Repository navigation
Releases
pbsgui v0.3.0-rc.7
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
Testing release
This is a pre-release for testing and has not yet been validated end to end on
real hardware. If you want the last tested, stable build, use v0.2.0:
https://github.com/sol1/pbsgui/releases/tag/v0.2.0
Please try this build in a non-production environment and report what you find.
New in this release
- Always On availability groups: log backups no longer get stuck. With
point-in-time protection, pbsgui took full backups on the preferred backup
replica, which SQL Server sets to a secondary by default. A secondary can only
take copy-only full backups, and those do not start the log chain, so log
backups could fail with error 4214 indefinitely and the log never truncated
(log_reuse_wait_desc = LOG_BACKUP). Now:- Full backups run on the primary, as normal full backups.
- Log backups run on the preferred backup replica, as before.
- When a database has no full backup to start its chain (newly added, or after
a recovery model change), the primary takes a full at its next log interval,
even if it is not the preferred replica. A secondary that cannot take its log
backup for that reason reports a warning, and fails the run if that lasts
over an hour, so a missing primary setup does not go unnoticed. - In a Basic availability group (Standard edition), which allows no backups on
a secondary, everything runs on the primary. - The replica's role is read from SQL Server directly, so a non-readable
secondary is no longer mistaken for the primary.
- Log backups start on the first interval. A new point-in-time job no longer
waits for its first scheduled full: its first log interval takes a full, then
log backups follow. A replica whose fulls run on another node takes log
backups straight away. - Read-only databases under point-in-time protection no longer attempt log
backups. - New Always On guide (docs/ALWAYS-ON.md):
setting up every replica, what runs where for each plan, checking and changing
the group's backup preference, readinglog_reuse_wait_desc, shrinking a log
once after it grew, and troubleshooting.
Upgrading
Install over the previous version as usual. The app and the service must be the
same version; the installer updates both.
Known limitations
- The availability group changes, VSS file backups, Windows-account SQL
connections, and the relay are new and unvalidated on real hardware; use v0.2.0
for production until they have been tested. - A Windows-account SQL connection signs in with NTLM, so it does not work where
the domain blocks NTLM. - Distributed availability groups are not supported yet.
- With Always On or a failover cluster, use the same PBS API token on every node:
PBS refuses backups into a snapshot group from a token other than its owner. - Restoring a very old snapshot (one taken before pbsgui stored the database file
list) under a new name or into a different instance is not supported over the
relay; restore it under its original name, or run the restore from pbsgui
installed directly on the SQL host.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.3.0-rc.6...v0.3.0-rc.7
pbsgui v0.3.0-rc.6
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
Testing release
This is a pre-release for testing and has not yet been validated end to end on
real hardware. If you want the last tested, stable build, use v0.2.0:
https://github.com/sol1/pbsgui/releases/tag/v0.2.0
Please try this build in a non-production environment and report what you find.
New in this release
- File backups no longer fail on open or locked files (#5, #6). pbsgui now
takes a Volume Shadow Copy (VSS) snapshot and reads the files from it, so
locked files such as a signed-in user's profile underC:\Usersare backed up.
The snapshot is a copy backup, so SQL Server and other VSS writers do not
record it as one of their backups. - Unreadable files are skipped, not fatal. A file that still cannot be read
(permission denied, or a OneDrive file that is only in the cloud) is skipped and
the rest of the backup continues. The run finishes with warnings: the job
card turns amber and names the skipped files, the run log lists them all, and
notifications report WARNINGS. - No more large temp file. File backups stream straight to PBS instead of
writing the whole archive to the temp folder first. - Weekly and monthly schedules (#5), for file jobs and for SQL full backups.
Weekly takes any set of days (for example weekdays at 22:00). Monthly takes a
day from 1 to 31; a day past the end of a shorter month runs on its last day. A
run missed while the service was down now runs as soon as it is back. - SQL connections with a Windows account or a SQL login. The SQL Servers tab
has a connection form (Add connection, and Edit on a saved one) with a choice of
the service account, a Windows account (DOMAIN\user), or a SQL login, and a
Test button that runs the readiness checks. Use this for remote instances
instead of changing the account the pbsgui service runs as. - Configuration is never overwritten after a failed load. Previously, running
the service as a different Windows account showed empty settings, and saving
anything then replaced the real configuration. pbsgui now leaves the files
untouched, refuses changes, and shows an error explaining the cause and how to
recover (set the service back to Local System). - Fix: a SQL connection entered as
HOST\MSSQLSERVERnow works as the
default instance for backup and restore.
