Repository navigation
Features
- New setting
CONSOLE_CSP_GIT_HOSTSlets the deploy wizard reach your
Git hosts without replacing the whole Content-Security-Policy. - Why it's needed: unless a registered endpoint is selected, the wizard's
GitHub and GitLab sources call the Git host from the user's browser. The built-in policy only allows the console's own
address, so those calls fail until the host is allowed. Until now the
only way to allow one was to replace the whole policy with
CONSOLE_CSP. That copy then misses every later hardening of the
built-in policy. - Set it to a comma-separated list of
https://addresses. Each one is
added to the built-in policy. For example:
CONSOLE_CSP_GIT_HOSTS=https://api.github.com,https://gitlab.com
| Wizard Git source | Add to CONSOLE_CSP_GIT_HOSTS |
|---|---|
| GitHub.com | https://api.github.com |
| GitLab.com | https://gitlab.com |
| GitHub Enterprise | https://<your GHE host> |
| Self-hosted GitLab | https://<your GitLab host> |
| Registered GitHub/GitLab endpoint | nothing (goes through the backend) |
| Public Git URL | nothing (the backend clones it) |
- Nothing to do on upgrade: the setting is empty by default and the
built-in policy is unchanged. - If you set
CONSOLE_CSPonly to allow Git hosts, you can remove it
and list the hosts here instead. You get the built-in policy back,
including its future updates. A customCONSOLE_CSPstill wins: this
setting is then ignored and a warning is logged at startup. - How each wizard source reaches its Git host, what to do when one
fails, and the rules for entries: Deploying from GitHub and GitLab.
- The service instance wizard no longer waits on a slow broker for as
long as it takes. After 30 seconds it says how long the operation has
been running and asks whether to keep waiting, for 30 seconds up to 10
minutes, or to continue in the background. Without an answer within 30
seconds, or when you choose to, it closes, says the operation carries
on and names where to watch it (the Services list, or the
application's Services tab). The outcome is still reported wherever
you are; a failure stays on screen until dismissed. Leaving the wizard
early hands the operation off the same way. - A service instance create or update handed off by the wizard is still
reported after a page reload or a new sign-in, signing out included:
Stratos says it is still running and for how long, then how it ended.
Only the wizard's create and update are kept this way; a bind, key,
delete or bulk operation that is still running when the page reloads
is not reported afterwards. If the session ends while one runs and the
page stays open (the session expired, or you signed out in another
tab), Stratos says its status will show after you sign in. With
several tabs open, one tab reports it; another takes over if that tab
closes.
The "still running" or "carries on in the background" message is taken
down when the message saying how it ended appears. - Stratos stops following a background operation (a slow broker, a
delete Cloud Foundry runs as a job) after a set time, 60 minutes by
default, and says it is still running in Cloud Foundry rather than
that it failed or that its result is unknown. A restage or rollback is
followed to the end, since Stratos itself moves it along. Set
ASYNC_JOB_TRACKING_LIMIT_IN_MINS (Helm: console.ui.jobTrackingLimitMins)
to change it; under 1 uses 1, over 360 logs a warning, 1440 or more
uses 360. - The backend logs a "job ended" line for each background operation,
with its kind, outcome and how long it took, so an operator can set
the tracking limit from their own foundation's numbers. scripts/secrets.shcan read the openssl passphrase from the macOS
Keychain instead of prompting for it. Store it once as service
stratos-e2e-secrets, accountpassphrase, then pass
--keychain stratos-e2e-secretsor setSTRATOS_SECRETS_KEYCHAIN.
Files stay compatible with ones encrypted at the prompt, and a
missing item is an error rather than a fallback to the prompt.- Steppers no longer show a greyed-out Previous button on their first
step. Completed step headers can be clicked to go back to that step. - Every tab of an application, space, organization and Cloud Foundry page
has a CF CLI menu with the cf commands that match the tab, names
filled in. Choosing a command copies it to the clipboard; the menu never
runs anything. Its last entry, CF CLI All, opens the full command
page for that level, which is the former CLI Info page with its commands
grouped by tab. The CLI Info buttons and the application summary's
"CLI Info" link are replaced by the menu. - The CF CLI All page of a Cloud Foundry, an organization or a space shows
the commands of that level only; it used to show organization and space
commands at every level. Its rename commands are shown to every user,
since the page only shows commands.
API Keys
- An API key is created in two steps:
- a description and an expiry: 30, 90 or 365 days, a date, or Never
(90 days by default) - the secret, hidden until shown, with a copy button. Escape does
not close this step. Done closes the dialog, and the secret is not
shown again.
- a description and an expiry: 30, 90 or 365 days, a date, or Never
- API keys can expire. Keys created before this release never expire.
A request that uses an expired key is refused with 401 and "API key
expired on ". - The API Keys page has an Expires column. A banner names a key from a
day before it expires until a week after, with "Create replacement".
Last Used notes attempts that were refused. - The API Keys page shows the key count and a labelled Create API Key
button above the list, as the other list pages do, in place of the
icon-only + in the page header.
Applications
- Display, filter, and manage Cloud Foundry application labels from the applications lists and edit page.
Endpoints
- An endpoint whose last connect failed says so after a reload: its
card shows "The last error was . Reconnect to get the latest
status." and the endpoints list shows the same on hover. The reason is
kept in the browser, per user and endpoint, until a connect succeeds
or the endpoint is unregistered. - Connect and Reconnect open the same two-step Register and Connect
dialog as Register Endpoint, on the Connect step with Register marked
done. This covers the endpoint card and row menus, the home page, the
Reconnect prompt and the Cloud Foundry login prompt. Previous, the
Register step's header or the "Review" link show the registration.
Going back never unregisters the endpoint. - The Register step ends with "Unregister…" for users who can change
the endpoint, and asks before removing it. In Register Endpoint the
form then reopens with its values, so a mistake can be fixed and the
endpoint registered again. - The Register step and Edit Endpoint show each CA certificate decoded:
- subject and issuer
- validity dates
- SHA-256 fingerprint
- a badge when it has expired or expires within 30 days
- "Check trust now" tests the endpoint address and its login server
(UAA) against the certificate on the form, and says for each host
whether it is trusted or why not. - Skip SSL validation and the CA certificate are set only on the
Register step and in Edit Endpoint. The Connect step's credentials
form no longer offers them. - Unregister from the endpoint card menu asks the same question as the
Register step, with Keep it and Unregister.
Organizations and Spaces
- The org and space summary pages show "–" in their Spaces and role
tiles while the data loads, instead of "0" or "None".
Performance
- One users cache per endpoint serves the summary tiles, the Users tabs
and the role dialogs, so the CF Users tab no longer reads every user
again after the summary pages or the home page have loaded them. - A space's summary page reads the space, its app and route counts and
its SSH setting at the same time, and takes each count from one small
request. On a space with 1,500 apps and 1,500 routes the page took
6.9 s; it now takes under 1 s. - An organization's Spaces tab reuses the spaces list the organization
page has already read, instead of reading it again, and the
organization page no longer waits for that list before showing its
tabs. - An application's instance stats poll (every 5 seconds while the
Instances section is open) now asks only for the stats, instead of
fetching all of the application's data again.
Autoscaler
- The Autoscale tab has a Latest Readings box beside Instances, with
each policy metric's latest value as a number, a dot in the gauge's
status colour, and when the oldest and newest readings were taken. - The Cloud Foundry Autoscaler tile in Register Endpoint registers an
autoscaler. The Register step asks which Cloud Foundry it serves and
which of that Cloud Foundry's service brokers it is; both are picked
for the usual names. With no Cloud Foundry registered on the
address's domain, the step says to register it first. Registration
is refused, with the reason on the form, when:- the address is not on that Cloud Foundry's domain, as the
autoscaler CLI requires - the address does not answer
GET /health, so is not an autoscaler
- the address is not on that Cloud Foundry's domain, as the
- An app's Autoscale calls go to the registered autoscaler whose
service broker the app is bound to, with the user's Cloud Foundry
token, which is never sent outside that Cloud Foundry's domain. Apps
with no registered autoscaler keep the address guessed from the Cloud
Foundry's (api.replaced byautoscaler.). - The Autoscale tab names the autoscaler it uses, or shows the guessed
address with a link to register it. It warns when that autoscaler is
registered more than once, and when the app is bound to more than
one registered autoscaler. - The Endpoints page banner counts autoscalers linked to the same
service broker as duplicates, whatever their addresses. - Autoscaler endpoints cannot be connected: they use the user's Cloud
Foundry connection. - Unregistering a Cloud Foundry endpoint also unregisters the
autoscalers linked to it.
