ct: add support for mark and labels #434
Conversation
atenart
force-pushed
the
at/ct-mark-label
branch
from
72cb70f
to
3f70aed
Compare
| /// u128 representation in the events. We can't use the Rust primitive as serde | ||
| /// does not handle the type well. | ||
| #[event_type] | ||
| pub struct U128 { | ||
| hi: u64, | ||
| lo: u64, | ||
| } |
There was a problem hiding this comment.
Due to serde limitations I chose to represent u128 as the above. Another way would be to use something like U128(u128) and represent it as a string in JSON.
atenart
changed the title
|
Feels like I'm getting some random garbage instead of labels. Every time I run the test they are different or not show up at all. For example: It supposed to be the same ct entry, but the labels are not shown the first time. And the The test that produced this is using either |
|
For the formatting of the parent entry, it looks like this: It might be better if the parent was indented further to the right, so it better aligns with the current connection and that it looks more like part of the |
Thanks for testing! I believe there are two issues:
(The above is probably a combination of the two issues). Also there is the issue reported in #445. [Edit] [1] I think the issue is the |
IMO we should avoid to shift things more, we already have events (in series) and their sections. Maybe instead we can use a special char to start the parent line? WDYT? |
As it contains helpers. Signed-off-by: Antoine Tenart <[email protected]>
Signed-off-by: Antoine Tenart <[email protected]>
While the mark is an optional feature that can be compiled out, let's not use a dedicated event section for just an u32. Instead we detect if the mark value should be seen as valid in the Rust side while processing the raw event. Signed-off-by: Antoine Tenart <[email protected]>
While labels are an optional feature that can be compiled out, let's not use a dedicated event section for just 128 bits. Instead we detect if the labels value should be seen as valid in the Rust side while processing the raw event. Also 0 means there is no label. Signed-off-by: Antoine Tenart <[email protected]>
atenart
force-pushed
the
at/ct-mark-label
branch
from
3f70aed
to
1221269
Compare
I like how the protocol and reply align in the shifted version, and it's easier to compare IPs this way. But I won't insist. Having a special character seems better than not having it. |
|
@igsilya the various issues should be fixed now. I refreshed the container image. |
|
Updated version seems to work: Tested with RHEL 9.4 stock kernel and 6.10-rc6 upstream kernel. I no longer see any garbage data and the labels are the labels match what OVS has in OpenFlow rules. Also works without kenrel config available. |
Thanks for testing! |
RFC as I don't want this to be merged before the bindgen PR (#419); this will need to be rebased.
A test image is available and can be used as follow:
Labels can be set for testing, eg:
A few questions / remarks:
nf_conn.@igsilya can you help testing to make sure this looks good? Thanks!
Close #407.