Plugins
The marketplace for swappable backends. Two categories ship today:
Activation modes: free, trial, license (handler-specific). Webserver activations rewrite every account's vhosts to the new backend — every hosted name, parked domains included, and a name's HTTPS from its certificate record (an uploaded certificate as well as a Let's Encrypt one; a name without a record but with a Let's Encrypt lineage keeps that lineage) — the live progress dialog (SSE) shows phase-by-phase status. A switch to LSWS Enterprise from Nginx or Apache installs the server first and then moves the sites into its configuration (the configuration exists once the server is installed). A switch to LSWS Enterprise while OpenLiteSpeed serves the sites switches OpenLiteSpeed off first (LiteSpeed's installer does not run beside it and LiteSpeed Enterprise does not read its site files): the sites go back to the web server they came from (Nginx + Apache), stay up there while LiteSpeed Enterprise is installed and then move into its configuration the same way; if the installation fails they stay there, and switching OpenLiteSpeed on again takes them back to it. Every switch to or from OpenLiteSpeed / LiteSpeed Enterprise first runs the configuration tests of the web server it moves to (and of any other web server running on the server): a test that already fails refuses the switch with no site moved and names the server. Accounts move one at a time, each one's old site files removed by its own web server (an Nginx + Apache account on an Nginx server included); if a site cannot be moved, the switch fails, every site goes back to the web server it came from, the plugin stays as it was and the dialog (or the API's error) names the site with the reason — when OpenLiteSpeed's or LiteSpeed Enterprise's configuration test refused the site, with the lines that refused it ([ERROR] lines, or the line naming the site with "configuration failed"; at most five, each cut at 300 characters). The agent log starts its record of the refusal with those lines and keeps the test's whole output, standard output and standard error, in the record that follows. An account's own requests (a PHP version, a domain, SSL) never quote the test's output, which can name other accounts' sites: they answer as before. The switch's last step stops the LiteSpeed server completely and has its service start it, and the switch counts as done only once that server generation has started its workers and holds the listener Nginx passes the sites to (port 8088) with no older generation still holding it; otherwise every site goes back to the web server it came from and the switch fails ("OpenLiteSpeed failed to start", "LiteSpeed failed to start"). A server the package installation started outside its service is started again by the service before anything reloads it. The switch also re-applies each home's and web root's mode and group (0750 and the web group on every web server: the LiteSpeed worker runs with the web group, so no other account passes through a home; homes an earlier release left open to other accounts on a LiteSpeed server (0711) are closed once at the agent's first start after the update, and the server is restarted gracefully when its configuration test passes); the web root is set through a handle, so a public_html that leads out of the account's home is left as it is and named in the agent log ("web root of … left as it is"). The LiteSpeed server (OpenLiteSpeed or LiteSpeed Enterprise) runs as nobody with the web group (www-data on Debian / Ubuntu, nginx on AlmaLinux / Rocky / CentOS Stream), set on the switch and checked at every agent start (a change takes one graceful restart), so it serves the account files the panel writes 0640 (the default page, File Manager uploads) the way Nginx and Apache read them; PHP keeps running as the account. While it serves the sites it writes their access and error logs in the accounts' log folders (see Account logs live outside the homes), and both OpenLiteSpeed and LiteSpeed Enterprise write the access line in Nginx's combined form with the visitor's address, so the Logs pages, the traffic counter and the web jails read it as under Nginx. One exception on LiteSpeed Enterprise: the 403 it gives on its own for a hidden file (a name that starts with a dot, such as /.env or /.git/config) is written with Nginx's address (127.0.0.1), which the web jails ignore, so those requests do not count toward a ban; other requests for files no site serves (such as /backup.sql) get a 404 with the visitor's address and do. The LiteSpeed server takes the visitor's address from Nginx's X-Forwarded-For on a connection that comes from 127.0.0.1, so while it serves the sites only root and the web server user (www-data / nginx) can open a connection to its listener behind Nginx (port 8088, on any of the server's own addresses); a process of a hosting account is refused there, so it cannot name the address an access line starts with and have the web jails ban it. A script that connected straight to that port gets connection refused; sites are reached by their names over ports 80 / 443 as before. The rule (a chain of its own in the packet filter, like the agent port's) is set when OpenLiteSpeed or LiteSpeed Enterprise is activated, at every agent start and when the firewall is enabled — each time the agent checks through Nginx that it still reaches the listener and takes the rule out again if it does not (the agent log says so) — and it goes when the server returns to Nginx or Nginx + Apache, once no LiteSpeed process listens on that port any more. Switching OpenLiteSpeed or LiteSpeed Enterprise off waits until the LiteSpeed server has stopped completely: a LiteSpeed process still finishing a request is waited for and ended after about half a minute, so the switch-off can take a few seconds longer while Nginx / Apache already serve the sites (if a process outlives that, the rule stays and the agent log says so). LiteSpeed Enterprise runs as its own system service (lshttpd, shown on the Services page): its vendor installer starts the