Changelog

The server, the dashboard and the client packages ship together under one version number.

Latest v0.29.0Nov 18, 2026
86
Releases
911
Changes
v0.29.0
Latest

One bar per release, as tall as its changes. Lime is a minor.

minor

storage · performance · postgresql · sqlite · sso

Smaller Storage

Events, transactions and spans take up to 75% less disk, PostgreSQL can give the space back, and OIDC groups map to project roles

Upgrading

  • Back up before upgrading. There is no downgrade: v0.28.0 does not start against a migrated database, and the down migrations refuse to run once a compressed payload exists. Going back means restoring the backup.
  • Migrations are near-instant: schema changes plus three small index builds (0.7 s on a 4 GB SQLite database). They run before the server listens.
  • Existing rows are not rewritten. They stay readable as before, but only rows written after the upgrade are smaller. Disk usage drops as retention removes the old ones.
  • PostgreSQL: once your retention period has passed, run Settings → Storage → Compact tables once to return the space to the disk. Do it at a quiet time: each table is locked while it is rewritten and needs free disk about the size of the largest table. See Reclaiming disk space, including pg_repack if you cannot lock.
  • Span start_timestamp, timestamp and duration_ms are now rounded to the nearest microsecond instead of truncated, so they can differ by up to 1 µs.

Smaller storage

Measured on the same dataset (50k error events, 100k transactions, 1.14M spans):

v0.28.0 v0.29.0
SQLite 4,022 MB 983 MB
PostgreSQL (after VACUUM FULL) 2,299 MB 843 MB
SQLite WAL peak during an ingest burst 8.0 GB 0.3 GB
  • Event, transaction and span payloads are stored zstd-compressed, with a dictionary each instance trains from its own rows
  • Spans are stored once: transactions no longer embed their spans array. API and MCP output is unchanged
  • Three span and transaction indexes nothing read are dropped, and the gen_ai span indexes only hold AI spans
  • Retention now deletes standalone spans past the transactions window; they were never deleted before
  • The SQLite WAL is truncated once it passes ~40 MB instead of growing for a whole ingest burst
  • PostgreSQL autovacuum runs at 1% dead rows on events, transactions, spans and logs instead of 20%

Compact tables on PostgreSQL

PostgreSQL never returns the space of deleted rows to the operating system. Compact tables on the Storage page, which did nothing on PostgreSQL, now rewrites events, transactions, spans and logs with VACUUM FULL, one at a time.

  • Manual only, never scheduled. A failed run leaves the table as it was
  • While a table is rewritten, error events keep being accepted. Transactions, spans and logs bound for it get 429 with X-Sentry-Rate-Limits and are dropped by the SDK
  • The storage response lists per-table sizes on PostgreSQL

Project roles from OIDC groups

OIDC_GROUP_ROLES gives roles from provider groups, on one project or on all of them (@rane-keba):

OIDC_GROUP_ROLES="shop-team=Viewer@storefront;platform=Developer"

Every SSO login syncs the account's assignments of the roles the mapping names. Roles it does not name, and password accounts, are never touched. See SSO.

Improvements

  • Deleting an issue returns to the issue list
  • lettre 0.11.24 fixes the RUSTSEC-2026-0560..0565 SMTP advisories
  • Transaction ingest is 15–40% faster, and the span list and trace view 4–5× faster on SQLite
minor

dashboard · pwa · mobile · auth

Installable PWA

The dashboard is now an installable PWA with a mobile shell, and sessions survive browser restarts

Installable PWA

The dashboard ships a web app manifest and a service worker, so it installs as a standalone app on phones and desktops.

  • Install from the browser, or from Install in the mobile menu
  • The service worker caches only fingerprinted /assets/*, cache-first. Navigations always go to the network and fall back to offline.html when it fails
  • /api, /auth and every other server path bypass the worker, so no signed-in response is ever cached on the device
  • The cache is keyed to the release, and a new version drops the old one on activation

Mobile shell

  • A bottom tab bar for issues, projects, search and more, with a sheet for switching projects and settings
  • Pull to refresh
  • Tables fit the screen and keep their sort and column headings on narrow widths

Persistent sessions

  • The session cookie is now persistent (Max-Age 30 days), so closing the browser no longer signs you out
  • Idle timeout goes from 1 to 14 days, sliding on last_seen_at; absolute lifetime from 14 to 30 days. Both are enforced server-side

Improvements

  • The bulk actions toolbar wraps on phone widths
patch

dashboard · preferences · performance

24-Hour Clock

Pick a 24-hour or 12-hour clock in Preferences, and a dashboard that loads far less on the first visit

24-hour clock

Preferences has a new clock setting next to language and timezone (@janatjak).

