feat(sentinèlla): p2p topology with DNS peer discovery

- Replace central sentinel with watcher: each node polls peers discovered
  via a single DNS name with multiple A records (e.g. peers.sentinella.com)
- Auto-detect own IPs via hostname -I; SELF env var available as optional
  override for NAT/floating-IP setups
- Fix Basic Auth bug in router.sh: compare tok against AUTH_TOKENS instead
  of unset $USER/$PASS
- Rename sentinel binary to watcher; drop unused shellplot dep
- Add inetutils to watcher runtime deps for hostname -I
- Update NixOS module: replace sentinel options with watcher p2p options
  (peersDns, self, peersPort, peersScheme, pollingIntervalSec)
- Add sentinèlla test suite: probe-status-empty, probe-disk, watcher-state-file
This commit is contained in:
2026-04-26 21:54:07 +00:00
parent 249d027f6b
commit 6035397e9b
15 changed files with 687 additions and 134 deletions

View File

@@ -1,3 +1,4 @@
{ system, inputs, self, pkgs }:
{ system, inputs, self, pkgs }:
(import ./migrator { inherit system inputs self pkgs; }) //
(import ./hemar { inherit system inputs self pkgs; })
(import ./hemar { inherit system inputs self pkgs; }) //
(import (./. + "/sentinèlla") { inherit system inputs self pkgs; })

View File

@@ -0,0 +1,44 @@
{ inputs, self, pkgs, system, ... }: let
lib = inputs.nixpkgs.lib;
mkTestDrv = name: type:
if type == "directory" then
pkgs.runCommand "test-${name}" {} ''
if ! [ -f ${./test + "/${name}" + /run.sh} ]; then
echo "no run.sh in test/${name}"
exit 1
fi
mkdir -p "$out"
cp -r ${./test + "/${name}"}/* "$out/"
chmod +x "$out/run.sh"
''
else if lib.hasSuffix ".sh" name then
pkgs.runCommand "test-${lib.removeSuffix ".sh" name}" {} ''
mkdir -p "$out"
install -Dm755 ${./test + "/${name}"} "$out/run.sh"
''
else
null;
testDir = builtins.readDir ./test;
testDrvs =
lib.mapAttrs' (n: v:
lib.nameValuePair (lib.removeSuffix ".sh" n) v
) (lib.filterAttrs (_: v: v != null)
(lib.mapAttrs (n: t: mkTestDrv n t) testDir));
sentinella = self.packages.${system}."sentinèlla";
mkTest = testName: testDrv: pkgs.runCommand "sentinella-test-${testName}"
{
nativeBuildInputs = [ pkgs.coreutils pkgs.gnugrep pkgs.gnused ];
buildInputs = [ sentinella pkgs.curl pkgs.jq pkgs.socat ];
} ''
${builtins.readFile self.legacyPackages.${system}.helpers.posix-shell.log}
export HECTIC_LOG=trace
test=${testDrv}
${builtins.readFile ./launch.sh}
mkdir -p "$out"
'';
in lib.mapAttrs (name: drv: mkTest name drv) testDrvs

View File

@@ -0,0 +1,45 @@
#!/bin/dash
# launch.sh — sets up helpers and runs the test pointed to by $test
# assert_eq(label, got, expected)
assert_eq() {
label=${1:?}
got=${2:?}
expected=${3:?}
if [ "$got" != "$expected" ]; then
log error "FAIL: $label"
log error " expected: $WHITE$expected"
log error " got: $WHITE$got"
exit 1
fi
log info "PASS: $label"
}
# assert_file_contains(label, file, pattern)
assert_file_contains() {
label=${1:?}
file=${2:?}
pattern=${3:?}
if ! grep -q "$pattern" "$file" 2>/dev/null; then
log error "FAIL: $label — pattern '$pattern' not found in $file"
exit 1
fi
log info "PASS: $label"
}
# wait_for_file(file, timeout_sec)
wait_for_file() {
file=${1:?}
timeout=${2:-10}
i=0
while [ $i -lt "$timeout" ]; do
[ -f "$file" ] && return 0
sleep 1
i=$((i+1))
done
log error "timeout waiting for file: $file"
exit 1
}
# run the actual test
. "$test/run.sh"