Bug Fixes
-
Deploying an application from a Git repository or folder no longer fails
when the source has nomanifest.yml. Likecf push <name>, the deploy
warns and continues with a manifest generated from the application name
entered in the wizard. Without a manifest or a name it now says so,
instead of a generic "Deploy Failed!" (#5987). -
A failed application deploy now keeps showing the specific reason
(invalid manifest, missing endpoint, ...). A generic "Deploy Failed!"
sent after it no longer replaces it. -
Route lists (an application's Routes tab, and the Cloud Foundry and
space Routes pages) show each route's full URL as a link with a copy
button, in card and table view. TCP routes copy as host:port. Route
URLs now showhttps://, matching the Visit button. -
The deploy wizard's Public tab can deploy from GitHub or GitLab without
a registered endpoint. Before, Next silently did nothing even after a
repository and branch were chosen. -
When the browser can't reach a Git host, the wizard names the host and
says how to allow it (CONSOLE_CSP_GIT_HOSTS, or registering it as an
endpoint). Before, the repository search showed no suggestions and no
reason, or said only "Git request failed". -
Owner and commit avatars from GitHub, GitLab.com and Gravatar now show
in the wizard and the application's Git tab. GitHub Enterprise and
self-hosted GitLab avatars show once that host is in
CONSOLE_CSP_GIT_HOSTS. -
bun run e2e:devnow targets the dev frontend on port 5440 instead of
4200, where nothing listens. -
Every frontend build now writes the
stratos.yamltitle, git branch,
commit and build date into the builtindex.html. Before, the build
rewrote the tracked sourceindex.htmlonce and later builds in the
same checkout kept those first values.npm run build-cfskipped the
step entirely. -
The source
index.htmlis no longer modified by builds, and setup no
longer hides it fromgit status. A checkout set up before this
change may still hold an old rewritten copy; setup says so, and
git checkout -- src/frontend/packages/core/src/index.htmlrestores
it. -
Release builds now record the release tag as the branch, so the
About page shows the tag (such asv5.5.5) rather thanHEAD. Release CI checks out
the tag, which leaves no branch; a build from a checkout with no
branch and no tag records the short commit. -
Creating a service instance with an application selected in the
wizard now binds it to that application. The binding was silently
skipped: the wizard looked for the new instance under a link name
Cloud Foundry does not use (#6001). -
Deploying an application keeps every field of its manifest. Stratos
adds aSTRATOS_PROJECTenvironment variable before the push and,
in doing so, rewrote the manifest with only the fields it knew,
dropping the rest beforecf pushsaw them (#6003). Among them:metadata(labels and annotations)processesandsidecarsreadiness-health-check-*andlog-rate-limit-per-second- the top-level
version
-
Opening an editor after a JSON editor had been shown no longer logs
"DropIntoEditorController depends on UNKNOWN service
treeViewsDndService" in the browser console. The editor's
drop-into-editor feature, which failed to start, now loads with every
editor. -
A failed write now reports the same message whether Cloud Foundry
answered at once or after a background job: Stratos's lead-in followed
by the service broker's or Cloud Foundry's own explanation, with any
HTML page it quoted reduced to its text. The Cloud Foundry error code
and title are logged to the browser console. -
Failures that showed no reason now show the one the server gave:
Where Showed Creating or editing an organization, quota or space quota [object Object]Deleting a service instance from a list, unbinding from its summary page [object Object]Binding a route service, creating or deleting a service key undefined (502)Attaching a newly created route to an application Http failure response for …Detaching a service instance's bindings unknown errorRolling back an application the error title only, or Rollback failedAny wizard step whose submit failed with an HTTP error [object Object]A detail page's error banner and header actions, adding an API key [object Object]Failed loads of an organization, space, endpoint, application or CF info [object Object] -
A service broker rejecting an operation no longer also raises the
"Cloud Foundry endpoint ... Request failed" banner: the backend now
answers a failed operation as 422 with the reasonoperation_failed
instead of 502, and the endpoint banner is kept for endpoint faults. -
Error snackbars are readable: a darker red background with a white
action and close button (the action was blue on red, 1:1 contrast). -
An operation whose result Stratos could not determine is no longer
reported as a success. Waiting for a background operation used to end
at the first failed status check (a network blip, a backend restart),
and the screen then said "created" or "deleted" regardless. Status
checks now carry on through transient errors, and an operation that
really ended without a known result says so and names where to see
the current state. -
Any Stratos instance behind a load balancer can answer for an
operation another instance started, so a status check no longer fails
when it reaches a different instance or one that restarted. -
An operation's status is shown only to the user who started it.
-
Creating a service binding or service key now reports the broker's
failure when the broker fails the operation after accepting it, as
service instances already did, in the broker's own words rather than
Cloud Foundry's internal error name. -
The service instance wizard names the instance in its messages,
failures included. It saidService instance "Service instance" created., named a renamed instance by its old name, and said "created
and bound" when binding an existing instance. A bind that fails after
the instance was created now says the instance was created. -
A bulk delete or unmap that partly fails names each item that failed
and gives the reason, and names any whose result could not be
determined. It said only how many failed. -
Deleting an application from its row menu with "Also delete this
application's routes and service bindings" reports how it ended, as a
multi-select delete does. A failure used to go unreported. -
Background operations are checked less often the longer they run:
every few seconds at first, then up to every two minutes. -
Starting, stopping, restarting, restaging or scaling an application
whose result could not be determined is no longer reported as done. -
An application action that fails in Cloud Foundry after it was
accepted (a delete, start, stop, restart or restage from the
application page) gives the reason Cloud Foundry gave, not its
internal error name such as CF-UnableToPerform. -
Deleting an application reached by its delete URL (a bookmark or a
refresh) no longer asks about, and reports on, org "?" / space "?"
when Confirm is pressed before their names have loaded. -
A setting given a value that does not parse (text for a number, say)
still stops the backend at startup, and the error now names the
setting. -
Screen readers read out pop-up messages, errors first, and name their
close button. They were silent before. -
The service instance wizard's "still running" question takes focus
when it appears, so a screen reader reads it out and it can be
answered from the keyboard. -
ASYNC_JOB_TRACKING_LIMIT_IN_MINS=0 follows jobs for 1 minute, as for
any value under 1, rather than the 60-minute default it got by being
mistaken for unset. -
The backend no longer keeps an unfinished operation in memory forever
once nobody is checking on it: after an hour without a status check it
is dropped. An operation someone keeps checking is kept however long it
takes. -
Signing in to a console deployed on Cloud Foundry works again for a
browser that still holds the cookie of an ended session (signed out in
another tab, cleaned up after expiry, or dropped by the 5.5.5 sessions
table migration). Since 5.5.5 the sign-in answered 200 but saved nothing,
so the next request was refused and the user was sent back to the login
page until they cleared the console's cookies. A cookie with no session
behind it now starts a new session, as the per-database stores did
before 5.5.5. -
Signing in now always starts a new session. On a console deployed to
Cloud Foundry a sign-in kept the session the browser arrived with, so
any copy of that earlier session cookie was signed in too. -
With no endpoints registered, the Endpoints page now shows the
Register Endpoint button and the "no registered endpoints" message,
and Home shows the message. Both pages were blank before. -
Fixed pages that waited forever:
- The UAA setup wizard checks the credentials in step 1 and shows an
error instead of an endless "Saving configuration" overlay. The local
account wizard gets the same overlay fix. - Kubernetes config import no longer stalls at the first Connect, and
moves on past a cluster that fails to register or connect. - The role picker in Add User and Manage Roles no longer shows
"Loading organizations..." forever; a failure shows Retry. - The navigation progress bar no longer stays at 95% after a page
fails to load. - After a failed session check near expiry, navigation no longer stops
working until a reload.
- The UAA setup wizard checks the credentials in step 1 and shows an
-
Fixed views that did not show or showed out-of-date content:
- SSH sessions show their progress and server errors.
- Deploy keeps its completion notice after you leave the page and keeps
every log line. A second deploy from the same wizard is labelled
"Deploy again". - The application Autoscale tab is back. It shows the reason when the
autoscaler refuses a request, and Create Policy only when there is no
policy. Its scaling history and metric charts load again; every
request for them had failed with an HTTP 400 error. - The git tab no longer stays on "Retrieving Git Details" after a
failed lookup. - Service broker cards show the Space row, and the broker URL as the
link text. - Home page application cards follow start, stop and delete.
- Helm release pages no longer show earlier data after a failed reload,
and the Helm catalog sidebar shows. - The Kubernetes resource viewer shows its header, and Delete works.
- Metrics charts with a time range selector render.
- An active user is no longer warned and logged out as idle after a
short session check failure.
