[techsupport] Skip switch-only collectors on Switch-BMC

[william8545]

What I did

generate_dump (the engine behind show techsupport) ran the full switch data-plane collection unconditionally. On a SONiC Switch-BMC — a control-plane-only platform with no ASIC/SDK, no front-panel ports, no FRR/vtysh, no swss/bgp/syncd containers, no STP, and an ARM/aspeed SoC that lacks some sysfs paths and tools — this produced a large number of spurious errors and irrelevant/empty sections in the techsupport bundle.

This change makes generate_dump Switch-BMC aware so show techsupport completes cleanly on a Switch-BMC, while behavior on regular switches is unchanged.

Part of the SONiC Switch-BMC enablement.

How I did it

• Add an IS_SWITCH_BMC flag, set from switch_bmc=1 in platform_env.conf (mirroring the existing IS_SUPERVISOR handling).
• Gate switch-only collectors on it so they self-skip on a Switch-BMC: FRR/BGP/EVPN/BFD, ASIC/SDK health events, front-panel interface/transceiver, gearbox, STP, swss/bgp docker logs, hdparm, the SAI dump, and the port-counter snapshots.
• Keep platform-agnostic collectors: the generic show ip interface L3 summary (incl. management), LLDP, NAT/conntrack, and platform info.
• Harden generic fallbacks so absent/empty paths no longer error:
  • record_missing_path replaces the broken echo > $TARDIR/$f fallback in save_proc/save_sys;
  • save_pstore handles absent/empty/populated without a failing glob;
  • sed_if_exists guards optional secret-scrub files.

How to verify it

On an AST2700-based SONiC Switch-BMC (switch_bmc=1):

1. show techsupport → completes with exit 0, no spurious data-plane errors, and a valid gzip tarball under /var/dump.
2. Confirm the switch-only sections are skipped (no FRR/BGP/STP/transceiver errors) while the platform/LLDP/NAT/management-interface sections are present.

On a regular (non-Switch-BMC) switch the behavior is unchanged: IS_SWITCH_BMC=false, so every collector runs exactly as before, and the empty-port-config path stays fatal.

Validated on a live AST2700 SONiC Switch-BMC: multiple full show techsupport runs (default, --silent, --debug-dump), all exit 0 and gzip-valid; the non-BMC fatal paths confirmed unchanged.

Previous command output (if the output of a command-line utility has changed)

On a Switch-BMC, show techsupport previously emitted many collector failures to the console/log (e.g. FRR/vtysh, STP/stpctl, front-panel interface transceiver, swss/bgp docker-log lookups) and wrote empty/error sections into the bundle.

New command output (if the output of a command-line utility has changed)

On a Switch-BMC, show techsupport now completes with exit 0 and no spurious errors; the inapplicable switch-only sections are omitted, and the platform/LLDP/NAT/management sections are retained.

[mssonicbld]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[oleksandrivantsiv]

@yxieca, could you please review the PR?

[yxieca]

LGTM. Reviewed the full diff — clean Switch-BMC enablement for generate_dump. IS_SWITCH_BMC mirrors the existing IS_SUPERVISOR handling (switch_bmc=1 in platform_env.conf), and the switch-only collectors self-skip on BMC (FRR/BGP/EVPN/BFD, ASIC-SDK health, front-panel interface/transceiver, gearbox, STP, swss/bgp docker logs, hdparm, SAI dump, counter snapshots) while ip-interface/LLDP/NAT/redis/platform are retained. Non-BMC behavior is unchanged and the empty-port-config path stays fatal. Generic fallbacks are hardened (record_missing_path, glob-free save_pstore, sed_if_exists for optional scrub files) with no secret-scrub regression. Validated on live AST2700. Nice work.

[mssonicbld]

The change is not in 202605 yet. @william8545, please manually create the cherry pick PR for branch 202605.
You can ping the release branch owner(github account: vaibhavhd) to approve your cherry pick PR.
If this change is already in 202605, please comment "already in 202605". Thanks!

_{---Powered by SONiC BuildBot}

[mssonicbld]

Cherry-pick PR to msft-202608: Azure/sonic-utilities.msft#388