View File

@@ -0,0 +1,35 @@
#!/bin/dash
# Test: probe GET /disk returns JSON with at least one volume entry
log notice "test case: ${WHITE}probe GET /disk returns volume data"
PORT=15989
export PORT URLS="" VOLUMES="/"
probe &
probe_pid=$!
trap 'kill $probe_pid 2>/dev/null; exit' EXIT INT HUP
sleep 2
response=$(curl -sS --max-time 5 "http://127.0.0.1:${PORT}/disk")
log info "response: $WHITE$response"
count=$(printf '%s' "$response" | jq -r '.volumes | length')
log info "volume count: $WHITE$count"
if [ "$count" -lt 1 ]; then
log error "expected at least 1 volume, got $count"
exit 1
fi
log info "PASS: at least one volume returned"
# each entry must have a mount field
mount=$(printf '%s' "$response" | jq -r '.volumes[0].mount')
if [ -z "$mount" ] || [ "$mount" = "null" ]; then
log error "volumes[0].mount is missing or null"
exit 1
fi
log info "PASS: volumes[0].mount = $mount"
log notice "test passed"

View File

@@ -0,0 +1,27 @@
#!/bin/dash
# Test: probe responds on GET /status with valid JSON when URLS is empty
log notice "test case: ${WHITE}probe GET /status returns JSON with empty checks"
# start probe on a free port
PORT=15988
export PORT URLS="" VOLUMES="/"
probe &
probe_pid=$!
trap 'kill $probe_pid 2>/dev/null; exit' EXIT INT HUP
# wait for probe to be ready
sleep 2
response=$(curl -sS --max-time 5 "http://127.0.0.1:${PORT}/status")
log info "response: $WHITE$response"
# must be valid JSON with summary.total == 0
total=$(printf '%s' "$response" | jq -r '.summary.total')
assert_eq "summary.total is 0 when URLS empty" "$total" "0"
ok=$(printf '%s' "$response" | jq -r '.summary.ok')
assert_eq "summary.ok is 0 when URLS empty" "$ok" "0"
log notice "test passed"

View File

@@ -0,0 +1,75 @@
#!/bin/dash
# Test: watcher writes a state file after polling a peer
#
# Setup:
# - Start a probe on 127.0.0.1:15990
# - Stub getent to resolve peers.test -> 127.0.0.1 (the probe) and 10.0.0.1 (fake peer)
# - Stub hostname to return 10.0.0.1 as the local IP so 10.0.0.1 is excluded
# and 127.0.0.1 (the real probe) is kept as a peer
# - Assert a state file appears in STATE_DIR within 15s
log notice "test case: ${WHITE}watcher writes state file after first successful poll"
PORT=15990
export PORT URLS="" VOLUMES="/"
probe &
probe_pid=$!
trap 'kill "$probe_pid" 2>/dev/null; kill "$watcher_pid" 2>/dev/null; rm -rf "$stub_dir" "$state_dir"' EXIT INT HUP
sleep 2
# Create stubs directory
stub_dir=$(mktemp -d)
# Stub getent: returns two IPs for peers.test
cat >"${stub_dir}/getent" <<'EOF'
#!/bin/sh
if [ "$1" = "hosts" ] && [ "$2" = "peers.test" ]; then
printf '127.0.0.1 peers.test\n'
printf '10.0.0.1 peers.test\n'
else
/usr/bin/getent "$@"
fi
EOF
chmod +x "${stub_dir}/getent"
# Stub hostname: -I returns 10.0.0.1 so watcher excludes it and keeps 127.0.0.1
cat >"${stub_dir}/hostname" <<'EOF'
#!/bin/sh
case "$1" in
-I) printf '10.0.0.1\n' ;;
*) /bin/hostname "$@" ;;
esac
EOF
chmod +x "${stub_dir}/hostname"
state_dir=$(mktemp -d)
export PEERS_DNS="peers.test"
export PEERS_PORT="$PORT"
export PEERS_SCHEME="http"
export TG_TOKEN="test-token"
export TG_CHAT_ID="test-chat"
export STATE_DIR="$state_dir"
export POLLING_INTERVAL_SEC="1"
export SPAM="0"
unset SELF # ensure auto-detection is used
PATH="${stub_dir}:${PATH}" watcher &
watcher_pid=$!
log info "waiting for state file in $state_dir ..."
peer_url="http://127.0.0.1:${PORT}"
state_file="${state_dir}/$(printf '%s' "$peer_url" | cksum | awk '{print $1}').state"
wait_for_file "$state_file" 15
state=$(cat "$state_file")
log info "state file content: $WHITE$state"
case "$state" in
up:*|down:*) log info "PASS: state file has expected format" ;;
*) log error "unexpected state file content: $state"; exit 1 ;;
esac
log notice "test passed"