-
Stopped background work that kept running after you left a page:
- the application stats poll (one request per Cloud Foundry every 30
seconds, until a reload); - the metrics polls;
- watches on endpoints, permissions and quotas.
- the application stats poll (one request per Cloud Foundry every 30
-
Disconnecting a Kubernetes endpoint clears its cached data and errors.
-
Removed the
users/removepage, which nothing linked to any more. The
users tabs remove users directly. -
The App Autoscaler pages work with autoscaler 14.x and 15.x:
- The policy editor opens, shows the application's name, and lets
you continue from the instance limits step. - The scaling history, metric charts and gauges load and draw.
- Saving a policy keeps its custom metrics settings, offers only
timezones the autoscaler accepts, and adds the cpuutil, disk and
diskutil metrics with their limits. - Changes made while a saved policy is still loading, or before going
back to the first step, are no longer lost. - Editing a rule with a percentage step shows the step instead of NaN.
- A schedule's initial instance count can no longer exceed its maximum.
- A refused change shows the autoscaler's reason.
- After a policy is detached, the tab shows the default policy the
autoscaler applies, if there is one. - The scaling history shows more than the newest 50 events.
- The policy editor opens, shows the application's name, and lets
-
Removed Manage Credentials from the Autoscale tab; it failed on
autoscaler 14.1.5 and later, which no longer have that API.
Applications get custom metric credentials from their service binding. -
Service instance lists and an instance's own page now follow an
operation that has not ended: while a create, update or delete is
queued or in progress, they read the instance again every 10
seconds, so its status changes to the result without a page reload.
They stop when the operation ends or the page closes. Before, a new
instance's status stayed at "initial" until the page was reloaded. -
Creating or updating a service instance now marks the services the
console caches for each endpoint as stale, so their next read
fetches them again. -
Service instance lists now say which operation a Last Operation pill
is about: "Create in progress", "Update failed", "Delete succeeded"
instead of the bare state. A failed operation's hover text adds the
broker's reason, with any markup removed. A queued operation shows
as in progress. -
A service instance's page shows "Last operation: Update failed" and,
only for a failure, the broker's reason as text on the page, so
screen reader and keyboard users get it too. -
A failed create in the service instance wizard reports the broker's
reason without its markup. -
Deleting a user-provided service instance reported "Delete failed:
deleting service instance: no Location header on async delete
response", although CF had deleted it. CF answers that delete with
204 No Content and no job; the CF API client now treats that as
done. This failed in every release since 5.0.0. -
The user-provided service instance wizard stayed on its last step
after Finish, with no message, although CF had created or updated
the instance. It now closes once CF has the change. -
The user-provided service instance form shows its Tags field, gives
each field's own error ("Invalid URL", "A service instance with this
name already exists...") instead of a generic one, and accepts
syslog://andsyslog-tls://drain URLs, as CF does. -
The user-provided service instance form accepted an
http://route
service URL, which CF then rejected; it now requireshttps, as CF
does. -
15 forms showed a generic line ("Invalid value", "Please enter a
valid format") instead of their own error text, such as a taken name
or a bad API address. They now show their own text, and screen
readers announce it as the field's error. -
The first navigation after loading or refreshing a page rebuilt the
whole page, side navigation and top bar included, instead of only the
part that changed. -
Helm chart catalog and chart values requests go through the app's
standard request handling again. Since the Angular upgrade they had
used a separate client that skipped it, so their responses did not
refresh the app's XSRF token and they were missing from the API call
counts and timings. -
When a service broker or a proxy answers with an HTML error page,
Stratos now shows the page's heading and message, one block per line,
instead of every word of the page run together ("502 Bad Gateway 502
Bad Gateway nginx"). Error snackbars, the stepper and a service
instance's "Last operation" reason show the heading in bold. Plain-text
errors read as before. -
An e2e run on other ports started its own servers there, but its
logins and API calls still went tohttps://localhost:5540, so it
logged into whatever console served 5540, such as another e2e run
or a dev server. Every fixture now takes the console URL from the
same place as the Playwright config. The e2e ports are now set with
E2E_FRONTEND_PORTandE2E_BACKEND_PORT, apart from the dev
servers'FRONTEND_PORTandBACKEND_PORT; an e2e run still reads
those old names for one release. -
When Cloud Foundry rejects a request as invalid (for example a list
with an unsupported sort order), Stratos now returns Cloud Foundry's
own 4xx status. The page that made the request reports the error, and
the "Cloud Foundry endpoint ... Request failed." banner no longer
appears for an endpoint that is working. -
When adding a user to a Cloud Foundry endpoint fails, for example
because the signed-in user is not a Cloud Foundry admin, only the Add
Users dialog's failure message appears. The endpoint "Request failed."
banner no longer appears with it. A failed user lookup is no longer
reported as "user not found". -
Several pop-up messages can be open at the bottom of the page at
once, as intended, but each one sat exactly on top of the one
before: the older message peeked out behind the newer one and could
catch a click meant for its close button. Stacked messages are now
usable as tabs. They share one box: when there is more than one,
numbered tabs along its top switch between them, and a new message
is shown as it arrives. Each message has one "Dismiss" button, in
place of a "Dismiss" or "Close" button and a "×" that did the same
thing; it closes just that message, and the other tabs keep their
numbers. A "Close all" button beside the tabs closes them all. The
same message arriving again straight after itself is counted
instead of getting a new tab: the start of its text shows how many
times it has been shown. The tabs work with the arrow keys and
screen readers, and a closed message is also cleared from what
screen readers can find. -
The application log stream tab now stops after three failed
connection attempts instead of opening a fourth connection and
closing it at once, and it starts counting again after a successful
connection. Before, a stream that connected and later dropped could
stop reconnecting. -
The application log stream tab now says when it is retrying the
connection (with the attempt number) and when it has given up,
instead of showing "Connecting to log stream..." for good. -
The metrics views never appeared, even with a metrics endpoint
registered and connected: the Metrics tab on an application and on a
Kubernetes node, the Cells tab on a Cloud Foundry, and the link to a
pod's metrics in the Kubernetes resource viewer. The check read a
flag that nothing has set since May 2026; it now reads the
endpoint's metrics link from the backend. -
When the Metrics tab opened, its charts asked for the time window
before the window was set, so Prometheus rejected the first request
and the charts said "No results found" until the next refresh ten
seconds later. The first request now carries the window. -
When a Cloud Foundry endpoint keeps answering with a server error
(5xx), the endpoint banner now says the endpoint is unreachable or
down and shows Cloud Foundry's status code, for example "HTTP 503: The
endpoint is unreachable or down.". It used to read "Request failed."
