Repository navigation
Nightly build (develop)
Pre-releaseAdded
-
An availability group can be failed over from its context menu: Fail Over…. Pick the replica to
fail over to, point DataTray at the saved connections for the instances involved, and it shows the exact
plan — every statement and the instance it runs on — before anything runs. Which plan you get follows the
group's cluster type:- WSFC: one statement on the target. A plain
FAILOVERwhen the target is synchronous and every
database is failover-ready; otherwiseFORCE_FAILOVER_ALLOW_DATA_LOSS, with a red warning, how far
behind each database is, and what to do about the old primary afterwards. - Read-scale (
CLUSTER_TYPE = NONE): the documented sequence across both instances — make both
replicas synchronous, wait until the target isSYNCHRONIZED, take the group offline on the primary,
promote the target, demote and resume the old primary, re-create the listener. It stops before taking
the group offline if the target never catches up. - Pacemaker (
CLUSTER_TYPE = EXTERNAL): DataTray refuses, because the cluster manager owns the
primary role, and shows thepcs/crmcommands with the target node filled in to run yourself.
A forced or read-scale failover asks you to type the group name. Each picked connection is checked to
really be the replica it was picked for, and when you confirm, the state is read again: if the plan
changed in the meantime (a target that fell behind turns a planned failover into a forced one), nothing
runs. Failing over is not reachable from the MCP server. - WSFC: one statement on the target. A plain
-
New availability groups can be created from the Availability Groups folder: New Availability
Group…. The instance you opened it on becomes the primary; tick the saved connections that should hold
the secondaries. A pre-flight over all of them comes first, and while anything blocks — one instance
only, Always On switched off, different versions or collations, an edition without availability groups,
a login withoutCONTROL SERVER, endpoints on only some instances — it shows only why, and no form. After
that the form offers only what can work on those instances: WSFC only when every instance is a node of the
same Windows cluster, read-scale (NONE) never preselected because it is not high availability, a basic
group on Standard edition, and only the databases that can join (the others are listed with the reason).
DataTray creates certificate-authenticated endpoints and exchanges the public certificates between the
instances itself, without writing a file anywhere, or reuses the endpoints if every instance already has
one. The databases are seeded automatically. As with failing over, the whole plan is shown per instance
before anything runs, and it is checked again when you confirm. -
Plugin releases are published by the build pipeline. Bumping a plugin's version now builds it,
packages it and puts it in the Plugin Store on its own, so a new plugin version reaches the store the
day it is finished instead of waiting for someone to build and upload it by hand. Store versions
carry a build number (0.8.0.41) identifying the exact build a plugin came from. -
The query editor and result grid can now be resized by dragging the splitter between them —
previously they held a fixed 2:3 height ratio. The chosen height is remembered across restarts, the
same way the resizable bottom tool panels (Output, Containers, AI activity) already are. -
SQL Server's Administration folder gets an Availability Groups entry, next to Agent Jobs. It is
badgeddisabledwhen Always On is switched off on the instance, and absent entirely on an
edition/platform with no Always On support at all (Azure SQL Database, for instance). -
An availability group now opens the same read-only dashboard SSMS shows: group info (cluster type,
quorum, listener, backup preference), a replica table, and a per-replica database table with a derived
"behind by" column — the primary's last commit time minus the replica's, in seconds, next to the DMVs'
own log-send/redo queue sizes in kilobytes. It is the only node dialog in DataTray that polls (every 10
seconds): queue depth moves while you are looking at it, unlike everything else this dialog family shows.
Failing a group over is its own action on the group (Fail Over…), and creating one is New
Availability Group… on the folder. -
Filter the connection tree from the sidebar. A field under Connections narrows the tree while
you type: connection names when everything is collapsed, and schema/table/view/column names inside
whatever you have expanded. Non-matching rows are hidden rather than greyed out, the path down to a
hit stays visible so you can see where it lives, and a footer line says how many of the loaded
objects matched.Escclears it andCtrl/Cmd+Fputs the caret in it (rebindable, under
Settings ▸ Keyboard). It never expands a branch by itself — what you have not opened has not been
fetched, and is therefore not searched. This is the counterpart toCtrl/Cmd+K, which jumps to
one object and closes again; the filter stays put. -
Text in a single-line input sits centred again. Every
TextBoxis a fixed 32px tall, but its
content was anchored to the top, so the slack piled up underneath and the text read as sitting high —
worst in the new sidebar filter, visible everywhere else. Multi-line boxes (SQL previews, cell values)
keep their first line at the top, where it belongs. -
The filter shows which part of a name it matched, highlighted in the row, and the filter box keeps
an accent ring while it is narrowing the tree so a partial tree never looks like the whole one. The
search glyph and the clear cross now sit inside the field. -
You choose where DataTray keeps your connection passwords. Onboarding has a third step, between
picking an engine and adding your first connection: the system vault your OS already runs (macOS
Keychain, Windows Credential Manager, Linux Secret Service) or an encrypted file in your DataTray
profile. Both options say who holds the key, who else can reach it and what you lose if it goes wrong,
so the choice is made on the trade-off rather than on which sounded safer. -
The encrypted file store works where there is no keychain at all. A headless or locked-down Linux
box has no Secret Service, and until now that meant DataTray could not save a password there at all —
it fell over at startup. That machine now gets the file store, with the vault option greyed out and the
reason spelled out. The file is AES-256-GCM under a master password you set in the same step; DataTray
never writes a secret into it without one. -
Preferences ▸ Security says which store you are on. And with the file store active, "Disable" for
the master password is gone — the store cannot work without it — replaced by the sentence that says so.