server by itself, so the activation and the LiteSpeed update stop that server completely and start it again through the service (one more short restart inside the switch), and restarting, redeploying or updating the agent leaves it running. At every agent start a LiteSpeed Enterprise server found running under the agent, or missing while its service still reads active (for example after updating from a release that left it under the agent), is started again through the service and the agent log says so; a service you stopped is left stopped. A change of the web server's maximum clients or keep-alive timeout (Settings → General) restarts OpenLiteSpeed / LiteSpeed Enterprise through the same service — gracefully while it runs, and a stopped server is started by the service. On OpenLiteSpeed a name no site serves gets the server's 404: the switch and every start take out the package's catch-all to its example site, whose PHP would run as the server itself. While OpenLiteSpeed or LiteSpeed Enterprise serves the sites, Nginx holds ports 80 and 443 in front of it: every hosted name has its own Nginx file that serves the site's HTTPS with the site's own certificate (Let's Encrypt or uploaded, over HTTP/2) and passes every request to LiteSpeed, which keeps serving the static files, the WAF, the site's logs, redirects and a suspension; a name without a certificate is passed on the same way over HTTP, and its HTTPS answers with the panel certificate. Request bodies up to 256 MB reach the site (as on Nginx and Apache sites), and a certificate renewal reloads Nginx as well. After an update from an earlier release the first agent start writes these files (one render, a configuration test and one reload, without interruption). Names no site file serves are passed to LiteSpeed by Nginx's default blocks: saving the host name, the server IP or the force-SSL setting keeps that, and every agent start puts it back where a save in an earlier release removed it. An SSL install (from either panel or the API, and the switch for a name with a certificate) writes no listener on port 443 into the LiteSpeed configuration, and the activation and every agent start take out a listener on port 443 found there (written by an earlier release or added by hand — LiteSpeed answering 443 itself without Nginx is not a supported setup) once LiteSpeed's configuration test passes, followed by one graceful reload; so whichever of the two starts first at boot, Nginx gets port 443 and the panel and the sites come up. A configuration whose test already fails for another reason keeps the listener (the agent log says why). LiteSpeed keeps a port it already holds across a graceful reload, so when the agent starts and finds LiteSpeed still holding 443 with no listener configured there (Nginx could not start at an earlier boot), it stops LiteSpeed completely, starts it again through its service and starts Nginx — the agent log says so and the panel and the sites come back. On OpenLiteSpeed, installing or removing a site's certificate leaves the site on the listener Nginx passes it to, and every agent start puts back a hosted site that an earlier release took off that listener (such a site answered 404) once the configuration test passes; the package's example site stays off it. OpenLiteSpeed gives its own configuration folder to the group it runs as at every start, so processes in the web group can read it (its WebAdmin password file stays lsadm-only). On OpenLiteSpeed the account's certificate files are copied into the vhost's own folder through a handle on the account's ssl folder: a certificate name that is a link stops that domain's SSL step (the agent log names the file) instead of being read. A site's .htaccess rewrite rules apply on both servers (WordPress permalinks and the like): OpenLiteSpeed reads the rewrite rules of a site's .htaccess files (its other Apache directives are not read), LiteSpeed Enterprise reads the same .htaccess directives as Apache sites (AllowOverride All). The redirects set in the panel apply on both servers too: they are written into the site's own configuration ahead of every folder and of the .htaccess rules and match the whole path, as on Nginx. After an update from a release whose sites left the .htaccess rewrite rules unread (pretty-permalink pages answered 404) or did not apply the panel's redirects (the redirected path answered 404), the first agent start writes every site's configuration again (one render, one configuration test, one graceful restart). The switch installs LiteSpeed's PHP (LSPHP) for every PHP version of the server with the extensions its PHP-FPM version has — the MySQL driver, mbstring, XML, GD, intl, OPcache, cURL and the rest, under the package names of LiteSpeed's repository; one the repository does not build for a version is left out, and one package that cannot be installed keeps out only itself. A server already on OpenLiteSpeed / LiteSpeed Enterprise when it is updated from a release that installed the bare LSPHP build (a MySQL site on that PHP version answered 500) gets the missing extensions in the background about half a minute after the first agent start — after a guarded OS update run or other package work has finished — followed by one graceful restart; the agent log lists what was installed, and a version still without its MySQL driver is tried again at the next start. Activating OpenLiteSpeed also installs its ModSecurity module (ols-modsecurity) and wires the WAF before the sites move, so the moved sites keep their WAF (see WAF (ModSecurity)); when the module cannot be installed the activation still completes, the agent tries again every 30 minutes and the WAF page names the reason. Runtime plugin activations are essentially flag flips that add the page's entry (a tile in the admin settings hub, a sidebar entry in the client panel) and unlock the matching API surface.
The "Mark as resolved" amber banner appears when an activation pipeline was killed mid-flight (agent crash, OOM kill). One click clears the marker; the operator should verify system state before doing so.