with no code. -
Cancel in a stepper (Deploy Application, Manage Users) runs the
step's clean-up: the upload step's deployer closes, and the error
message goes. -
Checkboxes, slide toggles and button toggles show the value and the
enabled or disabled state a form gives them. They could keep their old
look until clicked: a checkbox stayed greyed after its form was
enabled, and on the profile page picking Dark left Light highlighted
too. -
A form field whose label comes from the page after it is shown, such
as the Connect dialog's username and password, shows that label; the
fields were blank lines. -
Card titles wrap between words, not mid-word, and a second line is no
longer cut off. Hovering a card's title shows its full name. -
Debug logs no longer show endpoints' client secrets; the metrics
plugin printed every endpoint with its secret on each page load. -
With SSO set to skip the login page (
nosplash), the login page says
"Signing in…" while the browser goes to the login server; it showed
only the logo and looked stuck. -
With DIAGNOSTICS_ENABLED set, SSO login opens the login server's page
again. The console answered the browser's login request with an empty
page instead of sending it on, and other replies with no body, such as
redirects and 204s, lost their status the same way. -
Items per page shows a list's real page size. API Keys, which pages
by 12, showed 5. -
A list row's ⋮ menu closes on a click outside it. Before, only
Escape closed it. -
Errors while backing up or restoring endpoints, connecting, registering or
updating an endpoint, reading service key credentials and loading metrics
now read the same way as other upstream errors. An HTML error page from a
proxy or router shows its heading and message instead of its markup. A
Cloud Foundry error shows its reason instead ofundefined (422), and a
response with no body showsHTTP 502 Bad Gatewayinstead ofnull (502). -
A failed backup or restore that got no response from the server now says
so. The reason was being dropped. -
Registering an endpoint no longer says
undefinedwhen a 403 has no
Stratos error in its body, such as a page from a proxy. -
A failed action now gives the reason Cloud Foundry, Jetstream or a proxy
gave, instead ofHttp failure response for <url>: <status>. This covers
unmapping and deleting a route, saving and deleting a variable, terminating
an instance and changing the instance count of an application; start, stop,
restart, restage and delete in the application lists; delete in the
organization and space lists and on their summary pages; the last step of
Create Application; loading, creating and deleting API keys; and the URI
limit probe on the diagnostics page. -
A role change request that failed as a whole no longer reads
Failed to apply role changes: [object Object]. -
A failed Helm install, upgrade or delete now gives Jetstream's reason, such
as a release name that is already in use. -
The endpoint errors page, API keys and the setup wizard show an HTML error
page from a proxy as its heading and message instead of its markup. -
Manage Roles opened from an organization or a space no longer leaves the
organization name blank on the Confirm step. It readrole changes to organization '', and the Target column readOrg:with no name. -
An unknown address under a section such as Applications or Endpoints
(for example a mistyped tab name in an application's address) now shows
"Page not found". It used to show a blank page and keep the browser busy. -
The Delete Application wizard now opens on its first step. It used to
open on the last step, Confirm, with the Routes and Service Instances
steps appearing in front of it a moment later, so it was easy to confirm
without choosing which routes and service instances to remove. -
The CLI Info pages give commands that work when pasted. Scaling an
application's memory is nowcf scale "app" -m [N]M; it was shown as
-i -m [N], where-ineeds a value of its own. Organization and space
names are quoted in every command, so a name with a space in it no longer
breaks the command. -
The Getting Started steps on the CLI Info pages follow the endpoint:
cf apicarries--skip-ssl-validationwhen the endpoint is registered
with certificate validation skipped, andcf login --ssois offered when
the endpoint allows single sign-on. The sample login output shows the
Email:andPassword:prompts of the current CLI, and its lines are no
longer indented. -
A CLI Info page opened by its address, or reloaded, fills in the
organization and space names. It used to show[organization name]and
[space name]in their place. -
The CLI Info page of a Cloud Foundry or an organization no longer requests
an organization or a space namednull. -
The application's CLI Info page has a close button again. On the Cloud
Foundry, organization and space pages the close button is labelled "Back"
instead of "Back to App", and on a page opened by its address it returns
to the organization or space instead of "Page not found". -
The CLI Info pages show
cf space-userswith the organization as well as
the space. The CLI needs both, so the command as shown before failed with
"Incorrect Usage". -
The date beside each recent application is easier to read. It was a
faded grey at about 3:1 contrast; it now uses the theme's muted text
color, which meets the 4.5:1 minimum of WCAG 1.4.3 in light and dark
mode. -
Secondary text that was drawn faded now uses the theme's muted text
color, which meets the 4.5:1 minimum of WCAG 1.4.3 in light and dark
mode. This covers the profile settings descriptions, the type line of
each recent entity, the committed date range of a metrics chart, the
"connect later" hint when creating an endpoint, and the "no
applications" message on the home page. They were at 2:1 to 4:1 in light
mode. -
Colors set in
company-config.json(theme,navigation,layout) stay
in effect after the theme mode changes and after login; they used to be
replaced by the built-in colors until the page was reloaded. The brand
colors (theme) apply in dark mode too; the navigation and layout colors,
which the file has no dark variants of, apply in light mode only. -
The browser console no longer logs the whole theme on every theme change
and mode switch. Setdebug.logThemeChangesincompany-config.jsonto
turn those logs on. -
The browser console says once, at startup, which company config was
applied:company-config.json, the environment file, or the built-in
defaults. -
The first Tab stop on every page is a "Skip to main content" link, shown
while focused, that moves focus to the page content; the content is a
mainlandmark. -
The user menu button says whether its panel is open, and Escape closes the
panel and returns focus to the button. -
The header of a home-page endpoint card is a link, so the keyboard reaches
the endpoint's page as the mouse does. -
The CLI Info download step is the link text, instead of "see here".
-
The table header above the favorite star and row-actions column names the
column for screen readers. -
Recent Activity rows are links, so the keyboard reaches them, and show an
outline when focused. -
The application state icon is an image named after the state; beside its
text label it is not read twice. -
The About, Metrics and Invite Users page titles, the About section labels
and every meta card title are headings. -
The Instances card's icon buttons and the marketplace documentation links
have names; the icon link that repeats a text link is one tab stop fewer. -
Every element the keyboard reaches shows a 2px focus outline unless it
draws its own focus ring; the header and side nav draw it in their text
color. The light/dark toggle, the second-level menu items, the side nav
logo and logout buttons, the dropdown triggers and clickable cards no
longer hide keyboard focus. -
Text throughout the console now meets the 4.5:1 contrast minimum of
WCAG 1.4.3 in light and dark mode. The default blue is a deeper
#2563eb, so links in tables, active tabs and primary buttons read
clearly; dark mode keeps its lighter blue and draws dark text on
primary buttons. Placeholder text in inputs and dropdowns, text in the
status colors (such as an application's state), the About page's
admin badge, and the dimmed labels on the quota pages are darker.
Operators who configure their own brand colors are unaffected. -
The page tabs on application, space, organization and other detail
pages now work from the keyboard every time. Each tab used to take
two Tab stops, and the first one, where the focus outline showed,
ignored Enter; those stops also came before the rest of the page.
Each tab is now one stop in page order, a hidden tab leaves no empty
list item, and the current tab is announced as the current page. -
Status icons keep their colors with enough contrast. The warning
triangle had turned brown and the running circle dark green when
text in those colors was darkened; icons now use their own amber,
green, red, blue and gray shades at 3:1 or more in light and dark
mode. The application Visit button draws a white icon on blue again
instead of a dark one, and a dark icon on the lighter dark-mode blue. -
A card's name tooltip appears only when the name is cut off.
-
The home page's recently updated applications show each app's real
state. A started app showed a green circle, read as "Deployed, No
Instances", which turned gray a few seconds later once the console
learned its instance count. The list now carries each app's instance
count and loads its instance stats, so a running app shows green;
while the stats load, the row shows a loading placeholder. -
An application whose state is not yet known shows a hollow circle,
so it no longer differs from a running application's filled circle
by color alone. -
Tab still indents in the code editors, and each editor now names the
key that makes Tab move focus out of it, Ctrl+Shift+M on every
platform. Each editor also has a name for screen readers. -
The SSH terminal keeps Tab for the shell; Ctrl+Shift+M makes Tab move
focus instead, as shown under the terminal. After the session ends,
Tab leaves the terminal and "Disconnected" is announced. -
Leaving an application's SSH session with
exitno longer shows
"Error connecting to web socket". -
On narrow screens the page tabs open from their header by keyboard as
well as by hover; Escape closes them. -
The main side navigation has a name for screen readers ("Main").
-
Screen readers no longer read icon names ("home", "cloud_foundry")
as part of the side nav and page tab names. -
Screen readers no longer read icon names ("mode_edit", "cloud_done")
aloud. Icons that only decorate are hidden from them, and icons that
are the only sign of something, such as a status, a warning or a
report grade, are announced by what they mean. -
Every button and link that shows only an icon has a spoken name. The
error banner's dismiss icon is now a button that keyboard users can
reach. -
Text on success, warning and danger buttons meets the contrast
minimum, including with a company's own status colors. -
Error and warning messages (the endpoint error banner, list load
errors, dialogs, the deploy stages) meet the contrast minimum in light
and dark mode. -
Screen readers hear when the deploy page's Organization and Space
selects are not yet active, instead of being offered them as working
controls. -
The slide toggles on the profile page and in the org, space and
application editors are one switch each (the checkbox itself), not a
focusable box around a hidden checkbox. -
The completed step's check mark in wizards meets the contrast minimum.
-
The Copy UAA tokens and Recent activity buttons say whether their
panel is open, and Escape closes it and returns focus to the button. -
Notifications and the user menu's About, Profile and API Keys are
links, and the Notifications name includes the unread count. -
Every page has its own document title, such as "Summary - my-app -
Stratos" or "Create Space - norm-lab - Stratos", so browser tabs,
history and screen readers name the page instead of all reading
"Stratos". The operator's display name still ends the title. -
The Cloud Foundry summary card shows the foundation's description
and CF API version again; since 5.0.0 it showed "-" and an empty
version. -
With user endpoints on, a user's own endpoint now hides the system
endpoint at the same address even when the two were registered with a
different host case or an explicit default port, as connect and login
already treated them. -
Disconnecting the auto-registered Cloud Foundry now stays disconnected
at the next login when its registered address differs from
AUTO_REG_CF_URLonly in host case, default port or a trailing slash.