Existing installations are on the system vault, exactly as before, and nothing moves on its own. -
DataTray tells you what it is doing while it starts. A small window shows the real startup steps
(applying staged plugin updates, loading your plugins, putting your workspace back, starting
extensions) as rows filling a tray. A step that takes longer than five seconds says so, how long it has
waited and why DataTray needs it; after thirty seconds on your plugins you can start without the ones
from the Plugin Store. If a step fails you see the error and can retry, open the log, continue with the
built-in engines or quit, instead of the app closing without a window. The splash follows your theme,
shrinks to a compact layout on small screens and after a restart, and holds still when your system asks
for reduced motion. -
Plugins can now add a pane to query tabs, read the connection's tables and columns, and link the editor (host API 9). Plugins can also add items to table/view context menus.
-
Drag a table or view from the tree into the query editor to insert its qualified name at the drop point.
-
Extension plugins can contribute settings and shortcuts, and uninstalling a plugin now asks for confirmation first.
-
Long queries show what the server is saying, while it says it. A query tab gets a Messages tab
next to its results as soon as a run has something to report:PRINTandRAISERROR(also
WITH NOWAIT, line by line as the script runs) and(n rows affected)per statement on SQL Server,
NOTICE/WARNINGon PostgreSQL, and warnings on MySQL. Every line has its time; warnings and errors
are marked and counted, and an error that stops the run sits in the same timeline as the messages
before it. Filter by text or level, copy the selection or what is shown, clear it, and the list follows
the newest line until you scroll up. It keeps the latest 10,000 lines and says when older ones were
dropped. While a query runs, Run becomes Stop with a ticking clock beside it, and the status bar shows
the latest message; afterwards it says whether the run completed, was cancelled or failed, and where.
Messages are cleared when the next run starts, or kept behind a divider with
Settings › Query › Keep messages between runs. Engines without messages show only the clock. -
Prometheus provider (PromQL, read-only). Connect to a Prometheus-compatible endpoint and query
it in PromQL from a normal query tab. The tree browses metrics (with their type and help text) and
their labels; instant queries, range selectors (up[1h]) and subqueries all come back as a grid of
<labels…>, timestamp, value. Read-only by nature — the HTTP API accepts no writes, so nothing in
the grid or the DDL menus pretends otherwise. Debug-staged like the otherplugins/providers.