  • Choose 24-hour or 12-hour, and every time in the dashboard follows it
  • Until you choose, the clock follows your language: 24-hour in French, 12-hour in English
  • Stored per user, so it follows you to any browser. PATCH /api/auth/me accepts clock_24h, and null goes back to the language default

Lighter dashboard

  • Pages and their loading skeletons load when you open them, not up front
  • Assets ship precompressed as Brotli and gzip at the highest levels and are served as they are, so the server spends no CPU compressing them
  • Platform icons load as files when drawn instead of being inlined into the JavaScript
  • The code viewer carries only the languages it highlights, and the API client validates with Zod Mini
  • Relative times no longer re-render the whole app after every read

Improvements

  • Signing in applies your saved language, timezone and clock straight away (@janatjak)
  • A sign-in right after a sign-out is no longer undone by the older session answer
minor

performance · memory · ingest · views · breaking

Lighter Under Load

A third of the memory at rest, a backlog that drains in less than half the time, and a server that says "retry later" instead of falling over. The first start after upgrading builds new indexes and takes longer on a large database

Before you upgrade. This release adds indexes to the events table, and each one is built by reading the whole table. Migrations run before the server opens its port, so the first start after the upgrade takes longer on a large database: seconds for a few GB in the page cache, a minute or two on a slow disk, longer on a big PostgreSQL. While it runs, SDKs get connection refused and retry or drop the event, as they would during any restart. Don't stop or restart the container midway, and give a health check or liveness probe enough time. On PostgreSQL, run VACUUM (ANALYZE) events; afterwards. A numeric environment variable that doesn't parse now stops startup instead of falling back to its default.

Less memory, faster digest

Measured against v0.26.0 on Linux, SQLite, the same load:

v0.26.0 v0.27.0
Memory at rest 74 MB 24 MB (−68%)
Memory at peak, ingest burst 177 MB 98 MB (−45%)
Digest after an ingest burst 1,025 ev/s 2,212 ev/s (2.2×)
Time to drain the backlog 98 s 39 s
Disk written per event, burst 214 KB 131 KB (−39%)
Ingest p99 under mixed load 13.7 ms 9.2 ms
Dashboard read p99 under mixed load 22.8 ms 12.2 ms
  • The digest commits up to 256 events per transaction through one writer connection, instead of one transaction per event
  • Freed memory goes back to the system at once, and fewer copies of each payload are kept
  • The dashboard's event queries (releases, project stats, deleting an issue) get the indexes they need, read from the index alone
  • Seven redundant indexes are gone, so every write does less work

Holds up under pressure

  • Ingest answers 503 with Retry-After while the database is unavailable, too many events are waiting (INGEST_MAX_PENDING_EVENTS) or the disk is nearly full (INGEST_MIN_FREE_DISK_MB)
  • Shutdown finishes the running digests first; the compose files give it 30 seconds
  • An event that keeps failing to store moves to dead/ instead of blocking the queue, and a database outage never sends events there
  • Events accepted but not yet digested count against quotas

Saved views and every project's issues

  • Save a page's filters, search and sort as a named view, personal or shared, and pick the view a page opens on. The Views menu is on Issues, Logs, Performance, Releases, AI Agents and the project overview
  • Issues in the workspace sidebar lists the open issues and user feedback of every project you can read, with a projects picker. Also served as Sentry's GET /api/0/organizations/{org}/issues/

Improvements

  • OIDC_ADDITIONAL_TRUSTED_AUDIENCES lets Zitadel and Keycloak logins through when their ID token carries extra audiences (@lutek)
  • Dependencies updated to their latest versions
minor

performance · ingest · security · breaking

Faster and Lighter

Ingest takes about 20 times more events per second, the issue list loads about 118 times faster, and the server uses less memory. Everyone signs in once more after upgrading

Before you upgrade. Sessions now live on the server, so everyone signs in once more after the upgrade. Webhook and Slack alerts no longer reach private or loopback addresses: if yours point at an internal Mattermost, n8n or similar, set RUSTRAK_ALERTS_ALLOW_PRIVATE_NETWORKS=on.

