zuckerberg/nix-config
1
0
Fork 0
Code Issues Pull Requests Actions Activity

Fix VPN check alert limiting to only count failures

Browse Source 
StartLimitBurst counts all starts (including successes), so the timer
was getting blocked after ~15 min. Replace with a JSON counter file
that resets on success and daily, only triggering OnFailure alerts for
the first 3 failures per day.
This commit is contained in:
Zuckerberg
2026-03-05 21:28:39 -08:00
parent e3f78b460c
commit 77415c30fa
1 changed files with 52 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
common/network/pia-vpn/vpn-container.nix
View File

@@ -228,44 +228,73 @@ in
}; };
# Periodic VPN connectivity check — fails if VPN or internet is down, # Periodic VPN connectivity check — fails if VPN or internet is down,
# triggering ntfy alert via the OnFailure drop-in # triggering ntfy alert via the OnFailure drop-in.
# Tracks failures with a counter file so only the first 3 failures per
# day trigger an alert (subsequent failures exit 0 to suppress noise).
systemd.services.pia-vpn-check = { systemd.services.pia-vpn-check = {
description = "Check PIA VPN connectivity"; description = "Check PIA VPN connectivity";
after = [ "pia-vpn-setup.service" ]; after = [ "pia-vpn-setup.service" ];
requires = [ "pia-vpn-setup.service" ]; requires = [ "pia-vpn-setup.service" ];
path = with pkgs; [ wireguard-tools iputils coreutils gawk ]; path = with pkgs; [ wireguard-tools iputils coreutils gawk jq ];
unitConfig = {
StartLimitBurst = 3;
StartLimitIntervalSec = "1d";
};
serviceConfig.Type = "oneshot"; serviceConfig.Type = "oneshot";
script = '' script = ''
set -euo pipefail set -euo pipefail
# Check that WireGuard has a peer with a recent handshake (within 3 minutes) COUNTER_FILE="/var/lib/pia-vpn/check-fail-count.json"
handshake=$(wg show ${cfg.interfaceName} latest-handshakes | awk '{print $2}') MAX_ALERTS=3
if [ -z "$handshake" ] || [ "$handshake" -eq 0 ]; then
echo "No WireGuard handshake recorded" >&2 check_vpn() {
exit 1 # Check that WireGuard has a peer with a recent handshake (within 3 minutes)
fi handshake=$(wg show ${cfg.interfaceName} latest-handshakes | awk '{print $2}')
now=$(date +%s) if [ -z "$handshake" ] || [ "$handshake" -eq 0 ]; then
age=$((now - handshake)) echo "No WireGuard handshake recorded" >&2
if [ "$age" -gt 180 ]; then return 1
echo "WireGuard handshake is stale (''${age}s ago)" >&2 fi
exit 1 now=$(date +%s)
age=$((now - handshake))
if [ "$age" -gt 180 ]; then
echo "WireGuard handshake is stale (''${age}s ago)" >&2
return 1
fi
# Verify internet connectivity through VPN tunnel
if ! ping -c1 -W10 1.1.1.1 >/dev/null 2>&1; then
echo "Cannot reach internet through VPN" >&2
return 1
fi
echo "PIA VPN connectivity OK (handshake ''${age}s ago)"
return 0
}
if check_vpn; then
rm -f "$COUNTER_FILE"
exit 0
fi fi
# Verify internet connectivity through VPN tunnel # Failed — read and update counter (reset if from a previous day)
if ! ping -c1 -W10 1.1.1.1 >/dev/null 2>&1; then today=$(date +%Y-%m-%d)
echo "Cannot reach internet through VPN" >&2 count=0
exit 1 if [ -f "$COUNTER_FILE" ]; then
stored=$(jq -r '.date // ""' "$COUNTER_FILE")
if [ "$stored" = "$today" ]; then
count=$(jq -r '.count // 0' "$COUNTER_FILE")
fi
fi fi
count=$((count + 1))
jq -n --arg date "$today" --argjson count "$count" \
'{"date": $date, "count": $count}' > "$COUNTER_FILE"
echo "PIA VPN connectivity OK (handshake ''${age}s ago)" if [ "$count" -le "$MAX_ALERTS" ]; then
echo "Failure $count/$MAX_ALERTS today — alerting" >&2
exit 1
else
echo "Failure $count today — suppressing alert (already sent $MAX_ALERTS)" >&2
exit 0
fi
''; '';
}; };