[dhcpmon] Listen for membership updates

[Xichen96]

Description of PR

Summary:

dhcpmon currently reads VLAN_MEMBER and PORTCHANNEL_MEMBER only at
startup. Adding, removing, or reassigning a member leaves packet attribution
and counter inventory stale until the process is restarted.

This change registers both SWSS subscriber file descriptors directly with the
existing main libevent loop. Notifications mark topology refresh pending. On
the next health tick, packet handlers are suspended, the transactional
membership/counter reconciliation runs on the main thread, handlers resume,
and a successfully applied mixed-state health window is skipped. A failed
reconciliation rolls back and continues health checks against the last valid
topology. The process PID remains unchanged.

This is a stacked draft:

• [dhcpmon] Add packet event quiescing #91 provides safe packet-event
  suspension/resumption.
• [dhcpmon] Reconcile runtime membership state #92 provides transactional
  membership/counter reconciliation.
• Only the final listener commit belongs to this PR; the displayed diff will
  shrink as those prerequisites merge.

Work item tracking

• Microsoft ADO (number only): 38615412

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Documentation update
• Test improvement

Approach

What is the motivation for this PR?

Fungible F2 validation includes PortChannel/member add, delete, and
reassignment scenarios. Requiring a service restart loses counters and creates
an avoidable monitoring gap.

How did you do it?

• Create SubscriberStateTable instances for VLAN_MEMBER and
  PORTCHANNEL_MEMBER.
• Register their selectable file descriptors as persistent events on the main
  libevent base.
• Drain notification batches in the main-thread callbacks and coalesce them
  into one pending refresh.
• Recover failed subscriber FDs by suspending/removing tagged config events,
  rebuilding subscribers, and requiring a full reconciliation.
• Suspend packet callbacks, invoke the main-thread reconciliation API, and
  resume packet callbacks with no process restart.
• Defer refresh while counter clear is active.
• Skip a health window after topology state is successfully applied, so
  old/new counters are never compared together. A rolled-back failure keeps
  refresh pending while health checks continue on the prior topology.
• Reconcile once synchronously after subscriptions are registered and before
  socket event loops start, closing the startup notification gap.

How did you verify/test it?

Pending this stacked PR's Azure CI and master hardware validation.

Planned validation keeps the same dhcpmon PID while:

• adding/removing a physical PortChannel member;
• adding/removing a PortChannel from the monitored VLAN;
• deleting/recreating the same PortChannel with a new ifindex;
• adding a previously unseen PortChannel and members;
• running counter clear during membership churn.

Any platform specific information?

No.

Back port request

None. This repair targets master only.

Tested branch

Pending.

Test result

Pending this PR's Azure CI and master hardware validation.

Documentation

Not applicable; no command or counter schema changes.

[mssonicbld]

/azp run

[azure-pipelines]

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

[azure-pipelines]

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

[mssonicbld]

/azp run

[azure-pipelines]

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

[Copilot]

Pull request overview

This PR updates dhcpmon to react to runtime CONFIG_DB membership changes (VLAN_MEMBER / PORTCHANNEL_MEMBER) without requiring a process restart, keeping interface attribution and counter inventory consistent as topology changes.

Changes:

• Registers CONFIG_DB membership subscribers directly on the main libevent loop and coalesces notifications into a pending topology refresh.
• Adds packet-handler quiesce/resume support and bounds packet processing per callback to avoid indefinite drain under sustained traffic.
• Introduces a main-thread topology reconciliation path that refreshes membership mappings and reconciles cache/DB counters transactionally.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/sock_mgr.h	Adds APIs for suspending/resuming packet handlers and pruning cache counters by valid interface set.
src/sock_mgr.cpp	Implements packet-handler suspend/resume, adds a keepalive event per socket event base, and adds cache-counter pruning.
src/packet_handler.cpp	Bounds packet processing per callback to a fixed batch size.
src/event_mgr.h	Adds event suspend/resume APIs by tag.
src/event_mgr.cpp	Implements tagged suspend/resume and fixes fd logging during delete/free.
src/dhcp_mon.h	Adds dhcp_mon_reconcile_topology() API contract for main-thread reconciliation.
src/dhcp_mon.cpp	Wires CONFIG_DB subscriber FDs into the main loop and runs reconcile on the health tick with packet quiescing.
src/dhcp_devman.h	Declares a new transactional mapping refresh API.
src/dhcp_devman.cpp	Refactors membership mapping rebuild into dhcp_devman_refresh_mappings() and clears reverse maps on shutdown.

Comments suppressed due to low confidence (1)

src/dhcp_mon.cpp:887

• dhcp_mon_start() treats any non-zero return from dhcp_mon_reconcile_topology() as fatal. But that function returns 1 to indicate "counter clear active; defer reconciliation" (not an error). In that case, startup should continue with topology_refresh_pending left true so the next health tick can retry, rather than tearing down the main events.

    topology_refresh_pending = true;
    sock_mgr_suspend_packet_handler();
    if (dhcp_mon_reconcile_topology() != 0 || sock_mgr_resume_packet_handler() < 0) {
        goto unregister_main_events;
    }

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

[mssonicbld]

/azp run

[azure-pipelines]

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

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

src/dhcp_mon.cpp:480

• If resuming packet handlers fails, the callback returns while packet handlers remain suspended, leaving dhcpmon running but no longer processing packets. Consider failing fast (so a supervisor restarts the daemon) or implementing a recovery path that re-registers packet events before continuing.

        if (sock_mgr_resume_packet_handler() < 0) {
            syslog(LOG_ALERT, "Failed to resume packet handlers after topology refresh");
            return;
        }

src/event_mgr.cpp:103

• resume_all_events() re-adds events with a NULL timeout, which will drop any previously configured timeout for timer-based events in this tag (e.g., EV_PERSIST timers added via add_event(..., &timeval, tag)). Either store/restore the original timeout per event, or document/guard that suspend/resume is only valid for fd-driven events (no timeouts).

int event_mgr::resume_all_events(const std::string &tag)
{
    for (const auto &event : this->event_map[tag]) {
        if (event_add(event, NULL) < 0) {
            this->suspend_all_events(tag);

[Copilot]

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

src/dhcp_mon.cpp:892

• dhcp_mon_start() treats any non-zero return from dhcp_mon_reconcile_topology() as fatal. However dhcp_mon_reconcile_topology() explicitly returns 1 when counter clear is active (defer). On that path, startup will abort even though the daemon could continue running with the last-known topology and retry reconciliation on later health ticks.

    topology_refresh_pending = true;
    sock_mgr_suspend_packet_handler();
    if (dhcp_mon_reconcile_topology() != 0 || sock_mgr_resume_packet_handler() < 0) {
        goto unregister_main_events;
    }
    topology_refresh_pending = false;

[mssonicbld]

/azp run

[azure-pipelines]

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

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

[mssonicbld]

/azp run

[azure-pipelines]

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

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[mssonicbld]

/azp run

[azure-pipelines]

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

[mssonicbld]

/azp run

[azure-pipelines]

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

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

[Xichen96]

Closing after architecture review. Full CONFIG_DB listening is deferred until dhcpmon is first CONFIG_DB-driven for a single VLAN and then multi-VLAN. This listener draft skips those prerequisites and will not be merged as-is.