Accessibility
- Checkboxes and slide toggles placed without an id or name no longer
renderid="undefined"andname="undefined". Several on one form
(adding routes, quota definitions) shared that id, and assistive
technology relies on ids being unique to tie a control to its label
and error text. - Text fields that were named only by their placeholder now have a
name of their own. A placeholder disappears as soon as you type (and
a date or time field never shows it), so a screen reader had nothing
to announce and a sighted user had nothing left to read. Fields with
room for it get a visible label above the box; fields set inline in
a sentence, a toolbar or a list get a name screen readers announce:- Deploy Application, Options step: every field, and the Domain,
Stack and Health Check Type drop-downs (visible labels). - Configure User Invite Client: Client ID and Client Secret (visible
labels). - Autoscaler policy editor: the recurring and specific date
schedules' time, date and instance count fields (visible labels),
the custom From and To dates, and the scaling rule fields. The
Repeat On label now names the weekday or month-day list. - Install Chart: Name and Namespace (visible labels) and the cluster
drop-down. - Service Keys: the new key's name.
- Manage Users and Add User roles: the organization search and each
organization's space filter, which names its organization. - Metrics range selector: the start and end date and time fields,
grouped as Start and End. - File pickers: the path field.
- Forms built from a service's JSON schema: the key and value of each
map row.
- Deploy Application, Options step: every field, and the Domain,
- Controls that had no name at all now have one, taken from the
caption, prompt or column a sighted user sees beside them:- Lists: each row's select checkbox or radio (named after the row),
the select-all checkbox, the card view's Sort drop-down, the
Items per page drop-down, and the range filter's type, comparison
and value fields (named after the filtered column). - Confirmation dialogs that ask you to type a name: the field is
labelled by its "Please type ... to continue." prompt. - Register Endpoint and Edit Endpoint: the CA certificate box is
labelled by its "Specify CA Certificate:" caption. - Forms built from a service's JSON schema: yes/no checkboxes, the
JSON fallback box and the oneOf choice are labelled by the field's
label; array and tuple items and a oneOf value are named after the
field. - Kubernetes endpoint credentials: the certificate, key and token
boxes and the kubeconfig, certificate and GKE credentials file
pickers are named by their captions. - Import Kubeconfig: the box for renaming a cluster.
- Lists: each row's select checkbox or radio (named after the row),
- Drop-downs built on the shared select had no name at all: a screen
reader announced only the chosen value. Each is now named after its
placeholder, or where it has none after the caption or column beside
it: Backup Endpoints' Connection Details, the metrics Time window,
and Import Kubeconfig's Type and User. - Create Service: the Service, Plan and Bind App drop-downs had
captions that never appeared on screen. They now show "Choose
Service", "Choose Service Plan" and "Choose App" inside the empty
drop-down, which is also the name a screen reader announces. - Tooltips show for keyboard users: tabbing to a control with a tooltip
shows it after the usual delay, and moving focus away hides it.
Screen readers read the tooltip as the control's description. - A tooltip can be dismissed with Escape without moving focus or the
pointer, and the pointer can move onto a tooltip to read it without
it closing. In a dialog, the first Escape closes the tooltip and the
next one closes the dialog. - Tooltips on text, icons and other elements that cannot take keyboard
focus are still shown on hover only. - The Remove control beside each user name in Add User, Invite Users
and Manage Users is now a button, so keyboard and screen reader users
can reach and use it. - The Register Endpoint and Connect dialogs are announced as dialogs,
named by their titles. Opening Register Endpoint moves focus into it,
Tab stays inside, and closing it returns focus to the button that
opened it. - Icon buttons are announced by name (Back, Close, Refresh, First page,
Next page, Actions, Help on connecting, Backup/Restore Endpoints)
instead of their icon text, and menu items and dialog titles no
longer include icon names. - Endpoint cards can be opened from the keyboard through their title.
- The github.com and GitHub Enterprise choices on the Register step are
one radio group, so arrow keys move between them. - Select lists announce their options with their selected and disabled
states, and the control is linked to its list. - The closed side panel can no longer be reached with Tab or announced
as a button. - Escape closes the Connect and Create an API Key dialogs, and answers
"Yes" to "Are you still there?"; these dialogs ignored it. - The favorite star on endpoint and other cards is announced as "Add
to favorites" or "Remove from favorites", as in lists; it read the
opposite of its state, followed by the icon name "star_border". - Card action menus announce their entries as menu items; the menu
was announced with none. - Copy-to-clipboard buttons tell a screen reader "Copied to
clipboard" or "Copy failed"; only the on-screen tick or icon did. - On the Register and Edit Endpoint forms, the fields of a collapsed
Advanced Options or CA certificate section are out of the tab order
and the accessibility tree. - The theme button's spoken name now starts with the word it shows
(Light, Dark or System), so speech users can say what they see. - Checkboxes no longer hold a hidden second checkbox that screen
readers could land on. - The log viewer (application log stream and others) can be scrolled
from the keyboard: Tab reaches it and the arrow keys scroll it. - Pages that scroll but have no buttons or links (build packs, feature
flags, quotas) can be scrolled from the keyboard: Tab reaches the
page content and the arrow keys scroll it. - After a row action that opens a dialog (such as Terminate on an
application instance), closing the dialog puts focus back on the
row's actions button instead of the top of the page. - The copy button beside a detail such as the application GUID now
tells screen reader users whether the copy worked, and no longer
shows a failed copy as copied. - After a command is copied from the CF CLI menu, or a token from the
UAA tokens menu, focus goes back to the menu button instead of the
top of the page. - With the Instances section of an application's summary open, the
page no longer draws itself again every few seconds, which closed
the section and sent focus to the top of the page. - Refresh, the list pager, Clear in a list's filter popup, Clear filters
and sort, the bulk actions, and the instance count buttons keep focus
while they are off. They turned themselves off when pressed and sent
focus to the top of the page. - Closing a list's filter popup with Close or Escape puts focus back on
the button that opened it. - Clear in a list's bulk-action bar puts focus on the first selection
checkbox. - Cancel and Save in the instance count editor put focus back on Edit
instance count. - Dismiss on an endpoint error message puts focus on the main content.
- Copy to clipboard keeps focus after a press. The button was taken
out of the page while the result showed, and focus went to the top
of the page. The result now shows on the button. - A link to the page already shown keeps focus on it: the Status and
Uptime cards on an application's summary, and the breadcrumb and
Summary tab on a space's summary. The page was built again, and
focus went to the top of the page. Pressing the current tab or menu
item no longer redraws the page; use Refresh to load new data. - The theme toggle says "Dark mode activated" or "Light mode activated"
only when the theme is changed, not on every page change. - The keyboard focus outline is no longer cut off where a control
touches the window edge or sits in a box that clips: it is drawn
inside these controls:- the side nav links
- the app tabs (Summary, Log Stream and the others)
- the Instances bar on an application's summary
- the recent app names on the home page
- The count cards on the home page (Applications, Orgs, Routes) show
keyboard focus; before, they showed none. - A tooltip goes away at once on Escape, and when the pointer moves off
its control. Before, it stayed on screen until something else redrew
the page. - Text and colors reach 7:1. The muted grey is darker (#475569), and the
blue of links and buttons is darker (#1e40af). In dark mode, the muted
text (#cbd5e1) and the blues (#93c5fd, #7dd3fc) are lighter. - The edge of a field (the line under a form field, the border of a
list's filter and select boxes) reaches 3:1, in light and dark mode. - A focused form field has an outline round the whole field. A faded
control, such as a paging button with nowhere to go, shows its focus
outline at full strength. - Headings: every page title is a heading, a list's title ("Total
Applications: 4") is a heading, and a form step has a heading for
screen readers. The page header is a header landmark. - Screen readers say how many results a filter leaves, and the new
desired count after an app is scaled. - Links with the same text are told apart for screen readers: an app's
endpoint, org and space in the Applications list, and a page tab and
the side nav link with the same name. - The cards on an application's summary are named groups.
- Links stacked in one list cell are 24 px tall.
- CF, SSH, GUID, URL, EKS and AKS show their expansion on hover and to
screen readers. - The user menu button's name no longer includes the avatar initials,
and the Row actions button no longer claims to open a menu. - The user menu button's name starts with the initials its avatar
shows ("NA, user menu"), so a person using speech input can say what
they see. This replaces an earlier entry that said the initials were
taken out of the name; the name never had them. - The count cards on a space's summary show a focus outline.
- App names on a space's summary and in recent apps use the theme's
link color, so they reach 7:1 in dark mode too. - A screen reader says which column sorts a list's table, and which
way (aria-sort on the sorted column's header).
API Keys
- The docs explain API keys: who can create them (API_KEYS_ENABLED),
the "Authentication: Bearer" header they are sent in and the routes
that accept them. - The "Create an API key" prompt on an empty API Keys page points at
the toolbar button, below the page header, instead of overlapping "You
have no API keys"; the same prompt on an empty Home page is placed the
same way.
