What WPL7 does not do, does not contain or does not automate, as each page says it in its Limits section. Each heading links to the page with the details.
- The installer is for Ubuntu 26.04 on x86-64. The images are not built for ARM, so on an ARM
server it stops.
- Piped into
bash, the installer asks nothing. Every answer is a flag, and the panel
generates the admin password.
- The generated password is printed once, on the first boot. Editing
PANEL_ADMIN_PASSWORD
in .env afterwards changes nothing.
- Without a Cloudflare token, each dev site gets a certificate of its own. Let’s Encrypt
limits how many certificates one domain gets per week.
- Only Ubuntu 26.04 on x86-64. On another processor, such as ARM, the installer stops. On another
Ubuntu release,
setup.sh warns and carries on.
- The installer refuses a directory that already holds an install.
- Docker publishes ports past UFW: 80 and 443, UDP 443, plus the FTP ports while a site has an FTP
login.
- The generated password is printed once. Editing
PANEL_ADMIN_PASSWORD in .env later
changes nothing.
- An install from a checkout cannot use the update button in the panel.
- Lost the admin password? Reset a user’s password on the WordPress tab makes a new
one.
- A site’s slug cannot be changed after it is created.
- Going live allows search engines only with Allow search engines on. Without it they stay
discouraged until you untick Discourage search engines from indexing this site in WordPress
under Settings → Reading.
- The list has no search and no selection. To change many sites at once, use
Bulk management.
- The uptime check asks for one page. A site that answers it can still be broken elsewhere.
- During a job that starts, stops, rebuilds or updates the site, the badge keeps its reading from
before the job.
- Mode only says whether the site went live. It does not check that the domain’s DNS
still points at the server.
- The slug cannot be changed later. A few names are reserved, such as
admin, api, www,
new, bulk and import.
- A site has at most ten domains, and a domain belongs to one site only.
- Search engines stay discouraged until the site goes live, or until you untick the setting in
WordPress. Going live allows them unless you untick Allow search engines there.
- If the cleanup after a failure fails too, the site stays as Error. Delete site on its
Settings tab removes it.
- The browser keeps what you entered only until the page is reloaded. After a reload, a failed
site offers New site, which starts with an empty form.
- Multisite networks and WordPress in a subfolder cannot be imported.
- Symbolic links, and tables without the site’s table prefix, are not copied. Neither are views,
triggers and stored routines.
- A database row too large for the old host to read in one go is not copied. The job log names
it, with its table.
- The old site keeps serving during the import. Changes made there in the meantime can be
missing until a refresh.
- The old site serves visitors again when a refresh ends. Orders or comments taken there after
that are not copied.
- A refresh needs the plugin connected. After Disconnect, import the site again instead.
wp-config.php is the panel’s own. Database settings, keys, salts, paths and cache settings
are never carried over.
- The first copy of a big site can take hours.
- Multisite networks and sites on Windows cannot be connected.
- A backup is restored by hand. See Restore and download.
- A backup is not a copy of one instant: the site keeps changing while it is copied.
- Only commands that a plugin registered with WPL7 Connect run. WP-CLI’s own commands do not.
- WP Godmode does not register its command yet, so its chats cannot be reached on these sites.
- A rollback needs a writable
wp-content/mu-plugins. WordPress core updates are not rolled back.
- A host where WordPress cannot write its own files needs its FTP details in
wp-config.php for
updates.
- Files, FTP logins, caches, visitor statistics, malware scans, mail and site protection are for
sites the panel hosts.
| Limit |
Value |
| Cron lines per site |
10 |
| How often a line can run |
Every minute at most |
| Lines running at once, per server |
2. The others wait for the next minute |
| Stop after |
1 to 60 minutes, 5 by default |
| A command |
4,000 characters |
| What is kept of a run’s output |
The last 4 KB |
| Redis memory |
An eighth of the site’s memory limit, at least 16 MB. It counts toward that limit |
- There is no page cache, and none of WordPress’s tools:
updates, the plugin list, recipes,
one-click login and Bulk management.
- A site cannot switch between WordPress and PHP. A backup restores only into a site of its own kind.
- A PHP site has the PHP versions a WordPress site has. Older PHP is not offered.
- The site’s own database cannot be deleted without the site.
- Nothing keeps search engines off the dev address once the
site’s own files are up. Add a
robots.txt of your own until it goes live.
- A panel restart, or a lost connection to the server, loses the result of a line that was running.
