[xcvrd] Bootstrap the CMIS state machine for ports that miss their PORT_SET

[tahmed-dev]

Why I did it

A port only enters the CMIS state machine once a PORT_SET event assigns it an initial state through force_cmis_reinit(). get_cmis_state_from_state_db() reports an absent cmis_state as UNKNOWN, and process_single_lport() returns immediately on UNKNOWN:

state = common.get_cmis_state_from_state_db(...)
if state in CMIS_TERMINAL_STATES or state == CMIS_STATE_UNKNOWN:
    ...
    return

So UNKNOWN is treated as "already finished" rather than "not yet started". A port whose PORT_SET was never delivered is skipped on every iteration and left completely unmanaged: no CMIS state is ever written, the transceiver is never queried, and nothing is logged. There is no recovery path short of restarting xcvrd.

This was observed on a port created by a dynamic port breakout. It had no cmis_state field at all, while every other port on the switch reported READY. The transceiver stayed on the application it powered up with, and the link never came up. Restarting xcvrd fixed it, since the port was then present when the port map was built.

Reproduced against the current code with a fully configured port that never received a PORT_SET:

Scenario	cmis_state	Transceiver queried
PORT_SET missed	never written	never
PORT_SET delivered (control)	INSERTED	yes
PORT_SET without index	INSERTED, stuck	no, and KeyError('index')

How I did it

Bootstrap the state machine in process_single_lport() when a port has no CMIS state but a complete configuration, instead of skipping it. The port then follows the normal INSERTED path. This closes the window regardless of why the event was lost, and is a no-op for ports that already have a state.

The completeness check is factored into is_port_config_complete(), which also replaces the existing inline validity check so both use the same rule. Ports without a usable configuration, for example a port whose physical index is not known yet, are still skipped exactly as before.

Also always record a physical index in port_dict. on_port_update_event() only stored it when pport >= 0, while port_event_helper defaults a missing index field to -1. A port could therefore be admitted to port_dict without an index key, and is_decomm_pending() and is_decomm_lead_lport() index into it directly, outside the try block in process_cmis_state_machine(). This part is defensive, as I did not establish a reachable path to that KeyError in the current flow.

How to verify it

cd sonic-xcvrd && python3 -m pytest tests/test_xcvrd.py -k "is_port_config_complete or always_records_index or bootstraps_missing_cmis_state or skips_incomplete_config"

Added:

• test_CmisManagerTask_is_port_config_complete, a matrix over missing, invalid and non numeric index, speed, lanes and subport.
• test_CmisManagerTask_on_port_update_event_always_records_index, covering a PORT_SET with no index followed by one with a valid index.
• test_CmisManagerTask_process_single_lport_bootstraps_missing_cmis_state, which reproduces the reported case: a fully configured port with no CMIS state is now bootstrapped and its transceiver queried.
• test_CmisManagerTask_process_single_lport_skips_incomplete_config, which confirms a port without a usable configuration is still skipped and its transceiver never touched.

On hardware, create a port through a dynamic port breakout and confirm cmis_state appears in TRANSCEIVER_STATUS without restarting xcvrd.

Description for the changelog

Bootstrap the CMIS state machine for ports that never received a PORT_SET event, so they are not left unmanaged until xcvrd restarts.

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[mssonicbld]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Copilot]

Pull request overview

Bootstraps the CMIS state machine in sonic-xcvrd for logical ports that never received a PORT_SET, preventing them from remaining permanently unmanaged (no CMIS state written, no transceiver probing) until xcvrd restarts.

Changes:

• Add is_port_config_complete() and use it to consistently gate CMIS processing and (when needed) bootstrap ports with missing CMIS state to INSERTED.
• Ensure port_dict[lport]['index'] is always present (default -1) to avoid KeyError in decommission helpers.
• Add unit tests covering config completeness, always-record-index behavior, bootstrapping behavior, and skip-on-incomplete-config behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
sonic-xcvrd/xcvrd/cmis/cmis_manager_task.py	Adds config completeness helper, bootstraps missing CMIS state, and always initializes index in port_dict.
sonic-xcvrd/tests/test_xcvrd.py	Adds targeted unit tests to validate the new bootstrap and config completeness behavior.

Comments suppressed due to low confidence (1)

sonic-xcvrd/xcvrd/cmis/cmis_manager_task.py:1308

• process_single_lport() calls int() on index/speed (and reads lanes) before checking is_port_config_complete(info). If any of these fields are malformed (e.g. 'index': 'N/A'), this will raise and crash the polling loop instead of safely skipping the port.

        pport = int(info.get('index', "-1"))
        speed = int(info.get('speed', "0"))
        lanes = info.get('lanes', "").strip()
        subport = info.get('subport', 0)
        if not self.is_port_config_complete(info):

[prgeor]

@tahmed-dev are you saying the port update event was missed by Xcvrd during port breakout? There are important port attributes like lanes, speed, subport values that are obtained when new port is created(that how self.port_dict[]) without that even if CMIS_STATE_INSERTED is forced, the port cannot be reliably initialized(its working in one case doesn't invalidates the requirement). If Xcvrd is missing the event, how do we know that process_single_lport() is now operating on old port info versus newly created breakout port information?

[mssonicbld]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

Suppressed comments (1)

sonic-xcvrd/xcvrd/cmis/cmis_manager_task.py:1312

• process_single_lport() converts index/speed to int (and strips lanes) before calling is_port_config_complete(). That defeats the purpose of centralizing the validity check: if any of those fields are missing/invalid (e.g., non-numeric index/speed), ValueError/AttributeError can be raised before the helper gets a chance to return False and skip the port.

        pport = int(info.get('index', "-1"))
        speed = int(info.get('speed', "0"))
        lanes = info.get('lanes', "").strip()
        subport = info.get('subport', 0)
        if not self.is_port_config_complete(info):