recovery(ananke): verify flux suspension during thaw

This commit is contained in:
jenkins 2026-06-18 18:53:44 -03:00
parent bb07f1598f
commit e893af2a55

View File

@ -141,6 +141,8 @@ RECOVERY_FLUX_CRITICAL_KUSTOMIZATIONS="${RECOVERY_FLUX_CRITICAL_KUSTOMIZATIONS:-
RECOVERY_FLUX_OPTIONAL_SNAPSHOT_FILE="${RECOVERY_FLUX_OPTIONAL_SNAPSHOT_FILE:-${HOME}/${STATE_SUBDIR:-.local/share/ananke}/longhorn_unlock_optional_flux.tsv}"
RECOVERY_FLUX_RESTART_KUSTOMIZE_CONTROLLER="${RECOVERY_FLUX_RESTART_KUSTOMIZE_CONTROLLER:-1}"
RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION="${RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION:-1}"
RECOVERY_FLUX_SUSPEND_VERIFY_ATTEMPTS="${RECOVERY_FLUX_SUSPEND_VERIFY_ATTEMPTS:-6}"
RECOVERY_FLUX_SUSPEND_VERIFY_SLEEP_SECONDS="${RECOVERY_FLUX_SUSPEND_VERIFY_SLEEP_SECONDS:-10}"
STARTUP_SERVICE_CHECK_TIMEOUT_SECONDS="${STARTUP_SERVICE_CHECK_TIMEOUT_SECONDS:-10}"
STARTUP_SERVICE_CHECKLIST="${STARTUP_SERVICE_CHECKLIST:-}"
STARTUP_INCLUDE_INGRESS_CHECKS="${STARTUP_INCLUDE_INGRESS_CHECKS:-1}"
@ -1008,6 +1010,61 @@ patch_recovery_optional_flux_suspend() {
done < <(csv_each "${RECOVERY_FLUX_OPTIONAL_KUSTOMIZATIONS}")
}
patch_recovery_optional_flux_suspend_without_snapshot() {
local value="$1"
local name
while IFS= read -r name; do
patch_kustomization_suspend "${name}" "${value}"
done < <(csv_each "${RECOVERY_FLUX_OPTIONAL_KUSTOMIZATIONS}")
}
recovery_flux_unsuspended_list() {
local names=()
local name suspend
if [[ "${RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION}" == "1" || "${RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION}" == "true" ]]; then
names+=("flux-system")
fi
while IFS= read -r name; do
names+=("${name}")
done < <(csv_each "${RECOVERY_FLUX_OPTIONAL_KUSTOMIZATIONS}")
for name in "${names[@]}"; do
if ! kubectl -n flux-system get kustomization "${name}" >/dev/null 2>&1; then
continue
fi
suspend="$(kubectl -n flux-system get kustomization "${name}" -o jsonpath='{.spec.suspend}' 2>/dev/null || true)"
if [[ "${suspend}" != "true" ]]; then
printf '%s\n' "${name}"
fi
done
}
stabilize_recovery_flux_suspend() {
[[ "${EXECUTE}" -eq 1 ]] || return 0
local attempt unsuspended
for attempt in $(seq 1 "${RECOVERY_FLUX_SUSPEND_VERIFY_ATTEMPTS}"); do
if [[ "${RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION}" == "1" || "${RECOVERY_FLUX_SUSPEND_BOOTSTRAP_KUSTOMIZATION}" == "true" ]]; then
patch_kustomization_suspend flux-system true
fi
patch_recovery_optional_flux_suspend_without_snapshot true
sleep "${RECOVERY_FLUX_SUSPEND_VERIFY_SLEEP_SECONDS}"
unsuspended="$(recovery_flux_unsuspended_list | paste -sd, -)"
if [[ -z "${unsuspended}" ]]; then
log "recovery-flux-suspend=verified attempts=${attempt}"
return 0
fi
warn "Flux suspend state was overwritten during recovery thaw; reasserting attempt ${attempt}/${RECOVERY_FLUX_SUSPEND_VERIFY_ATTEMPTS}: ${unsuspended}"
done
unsuspended="$(recovery_flux_unsuspended_list | paste -sd, -)"
if [[ -n "${unsuspended}" ]]; then
warn "Flux suspend state is still not stable after verification attempts: ${unsuspended}"
return 1
fi
}
restore_recovery_optional_flux_suspend() {
[[ -f "${RECOVERY_FLUX_OPTIONAL_SNAPSHOT_FILE}" ]] || return 0
local name suspend
@ -1464,6 +1521,8 @@ resume_deadlock_automation_after_core_recovery() {
patch_recovery_optional_flux_suspend true
restart_kustomize_controller_for_critical_thaw
annotate_flux_kustomizations "${RECOVERY_FLUX_CRITICAL_KUSTOMIZATIONS}" || true
stabilize_recovery_flux_suspend || true
restart_kustomize_controller_for_critical_thaw
mark_checkpoint longhorn_unlock_automation_resumed
}