IP Family Inconsistency (EI0001)

Symptom

The CANN logs contain the keyword "rank[*] device ip family[2] is not same as others[*]," as shown below:

[ERROR] HCCL(144905,python):2025-04-20-00:26:54.435.048 [config.cc:413] [145735][InitGroupStage][RanktableCheck]rank[0] device ip family[2] is not same as others[10].

Possible Cause

The IP families obtained by the two ranks are different. For example, one is IPv4 and the other is IPv6.

Solution

Check whether IPv4 is configured.

hccn_tool -i {deviceId} -ip -g

Check whether IPv6 is configured.

hccn_tool -i {deviceId} -ip -inet6 -g

The IP families of all ranks in a job must be the same. By default, HCCL uses IPv4. If no IPv4 address is configured, it uses IPv6. You can use the HCCL_SOCKET_FAMILY environment variable to specify the NIC IP version.

Note: The family is printed as an enumerated value. The following table lists the enumerated values and their mapping IP versions.