Gately v0.6.8.4 released
November 19, 2026
Download v0.6.8.4 · GitHub Release
Security fix
- Notification channel keys could appear in delivery errors (low): Telegram Bot Tokens, Bark Device Keys, ServerChan SendKeys and similar secrets sit in the request URL. When a connection failed, the error message carried the full URL into the Notification module's log and the "Recent deliveries" records, where read-only accounts and API tokens with only notification read permission could see it. Channel keys in error messages are now replaced with
***. v0.6.8.2 and earlier are affected; "Recent deliveries" is kept in memory only and is cleared by the restart after the upgrade; the runtime logs may still hold old entries, so rotate the key of any channel whose delivery has failed before (GHSA-p368-wcp9-rv4q).
Changing configuration no longer locks you out
- Confirm after saving, automatic rollback on timeout: after you save port forwarding, protocol routing, Web services or global settings in the admin UI, the page opens a new connection to confirm the admin UI is still reachable. If it cannot be confirmed (for example you changed the listening port the admin UI is reached through), the configuration from before the change is restored after 90 seconds by default. When confirmation fails a countdown appears and you can roll back immediately; if 3 restarts pass without confirmation, it rolls back as well. "System settings › Access Control" lets you turn this off or adjust the wait (30–600 seconds). A rollback sends a "Configuration changes" notification.
- Access-path warning before saving: the admin UI traces which port forwarding, protocol routing and Web sites your current visit passes through. When a change you are saving would affect this path (changing a port, disabling, deleting, switching to an address this machine cannot listen on, and so on), it explains why first and highlights the affected hop. "System settings › Access Control" shows the current access path.
- Batch changes that take effect together: when several things must change together to work (for example change protocol routing first, then the Web listener), choose "Stage, apply together later". A banner at the top lists the staged rules and listeners; "Apply all" confirms the whole batch and rolls the whole batch back if it fails, and "Discard changes" drops it. Before applying, listeners that cannot start are listed first.
- Command-line rollback:
gately rollbacklists pending changes and the last 15 configuration saves, and you can undo one by number, or use-last,-pendingor-configtogether with-version. While Gately is running it restores and restarts within 2 seconds; when it is not running it rewrites the configuration file directly. - Deleting a configuration file now saves a copy to the config history first, so a deletion can also be undone from the config history.
Notifications: retry and resend after failures
- After a failed push, Gately retries after 10, 30 and 60 seconds (3 retries by default, 0–10 configurable); messages that still fail go into a resend queue and are sent in time order once the channel recovers, and the queue survives restarts (kept 24 hours by default, 0–168 configurable, 0 turns it off; at most 100 messages per channel).
- While a channel already has messages waiting to be resent, a new message is tried only once and, if it fails, queued behind the backlog instead of waiting through retries one by one.
- "Recent deliveries" shows "Pending resend", "Resent" and the number of attempts; a banner above counts the messages waiting to be resent per channel and can clear them.
Certificates: use the fnOS system certificate (#105)
- The certificate source list gains "fnOS system certificate": pick a domain from the certificates fnOS already manages, and every 30 minutes Gately reads the latest certificate to follow fnOS renewals (using the full chain when one is available); after a change it loads and deploys the certificate automatically.
- Certificates are managed in fnOS under "Settings → Security → Certificate". The native fnOS app (fpk) reads them directly; for Docker deployments, mount
/usr/trim/etc/network_cert_all.confand/usr/trim/var/trim_connect/sslsread-only into the container at the same paths, and run the container as root (the default in the official image). See SSL Certificates.
Storage usage
- "System settings › Network & Runtime" gains a storage usage card: the capacity of the partition holding the configuration directory, the total size of the configuration directory, and the usage and path of each kind of data, including run logs, visitor audit, config history, backups, certificates, managed programs, the notification resend queue and File Browser data.
Fixes
- Intrusion Prevention: the admin port is registered as an open port before the modules start; previously, when the trap port range contained the admin port, Gately's own trap listener took it first, so the admin listener failed and Gately exited on every restart. Trap ports are no longer set while the modules start; ports registered by modules such as frps, EasyTier and FTP passive ports are never used as traps even when not yet listening; trap ports are recalculated as ports change; the temporary HTTP-01 listener used when requesting certificates is no longer blocked by trap ports.
- Web services: when a site's backend is another Web listener of the same instance (for example a 443 site → this machine's 444 listener → the admin UI), the admin UI's sign-in cookie is forwarded unchanged, so you are no longer sent back with a 401 right after signing in.
Upgrade notes
- Docker: pull the new image and recreate the container. OpenWrt: install the new
.apk/.ipkover the old one. fnOS: install the new.fpkmanually in the App Center. Other deployments can upgrade online in "System settings". - Confirm after saving is on by default. Saving configuration from scripts or with API tokens is unaffected; only saves from an admin UI session that carries the confirmation request header need confirming.