Faster and lighter

Measured against v0.25.0 on the same machine, SQLite, the same load:

v0.25.0 v0.26.0
Events accepted per second 234 4,720 (~20×)
Ingest latency, p50 / p99 208 / 350 ms 9 / 18 ms
Issue list, p50 1,710 ms 14.5 ms (~118×)
Transaction list throughput 4,166 req/s 10,624 req/s (2.5×)
Memory at rest 86 MB 78 MB (−10%)
Memory after dashboard reads 369 MB 210 MB (−43%)
Database size, same events 180 MB 164 MB (−9%)
  • The issue list counts users from a column filled at ingest instead of reading every event's payload
  • SQLite runs without its global memory locks, so concurrent reads stop queuing behind each other
  • Every write on the ingest path takes the same turn, so a busy server no longer answers database is locked
  • Events reach the disk with a plain fsync instead of a full drive-cache flush on macOS

Sessions on the server

  • A session is a row in the database; the cookie only carries an opaque id, stored as a SHA-256 hash
  • Signing out really ends the session, and changing your password signs out every other device
  • Sessions expire after a day idle or two weeks in total

Closer to Sentry and Relay

  • A compressed body is decoded once, with the size checked on both sides: an unsupported Content-Encoding answers 415, as Relay does
  • The legacy /api/{project_id}/store/ endpoint accepts events again
  • Attachments sent with an error event are kept, listed and downloadable on the event page
  • Users are counted by Sentry's order: id, username, email, then IP
  • per_page above 100 answers 400 with Sentry's message, and every endpoint checks credentials before reading the request

Improvements

  • Webhook and Slack alerts only reach public addresses and ignore proxy variables; RUSTRAK_ALERTS_ALLOW_PRIVATE_NETWORKS=on allows internal ones
  • Long context values on the event page wrap instead of being cut off
  • The command bar lists the same pages as the sidebar, and hides products a project has turned off
  • Forms take you to the first field that needs fixing, including errors from the server
  • An attachment or user-feedback read that fails says so, with a retry, instead of looking empty
  • Removed: /auth/register, which always refused, and the deprecated issue fields is_resolved and is_muted
  • Dependencies patched for actix-files, actix-http, actix-web and rmcp advisories
minor

access · api-tokens · dashboard · breaking

Roles and Permissions

Build your own roles from granular permissions, give them per project, and limit API tokens to what a job needs. Some API behavior changes, so read the upgrade guide first

Before you upgrade. Read Upgrade to 0.25. Everyone keeps their project access, but members need a new permission to create projects, and tokens can no longer be copied again or create other tokens. Take a backup: going back needs it.

Roles and permissions

A role is now a set of permissions you define, under Settings → Roles. See Roles & Permissions.

  • 22 resources, from issues and logs to storage and integrations, each with its own actions
  • Give a role on all projects, including future ones, or on the projects you choose
  • A member's page shows everything they hold, with custom permissions for what no role covers
  • Nobody can give a permission they do not hold or change their own access
  • Screens hide or disable what you cannot use, and the server enforces it for the API and MCP too
  • Existing project levels become editable roles. Instance admins are now shown as Owner

Scoped API tokens

  • Limit a token to chosen permissions, chosen projects and an expiry date. It never exceeds its owner
  • Tokens are stored as a SHA-256 hash, so the value only shows once, when you create it
  • Creating a token needs a signed-in session: a token cannot create another, and the create_token MCP tool is gone

Invitations

  • Send an invitation by email, through your email integration or the SMTP_* variables, and resend it later (@LiJoeAllen)
  • An invitation can carry a role and a scope, held from the moment it is accepted

Improvements

  • The login page is redesigned
  • The issue list's tab says Archived, matching the Archive button
  • GET /api/access/me says what the caller may do, and GET /api/access/catalog lists every permission
minor

dashboard · design

New Visual Identity

The dashboard gets a new look, with charcoal surfaces, Inter, Hugeicons and one design system behind every screen

A new look

Every dashboard screen now runs on Rustrak's own design system, with a quieter, flatter look.