The command runs on to its stop time, and the line counts as running until then.
- When Redis is full, it drops the least recently used keys, whichever they are.
- A scan moves files to quarantine by itself only among
WordPress’s files. On a PHP site, move a finding by hand, as
Malware scans explains.
- The login limit and attack detection know only
WordPress’s login. An app’s own login page gets the request limit only.
- Attack detection reads a PHP site’s 404s as any site’s. An API that answers 404 for many addresses
can get a busy client blocked.
- Attack detection counts every path under a folder such as
/phpmyadmin/ or /adminer/ as
Probing. An app served from such a folder can get its own visitors blocked.
- PHP runs in any folder. No PHP in uploads is for WordPress sites.
- There are no Git or Composer deploys. Upload
composer.phar and run it with Run a command.
- An app that sends by SMTP gets no SMTP details. Send with PHP’s
mail() or sendmail, without
sendmail’s -bs mode.
- A server whose site image is from before PHP sites cannot take one. Update the server first.
| Limit |
Value |
| Routes, variables |
100 each |
| A variable’s value |
32 KB |
| Mounts, data folders |
20 each |
| Health check wait |
5 to 600 seconds, 60 by default |
| Stop grace |
0 to 300 seconds, 20 by default |
| Registry check |
Every 5 to 1,440 minutes, 15 by default |
| An uploaded image |
8 GB |
| A mount’s archive |
4 GB |
| Images kept per app |
3, for Roll back |
- One container per app, and no builds on the server.
- A deploy answers 503 while the container is replaced: the old one’s stop, up to its
Stop grace, and the new one’s start.
- Files the image has under a
VOLUME are not copied into its data folder. The app starts with it
empty.
- A mount’s archive with a link, a special file, an absolute path or
.. is refused.
- An uploaded image is one image from
docker save, podman or skopeo: files, folders and links
only, at most 10,000 entries.
- An upload that no deploy or mount takes is removed after a day.
- An uploaded image cannot be rolled back to once it is gone from the server.
- An image that names its user by name needs an
/etc/passwd of at most 1 MB. Or give the user
as numbers.
- A Docker Hub image is checked against Docker Hub’s pull limits. Give it a login, or use another
registry.
- Run a command needs
sh and timeout in the image.
- An app has no request limits or custom rules, only the block list and attack detection.
- An app is stopped while it moves, and its routes do not change meanwhile. An app with a path of
another site’s domain does not move.
- Plain HTTP needs the Traefik of this release on the server, and on the server a move goes to.
Update that server first.
- The uptime check starts a short-lived container every minute for each app.
- At most ten domains per site, and a domain belongs to one site.
- The panel checks DNS but does not wait for it. A domain that points elsewhere gets no
certificate until its record is fixed.
- The Cloudflare option only reaches zones in the account the token belongs to. Other domains
need their records created by hand.
- Changing the domains later leaves search engines as they are. Only going live allows them, and
only with Allow search engines on. Otherwise untick Discourage search engines from
indexing this site in WordPress under Settings → Reading.
- A domain a container app takes a path of cannot be taken off its site,
and that site cannot be deleted or moved, until the app’s route is gone.
- A container app does not go live here. Its hosts are the ones its routes name.
- Changes need the site to be running. A stopped site shows its last scan.
- The panel does not renew the maintenance page, and Log in to WordPress does not work
while it shows.
- On a hosted site, a failed update is not rolled back. Restore the backup from the Backups tab.
- Deleting a plugin leaves its database tables and options behind.
- Longer commands belong in your own scheduled jobs.
| Limit |
Value |
| A file in the editor, and one save |
8 MiB |
| One upload |
2 GiB |
| Free disk an upload or extraction must leave |
1 GiB |
| Entries listed per folder |
10,000 |
| One search |
1,000 matches or 45 seconds |
| Entries one compress may take |
200 |
| Downloads at once, per server |
3 |
- The site has to be running. A stopped site’s tab offers Start the site.
- While any job other than a backup runs on the site, changes are refused. Browsing and
downloading carry on.
- Binary files and text that is not UTF-8 open read-only or download.
- A folder keeps work tidy, not people out. PHP uploaded there runs as the site and can read
all of it.
- FTP needs the server’s public IPv4 address. Without one, the server offers SFTP only.
- A cloud firewall in front of the server has to allow the ports. Only IPv4 is served.
- Repeated failed logins ban an address for 30 minutes, and longer each time.
- An interrupted upload starts over instead of resuming.
