Skip to content

Want swadm subcommand for simulating physically reseating a transceiver #351

Description

@bnaecker

In some situations, we'd like to "fully reset" a transceiver. As an example, the Tofino software sometimes fails to fully power up a module. We haven't seen this much for a while, but we did just have another, similar situation at a customer site. In this case, we were also running Omicron's rack-setup process, which means the sled-agent is continually attempting to construct the links we've configured.

This complicates things. swadm sp xcvr reset will fail if the module is in automatic mode, but setting it to manual with swam sp xcvr set-management-mode manual will fail if there are extant links. One could instead use the lower-level xcvradm tool to reset the module, but that's out of band, and the Tofino doesn't reliably restart its module state machines.

We should add a subcommand to swadm and possibly a new Dendrite API, for fully effecting this kind of simulated reseat. It's a little tricky in light of the sled-agent reconciliation mentioned above, but this would essentially do:

  • probably set some bit to prevent API operations on the module
  • store the parameters for all the links on the port
  • delete all the links
  • set the module into management mode
  • reset the module
  • set the module into auto mode
  • recreate all the links
  • unset the bit from the top

I'm not convinced this could be done reliably, but it's worth pursuing. We often have physical access

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions