[Mellanox] Add a configuration to delay start xcvrd for fast-reboot#5643
[Mellanox] Add a configuration to delay start xcvrd for fast-reboot#5643liat-grozovik merged 1 commit intosonic-net:masterfrom
Conversation
|
retest vs please |
| "skip_ledd": true, | ||
| "skip_fancontrol": true | ||
| "skip_fancontrol": true, | ||
| "delay_xcvrd": true |
There was a problem hiding this comment.
Should this be configurable, or should we simply always delay the xcvrd start?
There was a problem hiding this comment.
IMO we should simply always delay, like it is done with counters.
There was a problem hiding this comment.
I heard that there are vendors that depend on the xcvrd to finish their fast-reboot, I am not so sure about that, I was not in that loop before. But if there is no vendor need that, we can simply always delay it. @jleveque any suggestion?
There was a problem hiding this comment.
@jleveque As we discussed last time, besides of just monitoring xcvrd dynamically tunes transceivers on some platforms. Thus, if not, on fast-reboot there might be port operationally down longer time.
Guohan can correct me if this is the case.
There was a problem hiding this comment.
@lguohan: My question is do we need to make this customizable, or should we simply delay it on all platforms? If delaying on all platforms will cause no harm with fast-reboot, then I think it should be delayed on all platforms to make things more universal and also to prevent adding extra complexity.
There was a problem hiding this comment.
@jleveque In our platforms xcvrd is used to program pre-emphasis and thus a delay would might even lead to ports not coming up. Moreover ins 400G platforms complex initializations need to be done to bring port up and delaying xcvrd there would affect fastboot. Hence we prefer having a platform knob and by default there should be no delay.
There was a problem hiding this comment.
Thanks, @dgsudharsan. Then we will proceed with this approach, adding a new delay_xcvrd parameter.
There was a problem hiding this comment.
does this means we can sign off it and have it taken to 201911 as well?
|
@lguohan could you please refer to the open issue above? |
|
|
||
| {% if not skip_xcvrd %} | ||
| [program:xcvrd] | ||
| {% if delay_xcvrd %} |
There was a problem hiding this comment.
what if delay_xrcvd is not defined? will this throw exception?
There was a problem hiding this comment.
Just did a quick check. It won't throw exception if delay_xcvrd is not defined.
|
@lguohan how can we proceed with this PR? we would like to have it in 201911 to ensure fastboot req are met. |
- Why I did it
Delay start xcvrd on mellanox platform to save CPU cost during fast-reboot. With this change, fast-reboot saves about 8~9 seconds.
- How I did it
Add a sleep before xcvrd start in pmon supervisor.conf
- How to verify it
Manual test
- Which release branch to backport (provide reason below if selected)
- Description for the changelog
- A picture of a cute animal (not mandatory but encouraged)