- The panel cannot see a change made over FTP or SFTP. A changed theme file reaches stored pages when
they expire, or after Clear cache. See Caching.
- A statement runs for at most 50 seconds from the SQL editor, and 20 seconds while browsing.
- The SQL editor shows the first 1,000 rows of a result, then stops the statement.
- A value over 1 MiB, and binary data, can only be changed with SQL.
- An import takes up to 4 GB, compressed, and up to two hours to load.
- A view, trigger or routine in a dump whose
DEFINER is another database user does not load.
Remove the DEFINER clauses first.
- Changes wait while a restore, import or move runs on the site. A backup does not block them.
- A change waits at most 10 seconds for a table the site is working in. Then it gives up and
changes nothing. Try again when the site is quieter.
- The forms take MariaDB’s usual column types. Change other types, triggers, routines and a
view’s query with SQL or in Adminer.
- MariaDB shows some characters of a default, such as an emoji, as
?. Such a default cannot be
kept as it is: the form asks for it again.
- A table with a full-text index cannot be copied for the try, and a foreign key is not tried:
those changes are checked when they run.
- A change cannot be undone, except from a backup.
- Countries follow address registrations. A company’s network shows where it registered its
addresses, not where the visitor sits.
- A crawler that presents itself as a browser counts as a visitor.
- Only requests that reach Traefik count. Pages a CDN serves from its cache never arrive.
- The first read of a server’s log goes back six hours at most. Nothing older is filled in.
- A request for a path a container app takes of another site’s domain
counts for the app, not for that site.
| Limit |
Value |
| A site’s page cache |
About 256 MB. The least used pages go first. |
| The longest a page stays |
10 hours |
| A 404 |
5 minutes |
| A site’s own rules |
50, of which 5 separate-copy rules |
| Redis |
An eighth of the site’s memory limit, at least 16 MB, out of that limit |
| A key in Redis |
4 hours |
- Logged-in visitors never get a stored page. The object cache is what speeds them up.
- A stored page never reaches
.htaccess or PHP. A change to the root .htaccess empties the cache
within 10 seconds. A block in PHP, such as a security plugin’s firewall, or in another folder’s
.htaccess, does not apply to a stored page until it expires.
- Site protection and
blocked addresses always apply, because they come before the
cache.
User-Agent, Accept, Accept-Language and X-Forwarded-* reach PHP but do not keep copies
apart. A page that changes by them without saying so in Vary freezes the first visitor’s version.
- A page that changes by what the request does not show, such as random content or location by
address, needs a skip rule or Off.
- A request with a cookie the cache does not know cannot store a page, though it can get one. Consent
cookies count as such. On a site where every visitor has one, nothing is stored.
- Behind a CDN other than Cloudflare, a header the cache does not know can stop pages being stored.
X-Cache-Skipped on a page that came through the CDN names it. Check a URL does not go through
the CDN, so it cannot.
- A store’s country header works only from Cloudflare.
- A clear empties the whole site’s cache. There is no clear for one page.
- A visitor’s approved comment clears the cache at most once a minute.
- After a clear or an expiry, every request for a page goes to PHP until one has been stored. PHP’s
workers take them in turn.
- A 404 at an address with no file extension stays for its 5 minutes, even if you upload a folder
there over FTP. Clear cache ends it.
- Tracking parameters are not put back on a redirect to another domain, or when the address already
has a query and the request has a separate-copy parameter.
- A site hosted elsewhere has no caches. Neither has a site that still runs
the old PHP setup: see Updating WPL7.
- A PHP site has no page cache. Its Redis is for its own code. The panel never
clears it, but every restart empties it.
| Limit |
Value |
| A request body |
1 GB, or post_max_size in uploads.ini plus 16 MB when that is more. Past it, a 413 |
| Bodies of 1 MB or more at once, per site |
16. The next one gets a 429 |
| One request header line |
8 KB |
| All request headers |
About 60 KB |
| One response header line |
8 KB. Past it, a 500 |
| All response headers |
32 KB. Past it, a 502 |
| One PHP request |
About an hour, at most 80 minutes |
| Running requests at a stop |
5 seconds. Recreate container ends them at once |
php_value and php_flag in a .htaccess give a 500 for its folder, and do nothing inside
<IfModule mod_php.c>. Use .user.ini.
- A change to a
.user.ini takes up to 5 minutes.
- A higher upload limit set only in
.user.ini gets a 413 past the body limit above.
LimitRequestBody in a .htaccess has no effect on PHP files.
- PHP runs only in files named
.php. A file a .htaccess hands to PHP under another name answers
403. One handed to a handler Apache does not have, such as cPanel’s ea-php, is sent as it is,
source included. The job log names each such line. Rename the file to .php.
- A download PHP serves holds a worker for as long as it runs.
- A PHP page with
Last-Modified or ETag can get a 304 or 412 from the server, even if its body
changed.
- A plain script that does not load WordPress gets a 500 for a header line with no colon, and
http_response_code() does not replace a status it set with header(). A plugin that registers
its own header_register_callback brings both back on WordPress’s pages.
apache_get_modules, apache_get_version, apache_getenv, apache_setenv, apache_note,
apache_lookup_uri, apache_response_headers and virtual do not exist under PHP-FPM.
SetEnv no-gzip does not stop Brotli. <IfModule mod_headers.c> blocks now apply, and
Header unset needs always.
- Repeated response headers are merged into one, except
Set-Cookie.
- A URL with
%00 in it gets a 400.
- A site gets at most the CPU cores its server has. A higher CPU limit gives it all of them.
- Switching PHP replaces the container, so the site does not answer for a few seconds.
- If a recreated container starts but does not answer, Recreate container leaves it in
place. The job log warns about it. A container app’s goes back to the old one instead, as a
deploy does.
- A vendor that cannot be reached keeps its license activation. The job warns, and the site is
deleted anyway.
- Docker’s WordPress image follows a PHP release by a few hours. Sites cannot move before it.
- Only the images of the newest release and of the newest edge build are rebuilt. An install on an older one gets a new PHP with a WPL7 update, and the card says so.
- A site busy with another job moves at the next hourly check.
- A coming release’s fixes are not known before php.net publishes it.
- php.net’s notice is read from its wording. A notice whose date does not read shows without one.
- Sites hosted elsewhere are not updated. Their PHP is their host’s.
- A dev site keeps serving while it is copied, so changes made meanwhile are lost.
- A site keeps its dev address. One created under another server’s dev domain keeps that domain,
so it needs an
A record of its own on the new server.
- Outgoing mail now leaves from the new server’s address. With direct delivery, add that
address to each domain’s SPF record.
- A hostname with an
AAAA record is never finalized automatically. Use Finalize now.
- Older backups stay on the old server. Restoring one needs its offsite copy, fetched back
first.
- The old server cannot be removed while a cleanup is pending.
- The site’s added databases stay on the old server until the move is finalized, as its own does.
- A live PHP site that is stopped for the move is offline while it is copied.
- A PHP site moves only to a server whose site image can run one. Update that server first.
- A container app is always stopped while it is copied, so its worker never
runs twice. Its image goes to the new server first: pulled there by its digest, or carried
through the panel when it was uploaded. Its mounts kept out of backups are copied beside the
snapshot. Its routes do not change while it moves.
- A container app with plain HTTP routes moves only to a server whose Traefik answers plain HTTP.
Update that server first.
- A container app that takes a path of another site’s domain does not move, and neither does that
site.
- A failed update is not rolled back. Restore the pre-update backup from the site’s
Backups tab.
- Pre-update backups stay until you delete them or their keep period ends, so they use disk.
- A site that has never been scanned cannot join a run.
- Bulk management changes what is installed. It does not install new plugins.
- PHP sites are not listed, because they have no WordPress.
- A database belongs to one site. There are no databases outside a site, and no other database
users.
- MariaDB has no port open to the internet. Only sites on the same server reach a database.
- Sizes are InnoDB’s estimates.
- A move stops when the new server already has a database with the name of one of the site’s.
- Deleting a database no site uses needs Full access.
- Adminer opens the databases of sites on your servers, not of sites hosted elsewhere.
- It needs an admin signed in to the panel in a browser. An API key or an AI app cannot open it.
- An import through Adminer takes up to 32 MB. Import bigger dumps on the site’s Database
tab.
- While a job other than a backup runs on the site, such as a restore or a move, Adminer can
browse but not run SQL or save changes.
- A change made in Adminer, such as a query or an import, runs to its end even if you close the
tab. The site’s jobs wait until it has finished. One that runs past 30 minutes is stopped.
- The panel cannot see a change made in Adminer. Choose Clear cache on the site’s Overview tab
afterwards, or visitors keep getting stored pages until they expire.
- Adminer looks like Adminer, with its own design and its own dark mode.
- On a server installed from source, the first open builds the Adminer image, which takes about
a minute.
- The panel emails when a site goes down, with Site down, and back up on in
Settings. It sends no email when a server goes down.
- The email for a server low on disk, at 10% free, watches the disk that holds
/srv. A backup
location on a disk of its own is not watched.
- The uptime check runs from the server itself, through its Traefik. It does not test DNS, or
the route a visitor takes from outside.
- The disk figures cover the filesystem that holds
/srv. A backup location on another disk
shows under Backups → Storage instead.
- Resource samples are kept for seven days. The charts show at most the last 24 hours.
- The dialog connects on SSH port 22. Add a server on another port through the API, with
sshPort.
- Without a Cloudflare token, the panel creates no DNS records and gives no reminder. Create each
site’s record by hand.
- The panel’s key is root on every server that trusts it. Protect
/srv/panel/ssh like a root
password.
- The panel does not move sites off a server that is down.
- At most ten terminals are open at once, across all servers and people.
- A session closes after 30 minutes without input or output.
- An open shell keeps running after you sign out, or after your account is removed, until it exits
or idles out.
- On the panel’s own server, the terminal connects to SSH on port 22. The SSH daemon must accept
root logins with a key.
- An API key needs Full access to open a terminal, and AI apps over MCP cannot open one.
- The shell is root. Nothing it does is checked, contained or undone by the panel.
- A site’s disk figure counts its files, not its database. Databases
shows the databases’ sizes.
- A backup needs free disk of one and a half times the site’s size. With less, it fails.
- Docker publishes its ports past UFW. A UFW rule does not close 80, 443 or the FTP ports.
- The FTP ports are published on IPv4 only.
- Housekeeping runs once a night. During the day, the panel only emails you when a server has
less than 10% of its disk free.
- Docker images under names of your own are left alone, used or not. A
container app’s images are its deploys’ to clear: each keeps the image it
runs and the two before it. On a server that builds its own site images, a WordPress base image
a newer pull left without a name goes too.
- While an update of the panel runs, the panel’s server keeps its unused images until the next
night.
- The panel shows no container logs. Read them in the Terminal.
- The numbers are as old as the last read. A plugin installed in wp-admin appears after the next
scan.
- Stopped sites are not read on schedule. Check now needs the site running.
- The panel takes at most ten Check now requests a minute.
- Must-use plugins and drop-ins are listed, but the panel does not manage them.
- A failed read keeps the previous list, and shows the error beside it.
- Malware that was in a zip when you uploaded it, and that AMWScan does not know, is vouched for
on every site that has the file.
- A zip with no single folder at its top, or one that unpacks to more than 1 GB, vouches for
nothing.
- Zips are checked only while Scan every site is on, in Settings → Security.
- The check never holds back an install. A new site can get a zip while its check runs, or
before you review what it flagged.
- In the panel, the catalog serves new sites only. A site’s WordPress tab installs from
wordpress.org. An uploaded zip reaches an existing site through the API only.
- A new advisory shows once the panel next refreshes that slug, up to a day later.
- A theme wordpress.org closed shows as closed only if the feed flags it.
- A temporary closure, pending wordpress.org’s review, shows like a permanent one.
- On a hosted site, a failed update is not undone automatically. Restoring the backup also undoes
every other change made to the site since.
- Where no fixed release exists, Fix vulnerable leaves the plugin as it is. Deactivating it is
your decision.
- An update from the WP-CLI console, a recipe, WP Godmode or WordPress’s own automatic updates
takes no backup first, whatever Back up before updates says.
- Every administrator of a site with the plugin can read its key. That is what licensing a site
means.
- Release on delete is best effort. If the vendor cannot be reached, the activation stays.
- Whether a dev address counts against a license is the vendor’s rule.
- Disabling or uninstalling a recipe leaves what it already activated. Its constants leave a site
at that site’s next recipe run.
- A command that prints a key in another form, cut short or in capitals, shows it in the job log.
- A recipe step runs inside the site’s container. It can do what a site administrator can, and
nothing on the server.
- A backup reads the files while the site serves visitors. A file changed meanwhile may not match
the database dump.
- A stopped site is not backed up on schedule.
- A backup restores onto a site of the same name only. There is no restore into a new site under
another name.
- The panel’s SSH key is not in the panel’s own backup. Keep a copy of
/srv/panel/ssh somewhere
else.
