IPv6-safe store address parsing (3 line change)

[BenCowen]

Motivation

Add support for IPv6 fabric.

Summary

create_store_client parses the rendezvous store address with partition(":"), taking everything after the first colon as the port. That's the correct assumption for host:port / IPv4, but it doesn't apply to an IPv6 address, for example on a flat-IPv6 fabric where containers address each other directly by IPv6:

store_addr = "fdaa:5137:3ebf:a75:...:29521/torchft/<quorum>/<rank>"
host, _, rest = store_addr.partition(":")     # host = "fdaa"  (!)
port, _, prefix = rest.partition("/")          # port = "5137:3ebf:...:29521"
TCPStore(host_name="fdaa", port=int(port))     # ValueError: invalid literal for int()

Fix

Split off the /prefix first, then take the port as the field after the last colon, and strip optional brackets from the host. Correct for IPv4, and for bare or bracketed IPv6 ([host]:port). One function, no signature change.

Testing

Verified end-to-end on a flat-IPv6 fabric (Modal i6pn), single-process replica groups across separate containers:

• Lighthouse quorum + cross-replica allreduce (Gloo on CPU, and NCCL on GPU).
• DiLoCo training to convergence.
• Fault recovery: a killed replica's replacement heals its checkpoint from a peer over IPv6 (HTTPTransport) and rejoins.

Existing host:port / IPv4 behavior is unchanged.

[meta-cla]

Hi @BenCowen!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[meta-cla]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[BenCowen]

Friendly ping @d4l3k @tushar00jain : 3-line, self-contained IPv6 parse fix, plus a unit test.

[d4l3k]

LGTM

[meta-codesync]

@d4l3k has imported this pull request. If you are a Meta employee, you can view this in D112226662. (Because this pull request was imported automatically, there will not be any future comments.)

[d4l3k]

@BenCowen can you rebase on main and fix the lint issues? trunk is green now

I tried to fix it but I don't have permissions to your branch

[BenCowen]

Thanks @d4l3k, all set!

[meta-codesync]

This pull request has been merged in 90d7f68.