An activation request is checked before anything runs, on both the blocking and the background path: an unknown plugin answers 404 PLUGIN_NOT_FOUND, an unsupported mode or a malformed serial key 400 PLUGIN_VALIDATION_ERROR, and no task is created for a refused request. Serial keys are trimmed and must consist of letters, digits, hyphens and underscores. Plugin state (including a masked serial key) lives in /etc/whost/plugins.json (root, 0600); the list endpoint never returns serial keys, the single-plugin endpoint returns a masked one. Webserver plugin installs fetch the LiteSpeed repository script and the LSWS tarball over HTTPS from LiteSpeed; the advertised version and build are accepted only as plain numbers. OpenLiteSpeed is updated the way it was installed: the update (POST /api/v1/plugins/openlitespeed/update; the OpenLiteSpeed card shows no update button) upgrades the openlitespeed package from the LiteSpeed repository with the server's package manager (apt on Debian / Ubuntu, dnf on the RHEL family; a configuration file you changed keeps your version) and, when the package version changed, restarts OpenLiteSpeed through its service; when the package already is the newest version the answer says Plugin is already the newest version and nothing is restarted. The answer's update_result names the package version before and after.
A webserver plugin's activation, deactivation and update change the web server's packages, files and units, so they run as package work started from the panel, one at a time: they are refused while the guarded OS update run or an OS package upgrade (409 OS_UPGRADE_IN_PROGRESS), a PHP version install or removal (409 PHP_PACKAGE_JOB_IN_PROGRESS) or another webserver plugin's work (409 PLUGIN_TASK_IN_PROGRESS) is in progress, and the toast under "Activation failed" (or "Deactivation failed", "Update failed") names what is running. A refused activation creates no task and its progress window closes; the panel falls back to the blocking route only where the background route does not exist (an older agent). A guarded run that starts while the work runs waits for it (an hour at most); one that started a moment before a background activation ends its task, failed, with nothing changed. An update of a plugin that is not active is refused first (400 PLUGIN_VALIDATION_ERROR). Runtime plugins only switch settings and are not held back.
An active web server plugin's card shows the installed server version (an inactive card shows the plugin's own version), and its Manage Plugin window has Restart LiteSpeed — a full restart through the server's service — and Restart PHP Processes: PHP runs under the LiteSpeed server (LSAPI), not as a service of its own, so this button restarts LiteSpeed gracefully, which starts every account's PHP processes again while the sites keep being served (the same restart the panel uses when a PHP setting changes); a refusal shows as Restart failed with its reason. Switching OpenLiteSpeed on while LiteSpeed Enterprise is active first takes LiteSpeed Enterprise away the way switching it off back to OpenLiteSpeed does: LiteSpeed Enterprise is stopped completely — no other web server serves the sites meanwhile, so a process still finishing a request is ended five seconds after the stop and that request is cut — then the openlitespeed package and its ModSecurity module package (ols-modsecurity) are installed again over the files LiteSpeed Enterprise's installer replaced and OpenLiteSpeed's service unit is put back, before the sites move to OpenLiteSpeed. The package reinstall makes this switch take about half a minute longer, during which the sites do not answer.
LiteSpeed Enterprise on a server that resolves names only over IPv6. LiteSpeed Enterprise checks its trial or licence with LiteSpeed's licence server through the host's resolver, and that check cannot reach the licence server when /etc/resolv.conf lists only IPv6 name servers (some provider images of AlmaLinux, Rocky Linux and CentOS Stream do). The activation therefore stops before it downloads anything and says why: LiteSpeed Enterprise was not installed: /etc/resolv.conf lists only IPv6 nameservers (…) — nothing is changed. Add an IPv4 name server to the resolver (for example the provider's IPv4 resolver) and activate again. When LiteSpeed Enterprise refuses to start for another reason (a licence the check rejects, for instance), the activation is undone and its error carries the server's own lines from the unit's journal, for example lshttpd did not start from its unit. LiteSpeed reported: [LICENSE] Failed to contact license server. [FATAL] license problem, try switching back to Apache!
LiteSpeed Enterprise's installer cannot reach LiteSpeed's hosts. The vendor installer downloads from LiteSpeed's hosts during the activation (for example the QUIC.cloud address list from www.quic.cloud). When one of them does not answer, the installer waits; after 5 minutes WHost stops it together with every download it started, nothing is switched, and the activation's error names what the installer was still running and its last line of output (for example … it was still running: wget … https://www.quic.cloud/ips?ln). Check that the server reaches that host over HTTPS (curl -sI https://www.quic.cloud/ips?ln) and activate again. Switching OpenLiteSpeed on after such an attempt first puts OpenLiteSpeed's package, its WAF module and its service unit back over the files the vendor installer left.
Endpoints used: GET /plugins, POST /plugins/{id}/activate/async, GET /plugins/tasks/{id}/events (SSE), POST /plugins/{id}/deactivate, POST /plugins/{id}/reset-state, POST /plugins/{id}/update, POST /system/service/lsws?action=restart|reload (the Manage window's two restart buttons).
Our support team is here around the clock for anything you can't find above.