  • Neutral charcoal surfaces in dark mode and a soft neutral light theme
  • Inter for text, Hugeicons for icons
  • Code and stack traces get their own surface, with syntax colours that follow the theme
  • Severity, status and trend colours stay consistent across every screen

Navigation

  • The project switcher moves to the top of the sidebar
  • Search in the sidebar is now a field
  • The overview gets separate scope filters, and one short note replaces the two empty session cards

Improvements

  • Form errors are announced to screen readers, and invalid fields are focused and outlined
  • Tags and column headers use sentence case
minor

server · dashboard · github

GitHub Issue Templates

GitHub issues open from an editable draft with the event, and templates set their labels, title prefix and footer

GitHub issue drafts

Creating a GitHub issue from an issue now opens a form with the issue already written, the way Sentry writes it. Edit anything before it is opened.

  • The description links back to the Rustrak issue and carries the latest event's exceptions in a code block, cut to fit GitHub's size limit
  • The assignee is the GitHub login of the Rustrak issue's assignee, when they have one
  • Clearing the description opens the issue with an empty one

Issue templates

Labels, a title prefix and a description footer that every GitHub issue Rustrak opens starts with, including the ones opened automatically.

  • Set them for the whole instance in the GitHub integration settings
  • A project can add its own in its repository settings: its labels join the instance's, and its title prefix or footer replaces the instance's
  • GitHub creates any label that does not exist yet

Fixes

  • Times fetched after the dashboard loaded no longer read as "in 3 minutes", and table cells keep their relative times current (@rookbreezy)
patch

server · dashboard · crons

Instant Navigation

Dashboard pages open on the click with their own skeleton, and cron runs that start just before their slot are no longer marked missed

Instant navigation

The dashboard used to keep the previous page on screen until the new one had its data, with no sign that the click registered. Every page now opens at once.

  • Each page shows its own loading skeleton while its data arrives
  • Changing a filter, a page or the event you are looking at keeps the current results on screen, dimmed, instead of flashing the skeleton
  • The projects and releases lists only send an out-of-range page to the last one for the page you are actually on

Fixes

  • A crontab monitor run that starts up to 10 seconds before its slot, for example from clock skew, now fills that slot instead of leaving it to be marked missed. A new environment's first run fills the first slot after it started. (@rookbreezy)

Improvements

  • Dependencies updated to their latest versions
patch

server · dashboard

Dashboard on TanStack Query

Every dashboard page now reads through one shared cache, and the Agents and Storage pages load quickly on large databases

One cache for the whole dashboard

Every page of the dashboard, from issues and events to alerts, crons, releases, logs, agents, storage and settings, now reads and writes through TanStack Query. There is one shared cache instead of a separate fetch on each page.

  • Going back to a page you just saw shows it immediately. A read stays current for 30 seconds. After that the page still shows what it had and refreshes in the background.
  • Data refreshes in the background when you come back to the browser tab
  • A change you save also refreshes every page that shows that data. A deleted record disappears from the cache straight away.
  • Signing out, or signing in as someone else, empties the cache, so one user's data is never on screen for the next
  • Every page uses the same error screen with a retry, and an expired session still sends you to login

Agents and Storage on large databases

On databases with many spans, both pages could time out. They now load quickly.

  • New partial indexes cover the Agents window, trace and conversation queries and the per-project count of standalone spans. The migrations run automatically on startup.
  • A page of conversations reads its agent names in one query instead of one per row
  • The Agents page defaults to the last 24 hours. Use ?period=all to see everything. The overview panels load after the header and the open tab.
  • Storage figures are read at startup, shown immediately and refreshed in the background. Only a finished cleanup makes a request wait for a new read.

Fixes

  • GenAI model, agent and tool names are cut to 255 bytes instead of 255 characters, so spans with long multibyte names no longer fail to insert on PostgreSQL. Conversation IDs keep their 255-character cut, so existing conversations are not split.

10 of 86 releases