Upgrading
Install over the previous version as usual. The app and the service must be the
same version; the installer updates both.
Known limitations
- VSS file backups, Windows-account SQL connections, and the relay are new and
unvalidated on real hardware; use v0.2.0 for production until they have been
tested. - A Windows-account SQL connection signs in with NTLM, so it does not work where
the domain blocks NTLM. - With Always On or a failover cluster, use the same PBS API token on every node:
PBS refuses backups into a snapshot group from a token other than its owner. - Restoring a very old snapshot (one taken before pbsgui stored the database file
list) under a new name or into a different instance is not supported over the
relay; restore it under its original name, or run the restore from pbsgui
installed directly on the SQL host.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.3.0-rc.5...v0.3.0-rc.6
pbsgui v0.3.0-rc.5
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
Testing release (new architecture)
This is a pre-release for testing a new architecture and has not yet been
validated end to end on real hardware. If you want the last tested, stable build,
use v0.2.0: https://github.com/sol1/pbsgui/releases/tag/v0.2.0
Please try this build in a non-production environment and report what you find.
New in this release
-
Remote SQL Server backup through a relay proxy (the new architecture). A thin
pbsgui agent on the SQL Server host now runs only the backup device and streams the
raw backup over an encrypted connection to a separate pbsgui "proxy" machine, which
does the deduplication, compression, encryption, and upload to PBS. The heavy CPU
work moves off the database server, so a fleet of SQL Servers can be backed up
without each one paying that cost during the backup window. Both backup and
point-in-time restore stream over the relay.Setup: on the proxy run
pbsgui-engine relay add-agent <name>(it prints the exact
command to run on the SQL host), run thatrelay joincommand on the SQL host, then
in the SQL Servers tab route a saved connection through the agent with the Relay
button. The SQL Servers tab shows each agent's live connection status. -
Snapshot group safety. A SQL job that would write colliding PBS snapshot groups
(for example two databases whose names reduce to the same id) is now refused when
saved, and the job wizard shows the exact snapshot groups a job will create so the
per-database naming is clear up front.
Known limitations
- The relay path is new and unvalidated on real SQL Server and PBS hardware; use
v0.2.0 for production until this has been tested. - Restoring a very old snapshot (one taken before pbsgui stored the database file
list) under a new name or into a different instance is not supported over the
relay; restore it under its original name, or run the restore from pbsgui installed
directly on the SQL host. - The installers and binaries are code-signed by Sol1 Pty Ltd.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.3.0-rc.1...v0.3.0-rc.5
pbsgui v0.3.0-rc.1
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
Testing release (new architecture)
This is a pre-release for testing a new architecture and has not yet been
validated end to end on real hardware. If you want the last tested, stable build,
use v0.2.0: https://github.com/sol1/pbsgui/releases/tag/v0.2.0
Please try this build in a non-production environment and report what you find.
New in this release
-
Remote SQL Server backup through a relay proxy (the new architecture). A thin
pbsgui agent on the SQL Server host now runs only the backup device and streams the
raw backup over an encrypted connection to a separate pbsgui "proxy" machine, which
does the deduplication, compression, encryption, and upload to PBS. The heavy CPU
work moves off the database server, so a fleet of SQL Servers can be backed up
without each one paying that cost during the backup window. Both backup and
point-in-time restore stream over the relay.Setup: on the proxy run
pbsgui-engine relay add-agent <name>(it prints the exact
command to run on the SQL host), run thatrelay joincommand on the SQL host, then
in the SQL Servers tab route a saved connection through the agent with the Relay
button. The SQL Servers tab shows each agent's live connection status. -
Snapshot group safety. A SQL job that would write colliding PBS snapshot groups
(for example two databases whose names reduce to the same id) is now refused when
saved, and the job wizard shows the exact snapshot groups a job will create so the
per-database naming is clear up front.