Applications
- An application opened from a Cloud Foundry's Applications list shows
the endpoint in its breadcrumb trail again, before Applications.
Two breadcrumb definitions shared one key and the trail showed only
Applications. - Creating or editing an application now shows the icon beside the
name while it is checked (busy, available or taken). It was never
displayed. - The Add Variable and Edit Variable dialog stays open when the save
fails and shows the error, so what was typed is not lost. - The application Variables tab no longer shows briefly to users who
may not view environment variables. - Deploy Application reuses the Cloud Foundry feature flags already
loaded to decide whether Docker images are allowed, instead of
requesting them again. - The new application route step shows the domains of the organization
picked last. A slower answer for an earlier organization could
replace them.
- Adding, renaming or deleting an environment variable that Cloud
Foundry refuses now says so, with Cloud Foundry's reason. Before,
the dialog closed as if the change had been saved, and the variable
list simply did not show it. - Name sort keeps the numbers in a name apart, so app-1-11 sorts
before app-2-1. Before, the numbers ran together (111 and 21).
Cloud Foundry
- An event's detail dialog has room around its contents; it ran to
the dialog's edges. - An event's data keeps its lines whole, scrolling sideways when one
is long, and wraps to show everything on hover or keyboard focus; it
broke values such as IDs in the middle. - The recently updated applications card on the Cloud Foundry summary
shows the newest apps as soon as the summary loads, instead of
"There are no applications" until the full app list arrives. - "View organization's quota" on the space summary card navigates once
per click; it used to navigate again whenever the quota reloaded. - The space page and CLI Info breadcrumbs follow an organization or
space rename instead of keeping the first name. - In the table view of the organization and space quota lists, a
quota's name links to its page again, as the card title does. - The Cloud Foundry name in a service card breadcrumb appears or goes
away when a second Cloud Foundry is connected or disconnected,
without reloading the page.
Endpoints
- A failed endpoint connect says why. A rejected password or token
keeps its 401 or 403 and says the credentials were rejected; an
endpoint that does not answer, answers 5xx or fails TLS reports 502
and what failed; the endpoint's own message is shown (a Kubernetes
Status, a GitHub or GitLab error). They all read "Could not connect to
the endpoint" before. - The endpoint admin role is granted only by the
stratos.endpointadmin
scope itself. Any scope containing that text, such as
stratos.endpointadmin_read, let a user register, edit and
unregister endpoints. - Edit, Unregister and Connect are offered only to users the backend
allows to use them, and the Edit page sends other users back to the
endpoints list. Every user saw all three on every endpoint. - Edit Endpoint shows the stored CA certificate. It showed none, and
saving the form removed the certificate. - The Register Endpoint dialog closes without reloading the page under
it (which left the page unable to scroll), and shows one Cancel and
one "registered" message. - Pressing Enter in the Connect dialog connects, once.
- A request to an endpoint that does not answer in time reports 504
"Request timed out" for that endpoint instead of leaving it out of the
answer, or a 500 on a passthrough. - Requests to an endpoint with a CA certificate reuse their
connections, and write and read requests get the right timeouts (they
were swapped). - In Register Endpoint, the Connect step shows the credentials and
endpoint options only while "Connect to now" is ticked, starts
unticked for each registration, and keeps Previous and Connect in view
when the step is taller than the dialog. - "Successfully registered ''" stays on screen for 7 seconds; the
endpoints page cleared it as soon as the endpoint list updated, and it
was gone before it could be read. - When two endpoints of one type share a URL, the endpoints page says so
above Total Endpoints, as the Cloud Foundry pages do, in place of a
message after registering that had to be dismissed. A Metrics endpoint
beside the Cloud Foundry it watches is not counted. - Logging in no longer reports "Couldn't check credentials" when the
login worked but connecting the auto-registered Cloud Foundry endpoint
failed. The login completes and the failure is logged. - With an endpoint of another type, such as Metrics, registered at the
auto-registered Cloud Foundry API's URL, logging in connects the
Cloud Foundry endpoint rather than trying the other one with Cloud
Foundry credentials. - With user endpoints enabled, a Metrics endpoint at a Cloud Foundry
API's URL no longer hides that Cloud Foundry endpoint. A user's own
endpoint now hides only a system endpoint of the same type at the
same URL. - With user endpoints enabled, connecting to an endpoint no longer
disconnects the user's other connections at the same URL. When a
personal endpoint is one of the pair, the connect's one message names
the other connections of the same type and offers "Disconnect
others"; dismissing it keeps them. Two system endpoints at one URL are
not reported. - After registering an endpoint, the endpoints list stays on screen
while it refreshes instead of briefly showing "Loading endpoints". - The Connect form labels the username and password fields of
endpoint types that give no labels of their own, such as Metrics. - The browser no longer fills the console login into an endpoint's
Connect credentials, or into the Client ID and Client Secret under
Advanced Options when registering or editing an endpoint, where it
was saved and made every connect fail with "Bad credentials". The
user-invite client, backup and restore passwords and setup wizards
are marked the same way. - Only Stratos admins are offered "Share this endpoint connection",
which the server refuses for other users. - A failed Reconnect keeps the Connect dialog open with its error,
instead of closing as if it had connected. - The Register dialog's connect step names what to enter for a
single-token login, such as "Personal Access Token", and its help
button no longer covers the title. - Opening help from the Register dialog no longer hides the dialog's
Connect and close buttons, and help that cannot be loaded says so. - The Artifact Hub registration text no longer shows backticks around
"Register". - When a Cloud Foundry login server answers "Bad credentials", Connect
says to check the username and password and the endpoint's Client ID
and Client Secret, since the server gives that answer for either. A
client that cannot log in with a username and password is named, with
the suggestion of "Allow SSO login" or Client ID "cf". Both read only
as a wrong password before. - A Cloud Foundry endpoint registered or edited with the console's own
Client ID and no Client Secret, as an endpoint's review shows the ID
or a browser fills it in, connects like one registered with Client ID
left empty. It was saved without the secret, and every connect failed
with "Bad credentials". - With several Cloud Foundry endpoints at the auto-registered URL,
logging in connects one that uses the console's own client, and no
longer gives the console's login to an endpoint on another client.
That endpoint showed Connected, but renewing its token failed with
"Authentication expired", and the failure was recorded against the
console's own login. - The Cloud Foundry endpoint docs say which client a blank Client ID
uses (the console'sCF_CLIENT, not alwayscf) and what a client
needs for password and for SSO connects, and the SSO guide explains
"Invalid redirect" when connecting an endpoint. - Register Endpoint shows the form again for Cloud Foundry, Metrics,
GitHub, Helm and the other types that use the common form. After
picking the type, the dialog showed only its header and footer. - Reconnecting an endpoint clears its Expired mark and closes its
re-authentication banners at once, instead of on the next page that
calls Cloud Foundry. - An endpoint update through the API that left out
ca_certremoved
the endpoint's stored CA certificate. Leaving the field out now keeps
the certificate; an emptyca_certstill clears it. - Endpoint addresses written differently are now treated as the same
address: host case, a default port (:443,:80), a repeated or
trailing/, and a query or fragment no longer make them differ.
Before, the login hook did not connect the Cloud Foundry endpoint
whenCF_API_URLwas written differently from its registered
address, and connect, the shared-URL banner, the kubeconfig import
and Git registration missed those duplicates. Stored and displayed
addresses are unchanged. - Registering a Cloud Foundry endpoint at an address that is already
registered now shows the "already registered" warning. It was never
displayed. - With several admin Cloud Foundry endpoints at the auto-registered
URL, logging in reconnects the one the user is connected to. With
none, it connects the auto-registered endpoint, named by
AUTO_REG_CF_NAMEor "Cloud Foundry"; before, it connected whichever
one the database returned first. A user who disconnected it is still
left disconnected. - The endpoints page reloads the endpoint list before each 30-second
health check. A connection dropped elsewhere, such as in another tab
or by an admin, stayed listed as Connected, and every check of it was
refused. - Registering an endpoint of a type the server does not know, or with
no type, says "This Stratos server cannot register endpoints of type
"" yet" instead of "Internal Server Error (500)". - The Cloud Foundry Autoscaler tile in Register Endpoint has the
Autoscale icon and comes after Cloud Foundry; it had the Cloud
Foundry icon and came first, so it was picked by mistake. - Icons on the Register Endpoint and Backup/Restore Endpoints tiles
are drawn at the size of the other tiles' logos; the Autoscaler,
Backup and Restore icons were a fraction of it. - In the table view of the Endpoints page, an endpoint's name links to
its page again, as the card title does, and an endpoint whose
connection the administrator shares shows the shared icon again. - The Endpoints page shows its loading message or its content straight
away. It used to show nothing for half a second, and every later
change appeared half a second late. - The Endpoints page Register button and the endpoint card Creator row
follow the user endpoints setting. They kept the value from the
first page load. - The Register Endpoint tiles stay current as endpoints are added and
removed. They kept the first list they were shown. - The home page endpoint card shows its links once they load. An empty
links panel could stay on screen.