- Only scheduled backups and panel snapshots are kept to a number. Manual and final backups are
never deleted by age.
- A restore replaces everything that changed since the backup: posts, orders, comments and uploads.
- A restore stops before it changes anything when a database in the backup is now another site’s,
or sits on the server unused.
- A backup restores only onto the site of the same name, on the server that holds the backup.
From another server, fetch it back from an offsite copy, or move the site back first.
- A PHP site’s backup restores only into a PHP site, and a WordPress site’s only into a WordPress
site. A container app’s restores only into a container app.
- A container app’s backup of an uploaded image no longer on the server does not restore. Upload
the image again first.
- A deleted site’s backup has no Restore. Create a site with the same name first.
- A site hosted elsewhere is restored by hand, as above.
- The panel’s own database is restored by hand, never from the panel.
- The files a restore set aside are deleted by the nightly housekeeping, a day or more later.
- An API key needs Manage access to restore, and Full access to download the panel’s own database.
- A location may not be
/, a system folder, or inside or around the live data under /srv:
sites, mysql, panel, mail, traefik and plugins.
- When the panel takes
BACKUP_ROOT at start, it moves nothing. Backups taken before stay where
they were, listed and restorable, until retention removes them or you delete them.
- The location of a server the panel cannot reach cannot be changed.
- Backups stay on their server when a site moves to another one.
- The panel lists only the copies it made. Older backups in a bucket you set up again are read by
hand.
- Removing a destination drops the backups kept only there from the panel’s lists. Their files stay.
- Credentials and the passphrase are stored in the panel’s database in plain text. Encryption
protects the copies from the provider, not from someone who has the panel.
- A panel snapshot holds every credential the panel has. Copy it only where you would keep them.
- A Custom rclone remote takes its other settings through the API only. A Provider cannot be
changed after saving.
- Copies are checked after upload, never test-restored.
- A deleted backup cannot be recovered, from the server or from a destination the panel cleared.
- A backup cannot be deleted while it is being written or uploaded, or while a job for its site may
be reading it.
- A destination that manages its own retention keeps its copies. Remove them with the provider’s
tools.
- Deleting a backup’s local files with Keep the remote copies leaves it restorable only after
Fetch back.
- The keep period deletes what is on the servers. Offsite copies follow each destination’s own
retention, which counts scheduled backups and panel snapshots only.
- A site’s own keep period cannot keep scheduled backups past Keep last N scheduled backups per
site.
- One deletion takes at most 5,000 backups. Delete old backups asks for fewer when more match.
- WPL7 sends mail and does not receive it. Mailboxes for your customers’ domains live elsewhere.
- A relay that stops takes
mail() away from every site on its server. Mail never fails over to
another server.
- Mail from a domain without a DKIM key goes out unsigned. Mail from a domain no site uses is
accepted and never signed, and the recipient’s own checks judge it.
- The smarthost is set per server. The panel does not copy it to the servers it adds.
- A site’s sendmail has no
-bs mode. An app that uses it needs /usr/sbin/sendmail -oi -t instead.
- Reverse DNS and a smarthost’s SPF value are always set by hand.
- Without a token, or for a zone it cannot reach, you add the records by hand. The panel publishes
through Cloudflare only.
- SPF is planned for the delivery mode of the panel’s own server, and checked for IPv4 only.
- The panel suspends a site by itself but never resumes one. A person decides.
- Mail the relay accepted before a suspension is still delivered, unless you empty the queue.
- The guard looks at the last hour only. A slower run stays under it and shows only as a mark.
sent means the receiving server accepted the message, not that it reached the inbox.
- Traffic comes from each relay’s log. If the panel cannot read a log for longer than the server
keeps it, the messages in between never appear.
- Mail deleted on the server with
postsuper -d ALL keeps its last status in Traffic, because
Postfix logs no message ids for it.
- The levels allow logins and XML-RPC in bursts of 2, because of an
open Traefik bug. Traefik gives an address that
pauses a few seconds a full burst again. Attack detection still blocks an address after 20 login
attempts in 10 minutes.
- Requests and assets keep large bursts, which a client renews the same way.
- Rate limits start from zero whenever a server’s rules change.
- Without IPv6 in Docker, IPv6 visitors are not limited one by one.
- A proxy that sends only
X-Forwarded-For cannot be trusted.
- Request bodies are never read. A POST exploit is refused by its path, or not at all.
- The settings inside the container stop a stolen admin login, not code already running in the site.
