Skip to content

Conversation

@Nieuwejaar
Copy link
Collaborator

No description provided.

@Nieuwejaar
Copy link
Collaborator Author

Notes:

  • This doesn't change the number of routes. We're still limited to 8k IPv4 and 1k IPv6.
  • This eliminates the protocol-multiplexing RouteTarget enum, and makes the arguments to the route APIs protocol-specific. The old way required pointless (un)marshaling on both sides, as well as requiring us to add explicit error checks that are now implicit in the type processing. This means this PR will require associated PRs in omicron, and maghemite.
  • This PR will also require a new version of sidecar-lite, which has also been updated to support IPv6 multipath.

Copy link
Contributor

@rcgoodfellow rcgoodfellow left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks Nils. LGTM just a few small comments. I think we'll also want to determine what the distribution looks like for the masked CRC8 so that is a known characteristic of the design.

meta.l4_dst_port,
meta.l4_src_port
});
}) & 0x3f;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a comment on the bit trading taking place through this mask (and for v6 below).

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done. Noted in the ipv6 path as well.

.map(|_| info!(s.log, "deleted ipv6 index"; "route" => %cidr))
.map_err(|e| {
error!(s.log, "failed to delete ipv6 index";
"route" => %cidr,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Something a bit wonky with the formatting here. And several spots below.

@Nieuwejaar
Copy link
Collaborator Author

Thanks Nils. LGTM just a few small comments. I think we'll also want to determine what the distribution looks like for the masked CRC8 so that is a known characteristic of the design.

A ran a test with a billion samples. Source and destination IPs were random within the 10.x.x.x subnet, and source and destination ports were random between 1000 and 10000. With the masked crc, there were 64 possible buckets into which these samples could land. The resulting bucket sizes looked like:

min: 15615310 max: 15634865 avg: 15625000 std_dev: 4071.0562 cv: 0.0002605476

ecmp support for IPv6
@Nieuwejaar Nieuwejaar merged commit 72461d3 into main Sep 3, 2025
6 checks passed
@Nieuwejaar Nieuwejaar deleted the ecmp branch September 3, 2025 15:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants