LLT-6859 adjust UAPI error handler #1618

LukasPukenis · 2025-12-05T11:46:47Z

Problem

Libtelio uses counter based mechanism when handling interface errors. The counter is used to ignore transient WireguardNT errors that happen around power-transitions and usually go away.

Current limitations:

Counter not reset after power-resume
Counter not aware of FeaturePolling::wireguard_polling_period_after_state_change
Hardcoded value of 10

These limitations allow libtelio to produce unnecessary critical failures that require libtelio restart.

Solution

Increase the counter threshold value.

☑️ Definition of Done checklist

Commit history is clean (requirements)
README.md is updated
Functionality is covered by unit or integration tests

mathiaspeters

+1

crates/telio-wg/src/wg.rs

stalowyjez

Looks rather ok, +1

crates/telio-model/src/features.rs

nat-lab/tests/test_events_link_state.py

crates/telio-wg/src/wg.rs

LukasPukenis · 2025-12-19T12:26:11Z

@stalowyjez I've addressed your comments

stalowyjez

lgtm, +1

Jauler · 2026-01-06T09:07:40Z

crates/telio-wg/src/wg.rs

+pub struct TUNHealthMonitor {
+    /// Minimum duration of a continuous UAPI failure period required to
+    /// classify the failure as non-transient
+    sustained_failure_threshold: Duration,


🤔 This is a nitpick, but maybe it is worth storing here just FeatureInterfaceHealth instead of having a copy of the fields?

This code no longer exists

Jauler · 2026-01-06T09:14:34Z

crates/telio-wg/src/wg.rs

+        // than expected, it indicates that libtelio(or device) might have been suspended. Since
+        // WireguardNT produces transient errors around device suspension(just before and right after),
+        // we must filter out those errors or else we can easily exceed sustained_failure_threshold.
+        let suspend_detected =


I am not sure about this part. As it seems like it is likely to be flaky.... 🤔 🤔 🤔 🤔 As while our period is roughly 1s, there are no guarantees that the only "interruption" can be suspension.

Although I do not have a better suggestion now 🫤.

Maybe it is worth having a small design/brainstorm session to see if we can come up to something less sensitive to timing (2s by default)?

As discussed - I just increased the counter value so this is resolved :)

LukasPukenis · 2026-01-07T14:39:37Z

@Jauler @stalowyjez @mathiaspeters we discussed a bit with @Jauler about the current approach and decided to go via simpler approach - increase the counter threshold. The benefits of such approach is that it's very deterministic while being very simple.

This reduces critical errors caused by transient wireguard-nt UAPI failures at the cost of longer detection time when interface is permanently gone. Signed-off-by: Lukas Pukenis <lukas.pukenis@nordsec.com>

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch 2 times, most recently from f9e65d3 to 4b9c0f6 Compare December 5, 2025 11:51

LukasPukenis changed the title ~~Llt 6859 configurable adapter gone error parameters~~ LLT-6859 configurable adapter-gone UAPI error parameters Dec 5, 2025

LukasPukenis temporarily deployed to Internal December 5, 2025 11:52 — with GitHub Actions Inactive

LukasPukenis changed the title ~~LLT-6859 configurable adapter-gone UAPI error parameters~~ LLT-6859 configurable adapter-gone error handler Dec 5, 2025

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 4b9c0f6 to 1c17d77 Compare December 5, 2025 12:09

LukasPukenis temporarily deployed to Internal December 5, 2025 12:09 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 1c17d77 to 2395810 Compare December 5, 2025 12:49

LukasPukenis temporarily deployed to Internal December 5, 2025 12:49 — with GitHub Actions Inactive

LukasPukenis mentioned this pull request Dec 5, 2025

Fix UAPI error tracker for backwards compatibility #1614

Closed

3 tasks

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 2395810 to 8cbe327 Compare December 5, 2025 12:50

LukasPukenis temporarily deployed to Internal December 5, 2025 12:51 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 8cbe327 to f581981 Compare December 5, 2025 14:51

LukasPukenis temporarily deployed to Internal December 5, 2025 14:51 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from f581981 to fa7b39c Compare December 5, 2025 14:54

LukasPukenis temporarily deployed to Internal December 5, 2025 14:55 — with GitHub Actions Inactive

LukasPukenis changed the title ~~LLT-6859 configurable adapter-gone error handler~~ LLT-6859 feature config for UAPI error handler Dec 5, 2025

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from fa7b39c to 5bc731b Compare December 9, 2025 14:20

LukasPukenis temporarily deployed to Internal December 9, 2025 14:21 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 5bc731b to 825b681 Compare December 10, 2025 12:51

