. For details, see (Optional) Configuring the Public Fault Level and Sender.	String	Yes
faults	Fault details	The value is a slice, whose length is greater than 0 and less than or equal to 100.	[]object, fault	Yes

**Table 2** Fault field description
Parameter	Meaning	Value	Type	Mandatory or Not
faultId	Fault instance ID	A string of 8 to 128 characters, including uppercase letters, lowercase letters, digits, hyphens (-), underscores (_), and periods (.). The value must be unique. NOTE: Even for the same fault instance, the value of faultId must be unique.	String	Yes
faultType	Fault type	The value can be NPU, Node, Network, or Storage. NPU: processor fault Node: node fault Network: network fault Storage: storage fault NOTE: This field is displayed as PublicFault in cluster-info-cm.	String	Yes
faultCode	Fault code	The value can be customized and must be unique. The value consists of nine characters. For details, see Fault Code Description. NOTE: The fault code for resumable training must exist in publicFaultCode of the fault configuration file. For new fault codes, configure the fault level in the fault configuration file. For details, see (Optional) Configuring the Public Fault Level and Sender. You are advised to define fault codes based on the fault code description table for subsequent maintenance. If the same fault code is generated on an NPU twice, the fault_code field in cluster-info-cm records the same fault code twice.	String	Yes
faultTime	Time when a fault occurs	The value is a 13-digit number in ms and must be later than 2025-01-01T00:00:00Z. NOTE: This field specifies the fault generation time only, regardless of whether a fault is generated or rectified. This field is displayed in the unit of seconds in cluster-info-cm.	Int64	Yes
assertion	Fault status	The value can be occur, recover, or once. occur: fault occurrence recover: fault recovery once: one-off event NOTE: To clear a public fault, write the recover event of the fault to ConfigMap. The clearance cannot be implemented by directly deleting ConfigMap. For a one-off event, the fault is automatically cleared several seconds later.	String	Yes
faultLocation	Fault location	Fault source, with a length less than or equal to 10 characters. The length of the key in the map is less than or equal to 16 characters, and the length of the value is less than or equal to 128 characters. For example: key: npuIp, value: ip	map[string]string	No
influence	Fault impact scope	The value is a slice, whose length is greater than 0 and less than or equal to 1000.	[]object, faultInfo	Yes
description	Fault description	The value is a string of 0 to 512 characters, containing non-whitespace characters and spaces.	String	No

**Table 3** faultInfo description
Field	Meaning	Value	Type	Mandatory or Not
nodeName	Node name. You can run the kubectl get nodes -owide command to query the node name.	The value is a string of 1 to 253 characters, including lowercase letters, digits, hyphens (-), and periods (.). It must start and end with a letter or digit. If this field exists, nodeSN is not used. NOTE: If the node name does not exist in a Kubernetes cluster, ClusterD does not display a message indicating that the node name is incorrect, nor does it write the fault information into cluster-info-device-cm.	String	Alternative
nodeSN	Node SN	The value is the node annotation written by NodeD. The key is product-serial-number. NOTE: If this field is used instead of nodeName, install NodeD in advance.	String	Alternative
deviceIds	Physical processor ID	The length range is (0, 32], and element range is [0, 32). The values must be unique. NOTE: If the faulty processor cannot be accurately located, enter the physical IDs of all processors on the node. If a physical ID of a processor that does not exist on a node is passed, ClusterD also displays its physical ID in cluster-info-device-cm.	[]int32	Yes

Parent topic: Public Fault APIs