PSA Update clone operations to use clone_session_id #515
Conversation
Some text in the clone section describes what the control plane must configure for each clone session.
|
@samar-abdi As for other issue updating PSA on multicast replication, this PR contains proposed details for PRE configuration of clone sessions. |
| recirculated at the end of egress processing, as a normal unicast | ||
| packet would at the end of egress processing. | ||
|
|
||
| Since it is an expected common case to clone packets to the CPU, every |
| clone sessions that the control plane may configure in the PRE. For | ||
| each clone session, the control plane may configure the `egress_port` | ||
| and `class_of_service` values that should be associated with packets | ||
| cloned using that session. |
There was a problem hiding this comment.
Thanks @jafingerhut
Are we mandating that these are the only two configurations that a PSA PRE will support? What about truncation?
There was a problem hiding this comment.
I am open to adding a truncation configuration value per clone session, if folks want it. @hanw @vgurevich thoughts on that?
There was a problem hiding this comment.
In general, I think of PSA as saying "at least this much, but an implementation is free to do more". Of course, taking advantage of more reduces portability, I know. If you know of other configuration options besides the ones listed that you want to discuss, this is a good place to ask about them.
There was a problem hiding this comment.
OK, but we will have packet length in the P4Runtime API regardless, with the assumption that either switch SW or ASIC will support it.
There was a problem hiding this comment.
We can have it in P4 Runtime if there is interest, but that kind of line-rate packet operation (I'm talking about truncation) has to be supported in the hardware.
There was a problem hiding this comment.
I saw this for the truncate operation:
For all copies of the packet made at the end of ingress processing,
truncate the payload to be at most the specified number of bytes.
Specifying 0 is legal, and causes only packet headers to be sent, with
no payload.
I suppose this applies to clones and multicast replicas.
There was a problem hiding this comment.
We may want a single place that summarizes how to truncate operations are supposed to work. For example, here is a little note stuck in Section 6.2 right now "Note: Truncation is not currently supported for cloned packets. The functionality of a truncated cloned packet can be replaced with sending a digest." There are also notes about truncation not needing to be supported for resubmit and recirculate packets.
I suspect it may be a good idea to add truncate options/parameters in the clone session configuration state. I will make an issue to discuss at next PSA WG meeting, so people have a chance to comment.
If there are no known issues with this PR other than whether truncate options should be included in the per-clone-session configuration state, I suggest we merge this in, and add that with a later PR after discussion.
There was a problem hiding this comment.
@samar-abdi Looks like the current PSA draft contradicts itself in different sections on whether truncation must be supported for cloned packets, or not. I have created this issue that will hopefully be discussed at next PSA WG, and resolved soon: #519
There was a problem hiding this comment.
@jafingerhut I am ok with merging this PR and dealing with truncation separately. Thanks!
|
@cc10512 OK to merge these changes in, do you think? There is now a separate issue specifically for the truncation behavior, for clones and all other kinds of packets, so that topic should be covered by the other issue. |
|
Yup, I didn't change my mind :) |
|
OK, merged it in, then. |
Some text in the clone section describes what the control plane must
configure for each clone session.