- A stored page never reaches PHP or
.htaccess. Site protection comes before the
page cache and still applies, but a security plugin’s firewall does not run.
- A container app gets no request limits and no custom rules.
- Detection reacts within about two minutes, not seconds.
- SSH and FTP are not covered. They keep their own protection.
- The firewall and Traefik come before the page cache, so a blocked address gets no stored page. A
block in a
.htaccess or in a plugin does not stop a stored page. See
Caching.
- A crawler on a domain the panel does not know is not verified, and can be blocked.
- A bot whose company publishes no address list is blocked like any visitor when it crosses a rule.
- Without IPv6 in Docker, direct IPv6 visitors arrive as one private address and are not detected
one by one.
- A scan reads files, not the database. A backdoor that rewrites its file from there comes back.
- Premium plugins and themes have no checksums. A backdoor in one is found only if AMWScan knows it.
- Malware already in a catalog zip when you uploaded it, and unknown to AMWScan, is vouched for on
every site that has it.
- Calls to
eval() and the like are not findings on their own. A backdoor that hides behind nothing
else is found only if a signature knows its shape.
- Signatures lag behind new malware. Scripts over 1 MiB are read only at their start and end.
- A
.user.ini that makes PHP load a file is reported wherever it is, apart from Wordfence’s firewall
at the site’s top. It is not moved on its own.
- Two-factor authentication guards the browser sign-in only.
- Removing an admin leaves every API key working, including keys that person created.
- A terminal already open outlives the session that opened it, until it exits or sits idle for 30
minutes.
- On an account without a second factor, whoever reads its recovery mailbox can reset its password.
- The published address lists and the country tables have no switch of their own.
- Traefik’s version check has no switch in the panel.
- Ubuntu itself, such as its package updates, is outside this list.
- A long job delays the other jobs on its server.
- The job console shows the jobs started in that browser tab. Jobs started by schedules, the API,
AI apps, other admins or other tabs are under Automations → All jobs.
- A bulk run shows its progress on Sites → Bulk management. A Run now that starts several
jobs lists them under its schedule.
- The job console’s place belongs to the browser, not to your account.
- The panel runs at most eight jobs at the same time, across all servers.
- A job stopped partway, by a cancel or a timeout, can leave its work half done. Its log shows how
far it got.
- An interrupted job is not resumed.
- Built-in schedules cannot be deleted, and their timing is changed in Settings, not here. The
housekeeping time is fixed.
- The tasks marked Always on cannot be paused.
- New monitoring intervals take effect after the panel restarts.
- Pausing scheduled backups also stops the nightly snapshot of the panel’s own database.
- Through the API, pausing, resuming or running a built-in schedule needs a Full key.
- At most 100 custom schedules.
- Each server runs one command at a time. A long command delays the other sites’ commands there.
Sites hosted elsewhere share two lanes of their own.
- A started command cannot be canceled. Only its time limit stops it.
- A
wp godmode command that waits on a chat is refused.
- Scheduled backups count toward the backup retention, so frequent runs push out older backups.
- A shell command, a WP-CLI command that changes something and a REST request other than GET empty
a WordPress site’s caches each time they run.
- Registry dates may differ from registrar billing dates. No WHOIS or registrar APIs.
- Domain lookups run daily. Failures keep the last date.
- Disable lookups in Settings → Monitoring → Look up domain expiry dates.
- Schedules pauses reminder processing and domain lookups separately.
- Paid support extensions are not detected.
- Calendars show active reminders’ next occurrence. Calendar apps control refresh intervals.
- API and MCP changes require Full access.
- A key belongs to the panel, not to a person. Two-factor authentication does not apply to it, and
endpoints that ask for your own password refuse it.
- A Manage key can read whatever a site holds, license keys included. What it creates inside a site
stays after you revoke it.
- The activity log keeps at most the newest 100,000 requests.
- The panel sends no webhooks. Your tool polls jobs.
- Some endpoints are never reachable through MCP, whatever the level: signing in, admin accounts,
API keys, the activity log, uploads, binary downloads and the terminal.
- What an app creates inside a site stays after you revoke it: WordPress users, application
passwords, FTP logins, changed files.
- The panel offers tools only. It has no MCP resources or prompts.
- On a site hosted elsewhere, WP-CLI commands run only if a plugin
registered them with WPL7 Connect. WP Godmode’s chats cannot be reached there yet.
- Giving the work to WP Godmode is a rule in the instructions the panel gives an app when it
connects. The panel does not enforce it, and an app that leaves the instructions out does not
follow it.
