bootstrap-test.sh

#!/bin/bash
# Watches tor's ControlPort for bootstrap progress. If it stalls past
# STALL_TIMEOUT without reaching 100%, it disables the currently-failing
# Bridge line in torrc and restarts tor on the next one, up to MAX_TRIES.
set -uo pipefail

CONTROL_PORT=9051
STALL_TIMEOUT=150    # seconds to wait per bridge attempt (snowflake's WebRTC/
                     # broker handshake routinely needs >90s on a cold start)
MAX_TRIES=3
TORRC=/etc/tor/torrc
COOKIE=/var/lib/tor/control_auth_cookie

authenticate_and_query() {
    # Uses the cookie file tor writes on startup; returns raw GETINFO output
    local cookie_hex
    cookie_hex=$(xxd -p "$COOKIE" | tr -d '\n')
    printf 'AUTHENTICATE %s\r\nGETINFO status/bootstrap-phase\r\nQUIT\r\n' "$cookie_hex" \
        | timeout 5 nc 127.0.0.1 "$CONTROL_PORT" 2>/dev/null
}

wait_for_bootstrap() {
    local waited=0
    while (( waited < STALL_TIMEOUT )); do
        [[ -r "$COOKIE" ]] || { sleep 1; ((waited++)); continue; }
        local out
        out=$(authenticate_and_query || true)
        if echo "$out" | grep -q 'PROGRESS=100'; then
            return 0
        fi
        if echo "$out" | grep -qi 'BOOTSTRAP'; then
            local progress
            progress=$(echo "$out" | grep -oP 'PROGRESS=\K[0-9]+' || echo "?")
            echo "[bootstrap-test] progress: ${progress}% (${waited}s/${STALL_TIMEOUT}s)"
        fi
        sleep 3
        ((waited+=3))
    done
    return 1
}

disable_current_bridge_line() {
    # Comments out the first still-active `Bridge ` line so the next
    # restart falls through to the next configured transport.
    local line
    line=$(grep -n '^Bridge ' "$TORRC" | head -1 | cut -d: -f1)
    if [[ -n "$line" ]]; then
        sed -i "${line}s/^Bridge /# FAILED-Bridge /" "$TORRC"
        echo "[bootstrap-test] disabled bridge on line ${line}, will retry with next transport"
    fi
}

for attempt in $(seq 1 "$MAX_TRIES"); do
    echo "[bootstrap-test] attempt ${attempt}/${MAX_TRIES}"
    if wait_for_bootstrap; then
        echo "[bootstrap-test] SUCCESS: bootstrapped to 100%"
        exit 0
    fi
    echo "[bootstrap-test] stalled after ${STALL_TIMEOUT}s, rotating bridge and restarting tor"
    disable_current_bridge_line
    pkill -HUP -x tor 2>/dev/null || true   # ask tor to reload torrc
    sleep 3
    pkill -x tor 2>/dev/null || true
    tor -f "$TORRC" &
done

echo "[bootstrap-test] FAILURE: exhausted all configured bridges"
exit 1