Changed
- SSH.NET updated to 2026.0.0. The library behind SSH tunnels was on a version carrying a known
vulnerability (GHSA-q939-rpr3-3284), which made every
build report it. The flaw is in SCP file transfer — an API DataTray does not use, so nothing was ever
exposed through it; this is about not sitting on a vulnerable version. Tunnels behave as before: password
and private-key authentication, host-key fingerprint checking, and the forwarded local port are unchanged. - Manage Connections reads as sections instead of one long scroll. The selected connection now has
a fixed header — colour, name, provider, where it connects, its folder, and pills for read-only and
AI access — that stays put while you scroll, so you always know what you are editing. Below it the
form is grouped into Identity, Connection, Advanced and Access, with labels beside their
fields instead of above them, which roughly halves the height of the form. The Advanced section is
now subdivided by the groups each provider already declares (Security, Connection, SSH tunnel). - The connection colour is a named dropdown. The same seven colours as before, but each one has a
name you can read, and "no colour" says so instead of being an empty circle. - AI access is three radio buttons rather than a dropdown, so which level is granted — and that a
read & write level exists — is visible without opening anything. The fail-closed default, the write
warning and the "never show to the AI" block are unchanged. - The action bar keeps one primary action. Duplicate and Delete moved to a
⋯menu next to Save;
they remain one click away as icon buttons in the toolbar top-left. The empty panel now offers New
connection and New folder instead of only a line of grey text. - Smaller form details from the design: a port or timeout gets a short box instead of a full-width
one, a dropdown no longer stretches across the whole column, file paths are shown in the monospace
font, and the required-field marker is red rather than another grey character in the label. The
collapsed Advanced header names what is inside it (Security · Connection · SSH tunnel), the test
button reports its result with a green/red dot instead of grey text either way, and a selected folder
shows how many connections and subfolders it holds. - Managing connections no longer opens a separate window. It is now a state of the main window: the
sidebar switches from the schema tree to the connection/folder tree — with its drag-and-drop
reordering and folders intact — and the connection form opens where the editor tabs are. Your open
tabs are untouched underneath and come straight back when you close it with the × next to Manage
Connections. Nothing was lost in the move; the tree, the form and every action are the same ones,
just in the window you were already looking at. - Both sidebar states now look like one sidebar. The connection tree kept the roomier row height and
larger icons it had as a 960x640 window of its own, so switching into Manage Connections made the
same column visibly change size. Rows, icons and the filter box now match the schema tree exactly. - The sidebar header is the same control in both states. A title on the left and one icon button on
the right, at the same coordinates: a settings icon that opens connection management, and the × that
leaves it again. New connection moved off the header down beside the filter, as the same small icon
button the manage toolbar uses — so switching states no longer changes the shape of the column. - Preferences is easier to find your way around. The eleven pages are grouped in the rail under
Workspace, Querying, Access and Extensions instead of being one flat list, every page opens with its
own title and a line saying what it is for, and the longer pages are split into numbered sections —
so General now reads "1. Language & startup", "2. Window & instances", "3. Schema tree",
"4. Updates" rather than a run of settings with the odd bold heading in between. - Preferences opens at the top of the page again. The Updates channel cards used to drag the
General page down with them as the window opened, so it started halfway through. - DataTray draws its own title bar. The menu moved up into it, which gives the window back the row
the separate menu bar used to take, and the bar itself now says which connection you are on and which
document is in front of you — something the system title bar could only ever spell "DataTray". On
macOS the traffic lights stay where they belong; elsewhere the minimise, maximise and close buttons
are DataTray's own. - Scrollbars are always shown in full. They used to be a thin overlay line that only widened into a
grabbable bar after you hovered it for a moment, so a quick drag often missed. Now every scrollbar, from
the result grid and SQL editor to the connection tree, lists, text fields and dialogs, is a full bar you
can grab at once, with content beside it instead of underneath. Prefer the slimmer look? Turn on
Preferences › Appearance › Overlay scrollbars. It applies live.
Fixed
- The macOS application menu says "About DataTray" again. It still carried Avalonia's own
fallback item, so the first entry under the Apple logo read "About Avalonia" and opened the
framework's about box instead of DataTray's. - "What's new" now actually shows what's new, and the Updates settings can start the download. The
release notes were never put into the update packages, so the dialog opened empty on every channel —
stable included. They are now rendered once per build and embedded, so the notes on the release page
and the ones the app shows are the same text. The Updates pane in Settings also gained a Download
Now button beside What's new, with the progress and the install step alongside it, so an update
can be taken without closing Settings to reach the banner. - A query tab opened from the tree now starts on that node's database. "Select top 1000", the
SQL command templates and "INSERT (with data)" opened their tab on the connection's default database
(master) instead of the one the table lives in — visible in the database picker, and a real failure
on engines that cannot reach across databases. New Query, reopening a file, duplicating a tab and
Ctrl+Shift+T keep the database too, and switching a tab's connection no longer drops it while the
new database list loads. - The setup wizard's "Install…" now opens the Plugin Store on the engine you clicked. Every tile
under "From the Plugin Store" opened the store on whichever plugin its own list happened to show
first, so clicking ClickHouse could land you on DragonflyDB. - Restarting from the Plugin Store during setup brings the wizard back. The store's own "Restart
now" ended onboarding on the way down, so an engine installed there left you with no wizard, no
connection and no sign of what had been installed. The wizard now returns to the engine step and
says which engine landed. - Typing a schema before the dot now narrows the completion list to that schema.