LukasPukenis temporarily deployed to Internal December 10, 2025 12:51 — with GitHub Actions Inactive

LukasPukenis changed the title ~~LLT-6859 feature config for UAPI error handler~~ LLT-6859 Dec 10, 2025

LukasPukenis changed the title ~~LLT-6859~~ LLT-6859 configurable UAPI error handler Dec 10, 2025

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 825b681 to 1049dc7 Compare December 10, 2025 13:25

LukasPukenis temporarily deployed to Internal December 10, 2025 13:25 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 1049dc7 to 5e76d91 Compare December 10, 2025 13:44

LukasPukenis temporarily deployed to Internal December 10, 2025 13:44 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 5e76d91 to 6c9662a Compare December 10, 2025 13:56

LukasPukenis temporarily deployed to Internal December 10, 2025 13:56 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 6c9662a to 45fe6a4 Compare December 10, 2025 13:58

LukasPukenis marked this pull request as ready for review December 12, 2025 07:18

LukasPukenis requested a review from a team as a code owner December 12, 2025 07:18

mathiaspeters reviewed Dec 12, 2025

View reviewed changes

crates/telio-wg/src/wg.rs Outdated Show resolved Hide resolved

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from cde43e3 to 853eb12 Compare December 12, 2025 12:40

LukasPukenis temporarily deployed to Internal December 12, 2025 12:40 — with GitHub Actions Inactive

stalowyjez reviewed Dec 18, 2025

View reviewed changes

LukasPukenis temporarily deployed to Internal December 19, 2025 11:06 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 6906a0e to 9d4063f Compare December 19, 2025 12:15

LukasPukenis temporarily deployed to Internal December 19, 2025 12:15 — with GitHub Actions Inactive

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 9d4063f to 3a76c6b Compare December 19, 2025 18:21

LukasPukenis temporarily deployed to Internal December 19, 2025 18:21 — with GitHub Actions Inactive

stalowyjez previously approved these changes Dec 22, 2025

View reviewed changes

LukasPukenis dismissed stalowyjez’s stale review via 6e68b3d December 30, 2025 07:58

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 3a76c6b to 6e68b3d Compare December 30, 2025 07:58

LukasPukenis temporarily deployed to Internal December 30, 2025 07:58 — with GitHub Actions Inactive

Jauler reviewed Jan 6, 2026

View reviewed changes

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 6e68b3d to 0d8e9e5 Compare January 7, 2026 14:36

LukasPukenis temporarily deployed to Internal January 7, 2026 14:36 — with GitHub Actions Inactive

LukasPukenis changed the title ~~LLT-6859 configurable UAPI error handler~~ LLT-6859 adjust UAPI error handler Jan 7, 2026

LukasPukenis requested review from Jauler, mathiaspeters and stalowyjez January 7, 2026 14:39

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from 0d8e9e5 to f79344c Compare January 7, 2026 14:45

LukasPukenis temporarily deployed to Internal January 7, 2026 14:46 — with GitHub Actions Inactive

Increase MAX UAPI failure count

d9db805

This reduces critical errors caused by transient wireguard-nt UAPI failures at the cost of longer detection time when interface is permanently gone. Signed-off-by: Lukas Pukenis <lukas.pukenis@nordsec.com>

LukasPukenis force-pushed the LLT-6859_configurable_adapter_gone_error_parameters branch from f79344c to d9db805 Compare January 7, 2026 14:46

LukasPukenis temporarily deployed to Internal January 7, 2026 14:48 — with GitHub Actions Inactive

LLT-6859 adjust UAPI error handler #1618

Are you sure you want to change the base?

LLT-6859 adjust UAPI error handler #1618

Uh oh!

Conversation

LukasPukenis commented Dec 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Solution

☑️ Definition of Done checklist

Uh oh!

mathiaspeters left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

stalowyjez left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

LukasPukenis commented Dec 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stalowyjez left a comment

Choose a reason for hiding this comment

Uh oh!

Jauler Jan 6, 2026

Choose a reason for hiding this comment

Uh oh!

LukasPukenis Jan 7, 2026

Choose a reason for hiding this comment

Uh oh!

Jauler Jan 6, 2026

Choose a reason for hiding this comment

Uh oh!

LukasPukenis Jan 7, 2026

Choose a reason for hiding this comment

Uh oh!

LukasPukenis commented Jan 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

LukasPukenis commented Dec 5, 2025 •

edited

Loading

LukasPukenis commented Dec 19, 2025 •

edited

Loading