Known limitations
- The relay path is new and unvalidated on real SQL Server and PBS hardware; use
v0.2.0 for production until this has been tested. - Restoring a very old snapshot (one taken before pbsgui stored the database file
list) under a new name or into a different instance is not supported over the
relay; restore it under its original name, or run the restore from pbsgui installed
directly on the SQL host. - The installer is unsigned, so Windows SmartScreen warns on first run.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.2.0...v0.3.0-rc.1
pbsgui v0.2.0
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
New in this release
-
Restore to a different SQL Server, or to a network folder. A SQL database can
now be restored into a different SQL Server instance: pick a saved connection as
the target, and pbsgui relocates the database files to that server's default data
and log directories after checking the target is reachable and a sysadmin. A
restore to files can also target a network (UNC) path, prompting for credentials
when the backup service cannot reach the share on its own. -
One database's failure no longer stops the rest. A SQL job covering several
databases now backs up every eligible one and reports a clear per-database summary,
instead of aborting the whole run on the first failure. A database in SIMPLE
recovery is skipped for transaction-log backups with a plain message (SQL Server
manages that log itself) rather than being treated as an error, and any database
whose log backups are failing is now visible instead of silently growing its log. -
Backups and restores leave CPU headroom. Compression and encryption no longer
run wide open across every core, so a backup, and especially a restore, no longer
saturates a live database server. The default is about half the machine's cores;
set the PBSGUI_WORKER_LIMIT environment variable to widen it for a dedicated backup
window. -
Compact notifications. Slack and webhook messages are now a compact one or two
lines led by a status symbol (success, warning, or failure), with human-readable
sizes (for example 716 GiB) instead of raw byte counts. Email and the in-app log
show readable sizes too.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.1.0...v0.2.0
pbsgui v0.1.0
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
New in this release
This release is about trust: a backup or restore is now either complete and
correct, or it fails loudly, and the place pbsgui keeps its settings is locked
down.
-
Restores are verified end to end. Every chunk pulled back from PBS is checked
against its content hash as it arrives, so a corrupt or tampered datastore can no
longer produce a silently wrong restore. A point-in-time restore also refuses to
run unless the backup chain actually reaches the moment you picked: if the logs
needed to get there are missing or unreadable, you get a clear error naming how
far the backup really covers, instead of a restore that quietly stops short and
reports success. -
Backups never commit a partial snapshot. If a SQL Server BACKUP fails partway,
or you cancel a running backup, the truncated bytes already streamed are discarded
rather than finalised, so a half-written snapshot can never be mistaken for a good
one. A backup that did finish is always recorded as the success it was, even when
a cancel arrives at the same instant. -
Hardened settings storage. The configuration directory is restricted to SYSTEM
and administrators, and the job and connection files are signed and written
atomically, so tampering or corruption is detected and a half-written file is never
loaded. The pre/post job hook scripts have been removed, closing the path they
offered for running arbitrary commands as the backup service. -
Steadier under load. Concurrent edits to jobs no longer risk losing an update,
and a GUI that disconnects mid-run no longer stalls the run: a manual backup always
finishes and is recorded, the same as a scheduled one.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.0.10...v0.1.0
pbsgui v0.0.10
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
New in this release
- Much faster restores. Restoring from PBS now downloads chunks concurrently
(the read-side of how backups already upload) and overlaps the network read with
the disk write, instead of fetching one chunk at a time and waiting a full network
round-trip for each. A restore that was bursty and stalled on network latency is
now steady and limited by your bandwidth or disk. Both the "Save to file" export
and the restore back into SQL Server benefit: the in-place restore now prefetches
chunks ahead of SQL Server so it rarely waits on the network.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.0.9...v0.0.10
pbsgui v0.0.9
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
New in this release
- Restore a SQL database to a file. A new "Save to file" action in the SQL
restore view writes the native SQL Server backup to a folder without touching a
SQL Server: pick a full restore point for its.bak, or a point in time for the
covering full plus its.trnlog backups and a steps file with the exact
RESTOREstatements. The files restore on any SQL Server (SSMS or sqlcmd), so you
can recover to a different or newer server, hand the backup to a DBA, or migrate. - Large databases now restore. The to-file export and the restore back into SQL
Server both stream the backup chunk by chunk instead of holding it in memory, so a
multi-hundred-GB or terabyte database restores without running out of RAM.
