feat: Go backup restore VF switchdev

[rollandf]

No description provided.

[coveralls]

Pull Request Test Coverage Report for Build 19696076708

Details

• 38 of 277 (13.72%) changed or added relevant lines in 1 file are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-1.9%) to 61.635%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
entrypoint/internal/netconfig/netconfig.go	38	277	13.72%

Files with Coverage Reduction	New Missed Lines	%
entrypoint/internal/netconfig/netconfig.go	1	35.49%

Totals
Change from base Build 19695900040:	-1.9%
Covered Lines:	1764
Relevant Lines:	2862

---

💛 - Coveralls

[greptile-apps]

Greptile Overview

Greptile Summary

This PR extends the network configuration backup/restore system to support switchdev mode for VF representors, implementing the Go equivalent of existing bash script functionality.

Key Changes:

• Added Representor struct to track switchdev representor devices (name, MTU, admin state, physical IDs)
• Implemented representor discovery during save phase by scanning /sys/class/net for devices matching pf{port}vf{id} pattern
• Modified VF restore logic to skip rebinding VFs in switchdev mode (deferred until after eswitch mode is set)
• Added representor restoration that finds devices by physical attributes, renames if needed, and restores MTU/state
• Added comprehensive tests for switchdev helper functions and VF rebinding logic

Issues Found:

• Critical: Bind delay is 3 seconds but bash script uses 4 seconds (commit 3ce9ba9), creating inconsistency
• Regex compilation happens in hot loops instead of using package-level compiled regex
• Previous thread noted concerns about bind delay impact on performance

Confidence Score: 3/5

• This PR has critical bind delay inconsistency with bash script that could cause VF initialization failures
• The bind delay mismatch (3s vs 4s) is a logic error that could cause real issues in production since the bash script was specifically updated to use 4s for stability. The regex performance issue is a style concern but won't cause failures
• Pay attention to entrypoint/internal/netconfig/netconfig.go lines 401 and 496 - bind delay must match bash script

Important Files Changed

File Analysis

Filename	Score	Overview
entrypoint/internal/netconfig/netconfig.go	3/5	Adds switchdev representor backup/restore functionality with VF rebinding logic. Issues: bind delay mismatch with bash script (3s vs 4s), regex compiled in hot loops
entrypoint/internal/netconfig/netconfig_test.go	5/5	Adds tests for switchdev helper functions and VF rebinding logic, covering both switchdev and legacy modes

Sequence Diagram

sequenceDiagram
    participant User
    participant NetConfig
    participant Sysfs as /sys/class/net
    participant Netlink
    participant Driver as mlx5_core

    Note over User,Driver: Save Phase (Switchdev Mode)
    User->>NetConfig: Save()
    NetConfig->>NetConfig: discoverSwitchdevRepresentors()
    NetConfig->>Sysfs: Read phys_port_name & phys_switch_id
    NetConfig->>Sysfs: Scan subsystem for representors
    NetConfig->>Netlink: Get representor config (MTU, admin state)
    NetConfig-->>User: Configuration saved

    Note over User,Driver: Restore Phase (Switchdev Mode)
    User->>NetConfig: Restore()
    NetConfig->>NetConfig: restoreVFConfigurations(switchdev)
    NetConfig->>Driver: Unbind VF from driver
    Note over NetConfig: Skip rebind for switchdev (handled later)
    
    NetConfig->>NetConfig: Set eswitch mode to switchdev
    
    NetConfig->>NetConfig: rebindVFsInSwitchdevMode()
    NetConfig->>Driver: Bind VF to driver
    Note over NetConfig: Wait 3s bind delay
    NetConfig->>Netlink: Restore VF MTU & admin state
    
    NetConfig->>NetConfig: restoreRepresentors()
    NetConfig->>Sysfs: Find current representor by phys attributes
    NetConfig->>NetConfig: Rename representor if needed
    NetConfig->>Netlink: Set representor MTU & admin state
    NetConfig-->>User: Restore complete

Loading

[greptile-apps]

_{16 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (2)

1. entrypoint/internal/netconfig/netconfig_test.go, line 102-123 (link)

   syntax: duplicate test case - this exact test already exists at lines 112-123

2. entrypoint/internal/netconfig/netconfig_test.go, line 223-255 (link)

   syntax: duplicate test case - this exact test already exists at lines 240-255

_{4 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

style: Check that the bind delay increase from 1s to 3s doesn't negatively impact restore performance for devices with many VFs

[greptile-apps]

style: Check that this bind delay matches the bash script value and is necessary for switchdev mode stability

[greptile-apps]

_{2 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

style: Regex is compiled on every call in a hot loop (called for every network device in findDeviceRepresentors). Consider compiling once at package level:

Suggested change

	re := regexp.MustCompile(`^pf(\d+)vf(\d+)$`)
	var representorRegex = regexp.MustCompile(`^pf(\d+)vf(\d+)$`)

Then use representorRegex.MatchString(physPortName) instead of recompiling each time

[greptile-apps]

style: Same regex compiled twice - once here and once in isRepresentorPhysPortName:1350. Use a single package-level compiled regex instead

[greptile-apps]

logic: Bind delay is 3s but the bash script uses BIND_DELAY_SEC=4 (see commit 3ce9ba9). This inconsistency could cause issues

Suggested change

	time.Sleep(3 * time.Second) // BIND_DELAY_SEC equivalent
	time.Sleep(4 * time.Second) // BIND_DELAY_SEC equivalent (matches bash script default)

[greptile-apps]

logic: Same issue - bind delay is 3s but bash script uses 4s (see commit 3ce9ba9)

Suggested change

	// Wait for bind delay (matches bash script)
	time.Sleep(4 * time.Second) // BIND_DELAY_SEC equivalent (matches bash script default)