Iteration 3 discovered a race condition: parallel CLI invocations of
change-lite-domain occasionally produced 'no secret in config' because
one process read config.json while another was mid-rewrite (jq -> tmp -> mv).
asyncio.Lock in bot.py only protects bot->bot races; CLI-level races and
bot+CLI mixed races were still possible.
- install.sh: bot_action_dispatch wraps dispatch in flock(1) on
/var/lock/gotelegram-bot-action.lock, 30s timeout, EX_TEMPFAIL on timeout
- common.sh: flock added to critical deps; apt_pkg_for_cmd/dnf_pkg_for_cmd
map flock -> util-linux
- common.sh: check_deps_present includes flock
- version bumped to 2.4.3 (common.sh + bot.py)
- install.sh: new bot_action_dispatch entry point for --action=X --json CLI invocation from the bot/scripts
- install.sh: bot_action_change_template — reuses download_template + deploy_template_to_nginx, updates config.json template_id
- install.sh: bot_action_change_lite_domain — regenerates telemt TOML with new fake-TLS mask domain, restarts telemt
- bot.py: run_bot_action() subprocess wrapper parses JSON result
- bot.py: cb_pro_confirm now performs real change-template when already in pro mode (fresh install still routes to CLI)
- bot.py: cb_lite_domain now performs real change-lite-domain when already in lite mode
- version -> 2.4.2
Bug report: 'статистика не работает' — submenu_stats showed the
'stats module missing' fallback even though gotelegram-stats.service
was active. Root cause: bootstrap.sh FILES[] never included
lib/stats.sh, so on every (re)install the file was absent from
/opt/gotelegram/lib/ on disk. The running collector was sourcing it
from an in-memory bash process from a previous install, but
install.sh's `source stats.sh` at menu entry failed silently and
hit the 'stats_module_missing' branch.
Also: stats_collect silently skipped history writes if
stats_history.csv was deleted post-init (only stats_init created it
once). Now stats_collect auto-recreates the file with header.
- bootstrap.sh: add 'lib/stats.sh' to FILES[]
- lib/stats.sh: stats_collect auto-creates HISTORY_FILE if missing
Tested live on VPS: show_traffic_stats now outputs real counters
(23.1 MB proxy, 10 MB site, 394 KB/s peak).
start_telemt() was a no-op when the service was already active, so
regenerating /etc/telemt/config.toml (e.g. reinstalling with a different
mask domain or switching lite<->pro) left the daemon running with the
OLD in-memory config. This caused the reported "lite-mode key does not
work" issue: after a fresh lite install over an existing pro install,
telemt kept tls_domain=anten-ka.com and dropped SNI=google.com clients
with unknown_sni_action=Drop even though /etc/telemt/config.toml said
tls_domain=google.com.
Fix: start_telemt now uses systemctl restart when the service is
already active, guaranteeing the fresh config is loaded.
- install.sh: fix {{NC}} -> ${NC} color escape on line 265
- bot.py: fix TOML parsing for telemt v3 [access.users] format
- bot.py: fix telemt config section [server].port instead of [config].listen_port
- telemt_config.sh: fix add_secret_to_config() for v3 format
- All frame boxes: correct 54-char width with emoji compensation
- Lite mode links now include ee-prefix + mask_host hex (fake-TLS)
- Added qrencode to ensure_deps() so QR codes work after fresh install
- Centralized link generation in generate_proxy_link()
- Fixed bot.py get_proxy_link() for lite mode ee-prefix
Lines 168-172 in show_template_preview() were missing >&2,
causing stdout leak into subshell capture chain.
This corrupted template_dir variable and broke Stealth mode deployment.
- All UI output (echo, printf) in interactive functions now goes to stderr (>&2)
- Only return values go to stdout for $() capture
- Affected: select_quick_domain, select_port, select_category, select_template, show_template_preview
- Also fixed: log_*, confirm, select_option in common.sh
