fix: bound LocalDNS restart budget so failures terminate deterministically - #9439

Open

Saewon Kwak (saewoni) wants to merge 28 commits into

mainfrom

fix/localdns-pod-service-recovery

Commits on Sep 17, 2026

fix: harden LocalDNS cleanup and lifecycle validation
Saewon Kwak (saewoni)
committed
b07dd85
fix: bound LocalDNS restart budget so failures terminate deterministically
Saewon Kwak (saewoni)
committed
fc320f6
docs: note that the localdns restart budget is shared with CSE provisioning

Saewon Kwak (saewoni)
and
Claude (claude)
committed
76c628b
fix: clear localdns StartLimit budget between provisioning restarts

Saewon Kwak (saewoni)
and
Claude (claude)
committed
fe3c99a
test(e2e): assert the LocalDNS restart budget bounds every failure mode

Saewon Kwak (saewoni)
and
Claude (claude)
committed
a071cae
test(e2e): explain the VHD-provenance failure mode in the directive assertion

Saewon Kwak (saewoni)
and
Claude (claude)
committed
ff80671
fix: restore the 100-attempt provisioning retry budget for localdns

Saewon Kwak (saewoni)
and
Claude (claude)
committed
9a4e3a0
fix: bound every systemd call in the localdns provisioning retry loop

Saewon Kwak (saewoni)
and
Claude (claude)
committed
2ee5a55
test(e2e): clear the start budget before each kill, and fix the fault harness copy

Saewon Kwak (saewoni)
and
Claude (claude)
committed
c1133e3
fix(e2e): stop leaking a Restart=no drop-in from the lifecycle test

Saewon Kwak (saewoni)
and
Claude (claude)
committed
4954e95
fix(e2e): retry the LocalDNS restore in lifecycle cleanup

Saewon Kwak (saewoni)
and
Claude (claude)
committed
ec5f390
test(e2e): address review on the restart-budget validation

Saewon Kwak (saewoni)
and
Claude (claude)
committed
fa30d82
test(e2e): harden the 70-localdns.conf assertion against a root-only dir

Saewon Kwak (saewoni)
and
Claude (claude)
committed
5017d1c
test(e2e): run the slow fault modes on shortened clocks so the matrix fits

Saewon Kwak (saewoni)
and
Claude (claude)
committed
fe19732
test(e2e): assert the budget's sizing, not just that it terminates

Saewon Kwak (saewoni)
and
Claude (claude)
committed
c0f7541

Navigation Menu

fix: bound LocalDNS restart budget so failures terminate deterministically - #9439

fix: bound LocalDNS restart budget so failures terminate deterministically#9439
Saewon Kwak (saewoni) wants to merge 28 commits into
mainfrom
fix/localdns-pod-service-recovery

Commits on Sep 17, 2026

fix: harden LocalDNS cleanup and lifecycle validation

fix: bound LocalDNS restart budget so failures terminate deterministically

docs: note that the localdns restart budget is shared with CSE provisioning

fix: clear localdns StartLimit budget between provisioning restarts

test(e2e): assert the LocalDNS restart budget bounds every failure mode

test(e2e): explain the VHD-provenance failure mode in the directive assertion

fix: restore the 100-attempt provisioning retry budget for localdns

fix: bound every systemd call in the localdns provisioning retry loop

test(e2e): clear the start budget before each kill, and fix the fault harness copy

fix(e2e): stop leaking a Restart=no drop-in from the lifecycle test

fix(e2e): retry the LocalDNS restore in lifecycle cleanup

test(e2e): address review on the restart-budget validation

test(e2e): harden the 70-localdns.conf assertion against a root-only dir

test(e2e): run the slow fault modes on shortened clocks so the matrix fits

test(e2e): assert the budget's sizing, not just that it terminates

Commits on Sep 18, 2026

fix: reap the watchdog sleep on SIGTERM, and test the invariants properly

fix: pin TimeoutStartSec so the restart-budget margin is distro-independent

test(e2e): fix the hung-start clock, and skip the budget check on older VHDs

fix: pin TimeoutStopFailureMode so the restart margin holds by design

test(e2e): gate the LocalDNS validations on the lane, not the unit under test

fix: restore the provisioning diagnostics the inlined retry loop dropped

Commits on Sep 21, 2026

docs: record what trapping SIGTERM did to a localdns stop, and pin it with a spec

test(e2e): keep the node scripts under the Bastion limit, and bound the burst check

fix: set TimeoutStopFailureMode in the drop-in, where systemd will honour it

test(e2e): address review on the script-size guard and the provisioning loop

test(e2e): replace the size-limit formula with the measurement, and unify the inventory

docs: record the measured cost of a provisioning retry iteration

fix: give TimeoutStopFailureMode its own drop-in rather than delegate.conf