Restoring under a new name relocates the database files using a file list now
recorded in each backup; backups made before this release restore under a new name
the previous way. - Clearer "Active runs" switcher. When more than one backup or restore is in
progress, the tabs that swap the output between them are now labelled and read
clearly as tabs.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.0.8...v0.0.9
pbsgui v0.0.8
pbsgui backs up Windows files and Microsoft SQL Server to a Proxmox Backup Server
(PBS), with browse and point-in-time restore.
New in this release
- See backups that are still running, even ones you did not start. Open the
app and any job with a run in progress shows "Running..." on its card with a live
progress bar and status line, including a backup the scheduler started in the
background. Click it to watch progress and Stop it if needed. - Run several backups at once and switch between them. The run output panel
keeps a separate live log and progress per run and has a tab bar to swap between
them, so a manual run, a scheduled one, and a restore can all be watched at once.
A job's Run button reads "Running..." for the life of its run. - Clearer status while a backup starts. Instead of a generic "starting..."
while a large database is read, the status shows the current phase (connecting,
reading a database, archiving files), so a long initial read no longer looks
stalled.
Installers are attached below. See the
README for which one to choose, plus
setup and upgrade steps.
Full Changelog: v0.0.7...v0.0.8
pbsgui v0.0.7
pbsgui is a Windows GUI for backing up Windows files and Microsoft SQL Server to a
Proxmox Backup Server (PBS), with browse and restore. It talks to PBS using a
clean-room reimplementation of the PBS backup protocol.
This is an early release. Please test against non-production data first and report
issues.
New in 0.0.7
- Fix: upgrading to 0.0.6 could briefly run every SQL job at once. 0.0.6 added
per-job scheduling timers that older state did not carry, so right after
upgrading the scheduler saw every existing SQL job as "never run" and ran them
all on its next cycle. Jobs that have run before are now gated by their last
successful backup, so only genuinely due jobs run. If you hit this on 0.0.6 it
was a one-time event - the extra runs were ordinary backups, not data loss. - Clearer token validation error. When a PBS token authenticates but still
cannot back up, the validation message now explains PBS privilege separation - a
token's effective rights are the intersection of the token's and its user's
roles, so both need the DatastoreBackup role on the path - and prints the exact
command to grant it to the user.
New in 0.0.6
- Backup compression. Backups are now compressed with zstd before upload, so
compressible databases and files take much less space on PBS and less bandwidth
to send. It is on by default and configurable per job; data that does not shrink
(already-compressed or high-entropy content) is stored as-is, so compression
never inflates a backup. It is a pure-Rust implementation, so there is no extra
runtime to install, and the format stays compatible with the PBS client. - Live backup progress and stats. A running backup now shows real-time
throughput, how much was deduplicated, the live compression ratio, and - for SQL
Server - a percentage against the database size. The same figures are exported to
the optional Prometheus endpoint (current bytes and throughput while running,
plus per-run compression and dedup stats). - Stop a running backup. A new Stop button cancels an in-flight backup cleanly:
the partial snapshot is discarded on the server and the SQL Server BACKUP is
aborted. A job can also no longer run twice at once, so a manual run cannot
collide with a scheduled one. - Reliability fixes.
- A backup that had actually succeeded could be reported as FAILED when PBS
closed the connection at the very end (right after committing the snapshot). It
is now reported correctly as a success. - A failed (or wrongly-failed) backup could be re-run by the scheduler over and
over, sending a flood of failure notifications. After a failure the scheduler
now waits for the next scheduled slot, so one failure is one notification.
- A backup that had actually succeeded could be reported as FAILED when PBS
- Token paste fix. A PBS API token id or secret pasted with a trailing space or
newline is now trimmed before use, so a token that "isn't working" only because
of invisible whitespace is accepted. This also repairs already-saved tokens. - Fix: "backup service is not reachable" after upgrading to 0.0.5. 0.0.5
restricted the engine's control socket to administrators (a security fix), but a
normally launched GUI runs with a UAC-filtered token and could not satisfy that,
so it could not connect. The GUI now ships with a manifest that requests
elevation, so Windows prompts for admin rights at launch and it connects again.
You no longer need to right-click "Run as administrator" - just accept the UAC
prompt. The background service keeps running scheduled jobs regardless.