- Signing in with a user name and password no longer signs the
auto-registered Cloud Foundry endpoint in again when the user is
already connected to it. Before, every login replaced that
connection with one made from the console credentials, even when
the user had connected with other Cloud Foundry credentials. - Endpoint names sort the same way on every page: the home page, the
endpoint pickers and filters, the Cloud Foundry list, backup,
diagnostics and Foundation Shape. Names such as "dup 3" now sort
between "dup2" and "dup4". - Registering an autoscaler at an address on no registered Cloud
Foundry's domain now names the address and lists the registered
Cloud Foundry domains. Before, it named a cut-down domain such as
"com". - When the service broker list for an autoscaler is empty, the
register form says why: the Cloud Foundry is not connected, or only
a Cloud Foundry admin can list brokers. - Unregistering a Cloud Foundry names the autoscalers registered for
it, which are unregistered with it.
Organizations and Spaces
- An organization with more than 50 spaces now shows all of them. Its
Spaces tile, the Applications, App Instances and Memory tiles (which
count apps by space) and the Apply Quota to Spaces dialog read only
the first 50 spaces. - A space's Quota tab follows the space's quota. It showed the quota
from when the page opened, and a removed quota stayed on screen.
Services
- Rebinding a route service: when removing the old binding succeeded
and the new bind failed, the page kept showing the old binding, and
the next attempt failed on the binding that was already gone. The
page now shows the route as unbound. - Binding to a service instance that is still provisioning no longer
says "bound to the application" before the bind has happened. It
says the instance will be bound automatically when it is ready. - When a user-provided service instance was created but binding it to
the application failed, the message did not say the instance had
been created. It now does. - In the add service instance wizard, a failed fetch of the space's
applications no longer stops later fetches: changing the space
fetches them again. - The plan step picks its plan once each time it is entered, not twice.
- Adding a service instance from the Marketplace for a service whose
broker is registered to a space now lists its plans. Before, the
plan step stayed empty.
Users and Roles
- A role change made while the users list was still loading is no
longer overwritten by that older load. Before, lists showed the
previous roles until the page was reloaded. - Manage Roles opened from a space lists only that space again, not
every space of the organization. - The Apply button on the Manage Roles confirm step is yellow, not
red: a role change is a caution, not a deletion. - In Manage Roles, each space's role options follow that space's
permission. Before, the per-space checks could overwrite each other
and show the wrong space as editable. - Removing a user's space role and their organization User role in one
change no longer fails on the organization User role. The role
removal now waits for Cloud Foundry to finish deleting the space role
first. - Giving a user a space role in an organization they are not yet a
member of works again: the organization User role is ticked with the
space role and added first. Before, Cloud Foundry refused the space
role, and the organization User box could not be ticked afterwards. - In Manage Users, Next on the role steps follows the selection without
a delay. - Manage Roles no longer accepts a role change while the user's roles
are still loading. A change made then enabled Next and confirmed
against an organization with no name. - Manage Roles keeps Next disabled until a role differs from what the
user holds, and disables it again when the changes are undone.
- Opening Add User no longer logs a failed request in the browser
console when the user may not list identity providers. The Identity
Provider Origin field is a free-text box in that case, as before. - The Identity Provider Origin field in Add User now looks like the
other fields: underlined, with a floating label, instead of a
separate label over a boxed input.
Autoscaler
- An autoscaler policy can pin an app to one instance count: the policy
editor now accepts a minimum equal to the maximum, for the default
limits and for both schedule types, as the autoscaler does. - The Autoscale tab is laid out like the other application tabs: the
rule count and the Edit, Delete and Create Policy actions sit in the
same row as on Routes and Variables. The Latest Metrics gauges no
longer run past their card, and the schedule tables scroll instead of
cutting off their last column, with the edge shade the lists use to
show there is more. - Scaling rule conditions read evenly (
>= 80% for 120 secs.): percent
had padding no other unit had, and the operator had no spaces. - The autoscaler's 401 for a user who is not a space developer of the
app reaches the user as a 401 with its message, instead of a 500. - Adding or editing a recurring schedule in an autoscaler policy works
again. The schedule row failed to open with an error because its
Effective Duration and Repeat By choices were not bound to the form.
Git
- GitHub and GitLab endpoints check the token when connecting. Any token,
even a wrong one, used to connect and read Connected. - A GitHub endpoint whose user could not be read shows no avatar,
instead of a broken picture. - The GitHub and GitLab connect help opens again instead of showing an
empty "Javascript is required" page, and the Git logo shows.
Helm
- A Helm repository's Synchronize action is back in its endpoint menu,
and the endpoints list shows a Helm repository's sync status instead
of "Disconnected". - Leaving a Helm workload page closes only its own "Errors were found
when parsing this workload" warning, and deleting a workload closes
only an earlier "Failed to delete release". Both closed every
message on screen, including other pages' messages. - A Helm release's resource graph redraws on every update. It redrew
only on the first. - A Helm release's graph and summary show the analysis report picked
last. An earlier, slower report could replace it. - The Helm catalog's Artifact Hub badge follows the connected endpoints.
It stayed off when the catalog opened before the endpoints loaded.
Kubernetes
- A Kubernetes list whose cluster could not be reached, or whose token
had expired or was refused, showed its "There are no …" message as
if the cluster were empty. The Nodes, Namespaces, node Pods and
resource lists now show "Failed to load" with the cluster's reason
(for example401 Unauthorized, or the connection error) and a
Retry button. Deleting a resource or creating a namespace that the
cluster refuses now reports the failure instead of appearing to
succeed. - Picking a namespace on the Kubernetes Services list or a resource list
(Config Maps, Secrets, Deployments, ...) now lists that namespace's
resources; it showed "There are no ..." whatever the namespace held.
Refresh and Retry reload the namespace on screen, Pods included. - Kubernetes lists show their loading message until their own load ends,
instead of a flash of "There are no ...", and show only the failure of
the list on screen (a cluster-wide failure no longer marks a node's
Pods page or a namespace view as failed). - A Kubernetes Summary tab or home card whose node, namespace or pod list
failed to load shows "-" and the reason instead of 0. - List error banners name the endpoint ("prod-cf: failed to load")
instead of showing its GUID, on Cloud Foundry and Kubernetes lists. - A failing metrics endpoint now shows "Failed to load metrics: "
on Kubernetes node and pod metrics and Cloud Foundry application
metrics charts, and on a node's usage summary, instead of empty charts
and 0 values. - The Helm releases list shows which Kubernetes endpoints could not be
listed and why; their releases used to go missing silently. - Kubernetes list failures reach the page-header error banner and the
endpoint's error page, as Cloud Foundry list failures do. - The Namespaces tile on the Kubernetes Summary tab and home card reads
"1 Namespace" for a single namespace (it said "1 Node"), and the
Summary tab's Pods tile opens the pod list (it linked to a page that
does not exist). - Kubernetes clusters registered with client certificates trust the
endpoint's CA certificate; they failed with "certificate signed by
unknown authority" unless Skip SSL validation was set. A cluster that
refuses the certificate reports 401 instead of a proxy failure. - Import Kubeconfig registers each cluster with its
certificate-authority-dataas the CA certificate. A kubeconfig that
pinned its CA had to skip validation or fail. - Choosing a file in Import Kubeconfig closes only its own earlier
"Failed to load Kube Config" message, not every message on screen. - Kubernetes and Helm pages show their icons again, such as the Helm
charts filter toggle, analysis report status and the kube terminal;
they showed the icon's name ("filter_list", "warning") as text. - The Helm charts page names its repository filter once, by the
field's own label; a heading "Filter By Repository" sat over that
label. - A Helm chart or release without a logo shows the placeholder image
instead of a broken image, and the Popeye and kube-score analysis
descriptions open. All four pointed at /core/assets, where the build
does not put the Kubernetes package's files, so the console page
answered instead. - The node tags card no longer fails on a node without that metadata.
- The cluster summary shows "-" for the node version when no node
reports one, instead of a blank. - The Kubernetes home card waits for its pod, node and namespace counts
instead of showing 0 until they arrive.