SELECT * FROM [dbo].offered every table in every schema — andFROM dbo.without brackets offered every
column of every table, because the schema name was read as a table alias that resolved to
nothing. A dot in FROM/JOIN is now understood as a schema qualifier, and[dbo],`dbo`,
"dbo"and baredboare all read the same way. - The tree's context menu no longer acts on the previously selected node. Right-clicking the empty
space below the last row left the old selection standing and opened the menu against it — so
Drop Schema… could be confirmed for a schema you never clicked. A click that hits no node now
clears the selection and opens no menu. - Drop and Alter dialogs name the database they will run against. The confirmation used to show a
bareDataSync, whileDROP SCHEMA [DataSync]names no catalog of its own; it now reads
rick-test › DataSync, so which database is about to lose the schema is visible before you confirm. - Dates and numbers follow the operating system again, not the UI language. Choosing English in
Settings › General also switched every date, time and number to English formatting, so a result
grid on a Dutch machine showed12/2/2025 11:10:40 AMinstead of2-12-2025 11:10:40. The language
setting now only changes the wording; the machine's regional settings decide how values are written. - The macOS application menu actually says "About DataTray" now, and gained Preferences.
SE-254's earlier fix compiled but never took effect on screen: it set the menu after
Avalonia.Native had already locked in its own fallback. Moving it earlier fixes that, and
macOS also lost its Edit/Help menus — Preferences moved into the app menu (Cmd+,, standard
macOS placement) and About was already there too, so both were left empty of anything else. - SQL Server tables whose only uniqueness comes from a filtered (partial) unique index — a common
soft-delete pattern,UNIQUE (OrgId, ItemId) WHERE IsDeleted = 0— can be edited again. The grid
showed these as read-only with "no primary-key column to identify rows", because SQL Server never
reports a filtered index's columns as a key. Editing now recognises the index and ANDs its filter into
every generated UPDATE/DELETE, so it can never touch a row outside the filter (e.g. a duplicate
soft-deleted row sharing the same key values). - A transaction script with
UPDATEstatements no longer fails on MSSQL with "Incorrect syntax near
the keyword 'ORDER'." ABEGIN…END/IF…ENDblock with aSELECTright beforeENDand no;
separating them (a common shape inside a conditional branch) confused the paging heuristic's
semicolon-only statement splitter: it glued the trailingSELECTto theEND/COMMITthat followed
into one "statement" that still started withSELECT, which then wrongly gotORDER BY … OFFSET … FETCH NEXT …appended for query-result paging.CASE…ENDinside a real singleSELECTis unaffected. - A query that leaves out its primary key no longer shows a phantom
Idcolumn full of NULLs.
SELECT Col1, Col2 FROM Tab1(SQL Server and Postgres) got an extra, unrequested primary-key column
in the result grid, every row NULL.CommandBehavior.KeyInfo— needed so the grid knows which rows are
editable — makes the driver append that column purely as key-resolution metadata; the provider counted
it as a normal selected column instead of skipping it. - F5 on a Data tab refreshes the table instead of running an empty query. Pressing Run/F5 while
browsing a table sent the query editor's (empty) text to the server, which MongoDB and Elasticsearch
rejected with "Empty query." while the grid kept showing the rows it had loaded before. It now
reloads the current page, the same way Discard already did. - Truncating a table clears its open Data tab. A truncate refreshed the object tree but left an
open Data tab for that table showing the rows that had just been deleted; the tab now reloads along
with the tree, as it does after any other change to the table. - Stop now stops a SQL Server script that already printed something. Once a script had sent a message
(for exampleRAISERROR … WITH NOWAITin a batch-delete loop), Stop only took effect when the whole
script had finished, so every remaining batch still ran. The script is now cancelled on the server
within a second, and what it printed until then stays in the Messages tab. - Updating no longer loses your open tabs. Installing an update closed DataTray without saving the
session, so the new version started with empty or outdated tabs. The tabs, their text (also unsaved),
the active tab and the window layout are now saved before the update installs, the same way as when
you quit. The open tabs are also saved every 30 seconds while something changed, so a crash or a forced
quit costs at most that.
Security
- A read-only AI connection can no longer be changed through a SQLite
PRAGMA. Over MCP, every
PRAGMAcounted as a read, so an AI client could runPRAGMA user_version = 999or
PRAGMA journal_mode = OFFagainst a SQLite connection whose AI access was set to read-only, and
change the database file. Only introspection pragmas without a value, such asPRAGMA table_info('x'),
are still allowed there. Every other pragma is refused unless the connection is a sandbox.
These builds are unsigned / not notarized.
macOS: after moving to /Applications, clear quarantine: xattr -cr "/Applications/DataTray.app".
Windows: pick the -Setup.exe (per-user, no admin) or the -Portable.zip; SmartScreen may warn on first launch.