New in 0.0.5
- PBS server validation. A saved PBS server is now checked the moment you add
it: reachability, the pinned TLS fingerprint, that the token authenticates, and
that it actually holdsDatastore.Backupon the datastore (and namespace). The
server's indicator turns green only once it passes, and a Test button reports the
precise reason on failure (unreachable, fingerprint mismatch, bad token, or
missing DatastoreBackup) so you catch permission problems at setup, not at 2am. - Clean in-place upgrades. Installing a new version over an old one now stops
and restarts the background service cleanly, and your configuration (jobs,
connections, servers, notification and metrics settings) and stored secrets are
preserved automatically. No need to uninstall first. - Prometheus metrics (optional). The engine can export per-job and per-database
metrics (last run, result, duration, size, dedup counts, and the point-in-time
chain freshness and stall state) either over an HTTP/metricsendpoint or as a
pbsgui.promtextfile for a node/windows_exporter collector. Off by default,
bound to localhost, and the metrics carry no secrets. - Clearer PBS errors. A permission or credential failure when starting a backup
now explains itself (for example, that the token lacks DatastoreBackup on the
datastore, or that another owner holds the backup group) instead of showing a raw
status code. - Hardening (security review). The engine's control socket is now restricted to
administrators, so only an administrator can drive backups (see Important
below). Several internal cleanups removed dead code paths.
Removed: the one-off per-database "To PBS" / "To file" buttons on the SQL Servers
tab. Back up SQL Server through a job (a saved connection plus a protection plan),
which uses your saved credentials, can encrypt, and is browsable and restorable.
Install
Two installers are attached; pick one:
pbsgui_<ver>_x64-setup.exe(full, ~200 MB) - bundles the WebView2 runtime
and installs with no internet access. Use this for air-gapped servers.pbsgui_<ver>_x64-setup-online.exe(small) - downloads the WebView2 runtime
at install time if it is missing (most Windows 11 and updated Windows 10 already
have it). Use this when the machine has internet; it is a much smaller download.
- Download and run one of the installers.
- The installer is unsigned, so Windows SmartScreen may warn: choose
More info -> Run anyway. - The installer registers and starts the
pbsgui-enginebackground service, then
opens the app. Add a PBS server and (optionally) a SQL Server connection, then
create a job.
What works
- File and folder backup to PBS with content-defined chunking and incremental
deduplication; browse and restore (full or selected files). - SQL Server: local and network discovery, connection probe and readiness checks,
outcome-driven protection plans with point-in-time recovery (fulls plus log
backups streamed over the Virtual Device Interface), and restore to a moment in
time or a chosen full, over the original database or a new name. Always On and
Failover Cluster aware; system databases supported. - Optional client-side AES-256-GCM encryption, byte-compatible with the PBS scheme;
keys are stored in the Windows Credential Manager. - Notifications on job success, failure, and a stalled point-in-time chain, via
email (SMTP) and a Slack-compatible webhook, each with a Test button. - A jobs dashboard with status and size-over-time graphs; PBS servers are validated
(reachable, fingerprint, token, DatastoreBackup) with a health indicator. - Optional Prometheus metrics (an HTTP
/metricsendpoint or a textfile), off by
default. - Runs as a Windows service with a tray icon; scheduled jobs run without the GUI
open.
Important
- The app prompts for elevation (UAC) at launch. It manages backups through the
engine, which runs asLocalSystem, and the engine's control socket is restricted
to administrators (a security fix); the GUI requests admin rights so it can
connect, so accept the prompt. The background service runs scheduled jobs
regardless of whether the GUI is open. - Upgrading from an earlier version: just run the new installer over the old
one. It stops and restarts the service for you, and your jobs, connections,
servers, settings, and secrets are kept. - Encryption keys cannot be recovered. If you lose an encryption key, backups
made with it cannot be restored by anyone. The key is shown only once, when you
create or import it, so copy it into a password manager then. - SQL Server backup requires TCP/IP enabled on the instance and the engine's service
identity (NT AUTHORITY\SYSTEM) in thesysadminrole. See the docs. - Always On: install pbsgui on every replica and give each replica's job the
same backup id (it defaults to the Availability Group name) so they share one
continuous chain. Exactly one node backs up at a time, and it follows failover
automatically.
See the documentation for setup
and details.
Full Changelog: v0.0.6...v0.0.7