You do not need to declare the dstPrefix variable. Just reuse the dstPrefix from the function input parameters.

Fixed - using dstNamePrefix now. dstPrefix is for the name inside the host - thus using it here is wrong and confusing.

As this is internal, please rename to ContainerIfacePrefix so that we can reduce this change's footprint.

Ack

For readability, I would rename the constant to defaultContainerVethPrefix and here have

containerVethPrefix := defaultContainerVethPrefix
if network.config.ContainerIfacePrefix != "" {
    containerVethPrefix = network.config.ContainerInterfacePrefix
}
err = iNames.SetNames(endpoint.srcName, containerVethPrefix)

Fixed

Again to reduce the fix footprint, I would leave the variable name as is, even if it now is a map. At the end it is one index per interface.

Ack

please remove this line
and modify next one to

i.dstName = fmt.Sprintf("%s%d", dstPrefix, n.nextIfIndex[dstPrefix])

dstPrefix is the prefix of the interface in the host system - I am trying to modify the one in the container. If I override this it will change the logic in the code following. I do not think this is correct - please clarify.

The only reason I assign dstName to a new dstNamePrefix is that the original data of dstName is to be modified and dstNamePrefix is needed for the increment.

i.srcName is the original name of the veth pipe end which is going to be moved into the container's netns.
i.dstName is the new name the above interface will be renamed to, once it is moved.

dstPrefix is the prefix of the interface in the host system

No, dstPrefix the network driver chosen prefix for the new interface name once we move it into the netns.
func (n *networkNamespace) AddInterface(srcName, dstPrefix string, options ...IfaceOption) error {

At this point of the code i.dstName == dstPrefix (L229 i := &nwIface{srcName: srcName, dstName: dstPrefix, ns: n})
IMO, original code was already confusing and should have been written as
i.dstName = fmt.Sprintf("%s%d", dstPrefix, n.nextIfIndex)

You are of course correct. Should have read up a few lines in the code. Have made this modification now as well.

Looking at this again, it makes me think people may want the same option while using different network drivers in future.

In a way it s not specific to a network driver (like the MTU option we have now).

So I was thinking it may be better we make this option driver agnostic from the beginning, and define it in https://github.com/docker/libnetwork/blob/master/netlabel/labels.go

as

ContainerIfacePrefix = "com.docker.network.container_iface_prefix

Also wondering whether we should keep a 1-to-1 match to the networking objects (here would be sandbox_iface_PREFIX) or stick to container given it is a user input option. It is probably better to leave it as container_iface_prefix, more clear.

That is now done as well.

j/c. are you relying on the fact that the map access on new prefix will return 0 ?

Correct - basically just moved the original method towards using a map to track all prefixes independently.