- The panel writes
A records only, never AAAA, because servers are added by their IPv4
address.
- The panel writes records through Cloudflare only, in the zones the token reaches. Other domains
need their records created by hand.
- The records it writes are not proxied through Cloudflare.
- Every server issues its own wildcard certificate, for the same names. Turn it on for at most
five servers a week.
- Settings apply to the whole panel. A site’s own PHP version and protection are on its own page.
- Every admin can change every setting.
- The panel’s domain, the mail relay and other values in
deploy/.env are not here. See
Configuration.
- The backup schedule and other values read from
deploy/.env on the first boot belong to the panel
afterwards. Editing the file changes nothing.
- A panel that always has a job running keeps updates waiting.
- The button needs the panel’s SSH access as root to its own host. A host that refuses root logins
updates from the command line.
- An update that brings new site images rebuilds every site once. Each site is unreachable for a second or two.
- A site that cannot move to PHP-FPM keeps its old image’s PHP, and the PHP security releases in newer
images do not reach it until it moves.
- The move cannot tell what a site’s visitors should see. A page that changes by something the request
does not show needs a rule or Off, as Caching says.
- There are no lesser roles. Every admin controls the whole panel and every server, terminal
included. Give an account only to someone you would trust with root.
- Two-factor authentication guards the browser sign-in only. API keys are a separate credential.
- An owner who forgot the password and has no recovery email, or lost both the phone and the
recovery codes, needs a shell on the server to get back in.
- The theme and accent belong to the browser, not to your account. They do not follow you to
another device.
- The Machine card describes the panel’s own server, not the other servers.
- The Enterprise package is a premium service offered on wpl7.com, not inside the panel.
setup.sh is written for Ubuntu 26.04. On another system it prints a warning and carries on.
- There is no uninstall script.
--dns-provider, --dns-token-stdin, --admin-user and --admin-password only count on the
first run, when deploy/.env is written.
--dns-provider=hetzner writes HETZNER_API_KEY. Traefik’s ACME library reads that variable
as a key for Hetzner’s legacy DNS API, which it marks as deprecated.
install.sh refuses a directory that already holds an install. Use update.sh there.
- Each site has a Docker network of its own, and the panel creates it from Docker’s default
address pools. On Docker 29 these hold 31 networks.
- The stack’s own networks and Docker’s default bridge take four of them, and FTP two more on a
server with FTP logins. That leaves room for about 25 sites per server.
setup.sh replaces /etc/docker/daemon.json with provision/daemon.json whenever they
differ, so a larger address pool added there by hand does not survive the next run.
- The panel is root-equivalent on its own host. Anyone who can sign in to it can run anything
on its servers.
- A step runs inside the site’s container as
www-data, so it can do what a site administrator
could do and nothing on the host.
- Site administrators can read a license key once its plugin is on the site. That is what
licensing a site means.
- Job logs hide a secret only where a step prints it exactly. A vendor’s command that prints it
changed, such as in capitals or cut short, can reveal it. Callers with Read only access never
see a failed step’s output.
- Each plugin has one recipe at a time.
- The steps restore the files and the database. A site rebuilt on another host runs without what
the panel adds: its own mail login, its protection rules and its resource limits.
- The database password of a WordPress site is not in its backup. On another host, the site gets
a new database user.
- The
-apache image’s mod_php does not read .user.ini, so settings the panel moved there do not
apply. In each .htaccess, take the # Moved to .user.ini by WPL7: mark off the line to bring one
back.
- Restoring the panel’s database rolls back everything the panel knows to that night, users and
API keys included.
- The relay and the DKIM signer follow
latest. An install that pulls its images gets their
newest release whenever setup.sh runs, which every update does.
- Traefik and MariaDB follow their minor versions, so patch releases arrive the same way.
- AMWScan runs with its signature updates switched off. Its signatures are the ones in the
pinned image, and they age until a WPL7 release moves the pin.
- The site image’s Debian packages have the fixes of the day a WPL7 release builds it. Redis stays
at the version in Debian’s stable release, with Debian’s security fixes, not Redis’s newer releases.
- The Redis Object Cache plugin updates like any plugin on the site, and WPL7 tests it at one version.
- Only the newest release gets fixes. There is no backporting while WPL7 is at 0.x.
- The bug and security links on the Support page follow
WPL7_REPO, so an install from a fork
sends its people to the fork’s repository.