Metrics
- The metrics routes for application and cell pages accept only the
queries those pages send. Through the metrics endpoint's stored
credentials, any signed-in user could add any PromQL to a cell query,
and a user who could see one application could read the metrics of
every application. - A metrics endpoint can be registered with "Connect (Optional)" again,
its card shows "N / M sources" again, and its page
(/endpoints/metrics/<id>) opens again instead of failing. - Registering a metrics endpoint that answers 404 or 500 fails instead
of succeeding, and connecting one that cannot be reached reports it
instead of failing the request. - A Cloud Foundry cell's Summary page opens again; it failed to load.
- A cell's App Instances list shows the cell's applications; it always
read "There are no applications". - A cell's health history lists the past hour's samples; it always read
"Cell has no health history". - A cell metrics request that fails shows its reason: the cell lists
show "Failed to load" with Retry, and the cell page an error above its
tabs. A cell with no health sample reads "Unknown", not a red ERROR
and "Healthy: No". - An application's instances list shows its Cell column again (each
instance's cell, linked to the cell's page) when the Cloud Foundry has
cell metrics. - The Metrics page shows its "provides no metadata file" notice again.
Core
- A wizard step that fails, such as a refused endpoint registration,
gives its reason above the step's buttons. It was a toast that
covered a dialog's Cancel and Next or Finish. - After a wizard step fails, keyboard focus returns to the step's Next
or Finish button, which assistive technology describes by the reason.
Focus used to drop to the page, sending a keyboard user back to the
top of it. - Icons draw at the size set for them. The side nav's are 18px,
Diagnostics' 16px and icons in buttons 20px; they were all 24px. A
type's font icon on the Home and Endpoints cards matches the images
beside it, the "no applications" panel's icon is 48px again, and an
empty page's icon is 96px again. - Home shortcut icons show a pointer and return to 70% opacity. A
service card's link icon is 16px, and the CaaSP version warning is
20px with its gap and help cursor. - An address wraps between its parts, not mid-word, on the Instance
Address, the broker link, a route service URL, an error's URL and a
chart's links. The Endpoints card shows an ellipsis on a cut-off
address. - A metrics endpoint's page draws each endpoint's icon in the right
font. A Cloud Foundry card showed a cloud and a hand with a gap
between them, which pushed its title to the right. - Home's "Card layout" menu closes on Escape, which returns focus to
its button, and on a click outside it. Its tick is 20px, so each
option is 36px tall, not 40px. - A failed favorites load no longer raises an error on every page that
reads favorites; they show as empty and the failure is logged once. - The favorite star shows its state from the first frame.
- A wizard's busy spinner takes the color of its button's text, so it
stays visible on every button style.
Maintainability
- Removed unused build tooling:
stratos-buildpack.yml,
deploy/cloud-foundry/build.shanddeploy/cloud-foundry/start.sh.
They served the old custom Stratos buildpack, which built Stratos from
source duringcf push. Stratos no longer deploys that way: the Cloud
Foundry deploy pushes the pre-builtstratos-cf-<version>.zipwith
thebinary_buildpackand never runs these files. Nothing changes for
a deploy that follows the current instructions. - The last release that has them is v5.5.5. To restore them:
git checkout v5.5.5 -- stratos-buildpack.yml deploy/cloud-foundry/build.sh deploy/cloud-foundry/start.sh - The backend no longer depends on
github.com/SermoDigital/jose, an
unmaintained JWT library that panics at startup under Go 1.27. It
came in through the CF CLI libraries used for app push; the CLI now
usesgolang-jwtv5, so the patched fork Stratos substituted for it
(norman-abramovitz/fw-jose) and both of itsreplacelines are gone. - Removed the long-running request path from the backend proxy. Nothing
has sent itsx-cap-long-runningheader since the service instance
actions that used it were replaced; Cloud Foundry v3 answers those
writes with a background job instead. The
HTTP_CLIENT_TIMEOUT_LONGRUNNING_IN_SECSsetting is gone from the
shipped configs and Helm chart; a deployment that still sets it is
unaffected, as the value is ignored. - The library type checks pass again (TypeScript 6.0 had broken them
all) and now run in the gate and on every pull request. - The store's signal-to-observable helper (
observeSignal) now lives
exactly as long as its subscription, whatever injector it is given.
With a component's injector it used to end with that view and leave
any later subscriber waiting forever. No current caller outlives its
view; this removes the trap for the next one. - The Helm upgrade page ends its upgrade-target subscription when the
page closes, so a late answer no longer sets up a page that is gone. - Replaced the RxJS APIs deprecated in RxJS 7 (publishReplay/refCount,
retryWhen, toPromise, multi-argument combineLatest and subscribe,
the combineLatest operator, throwError with a value), the deprecated
HttpClientModule, and the NgZone wrappers that do nothing in the
zoneless app. Lint rules now reject them, and the theme package is
linted. - The shared card header shows the shaded title band the other cards
use by default; the cards that had no band keep their look with the
newplainoption. - Most component inputs and outputs moved from the
@Input()and
@Output()decorators to Angular's signal-basedinput()and
output():
| Kind | Decorator before | Decorator after | Signal after |
|---|---|---|---|
| Inputs | 602 | 195 | 454 |
| Outputs | 62 | 4 | 60 |
- Pages should look and behave as before. The 195 inputs left on the
decorator need design changes and follow in a later release. - 47 more component inputs moved from the
@Input()decorator to Angular's
signal-basedinput(): the ones a template reads inside an@if. Pages
should look and behave as before. 148 inputs remain on the decorator. - 21 more component inputs moved from the
@Input()decorator to Angular's
signal-basedinput(): inputs written as a setter that only stored the
value or worked out what to show from it. Pages should look and behave as
before. 127 inputs remain on the decorator. - 22 more component inputs moved from the
@Input()decorator to Angular's
signal-basedinput(): inputs written as a setter that worked out what to
show or passed the value on to a service, among them the page header's
tabs, breadcrumbs and favorite. Pages should look and behave as before.
105 inputs remain on the decorator. - 13 more component inputs moved from the
@Input()decorator to Angular's
signal-basedinput(): inputs written as a setter that emitted an event,
filled in a form or started something when set (the connect form, the log
viewer, parameter forms). Pages should look and behave as before.
92 inputs remain on the decorator. - The wizard step's 12 setter inputs (
hidden,valid,canClose,skip,
the button texts and others) moved from the@Input()decorator to
Angular's signal-basedinput(). The stepper now follows a step's inputs
directly, so the step's four change events are gone. Wizards should look
and behave as before. 80 inputs remain on the decorator. - The last 80 inputs moved from the
@Input()decorator to Angular's
signal-basedinput()andmodel(): the list cell base class and its
cells and cards, the select, checkbox, toggle and date controls, the
endpoint auth forms, the wizard step's remaining inputs, the sort and
name-check directives and a number of single components. No@Input()
decorator remains in the frontend source. Pages should look and behave
as before. - Extension authors: an endpoint auth form's
formGroup(andconfig,
where it has one) is now a signal input, declared withinput()and
read asformGroup(). TheIAuthFormtype reflects this. - The view and content queries (
@ViewChild,@ViewChildren,
@ContentChild,@ContentChildren) and the last three@Output()
decorators moved to Angular's signal-basedviewChild(),
contentChild()andoutput(). With the inputs done earlier, no input,
output or query decorator remains in the frontend source. Pages should
look and behave as before. - Seven components that wrapped a signal in
toObservableonly to read it
back with the async pipe now read the signal in the template: the logout
page's error view, the page header's mobile menu toggle, the three
profile settings toggles, the Kubernetes analysis report breadcrumbs,
the Helm release analysis report, the kubeconfig certificate cell and
the service instance wizard title. Each now renders in the same
change-detection pass instead of one microtask later. Pages should look
and behave as before. - The loading overlay, the card status bar and the card progress overlay
take a plain value as well as an Observable. Seventeen pages and cards
that held a signal and wrapped it intoObservableonly to feed one of
those inputs now pass the signal's value directly: the two setup
wizards, the metric card, the org, space, quota, git tab and
application delete pages, the app usage, instances and status cards,
the service plan step, the Kubernetes dashboard and its configuration
card, the Kubernetes summary and the Helm release summary. Pages
should look and behave as before. - The recently updated applications card takes its app list and
loading flag as plain values as well as Observables; the home card
passes its signal value directly. The card should look and behave as
before. - The application pages read the app, org, space, endpoint and loading
state as signals from the detail data service instead of through the
ApplicationService Observable facade: the action bar, the page header
and its breadcrumbs, the edit and autoscaler policy titles and the
build tab's endpoint name. They should look and behave as before. - The application pages and cards read the application detail data
service's signals directly: the CLI info, SSH, edit and autoscaler
pages, the build, git, services and routes tabs and the instances
card. The ApplicationService Observable facade and the unread
ApplicationMonitorService are removed. Pages should look and behave
as before. - The Cloud Foundry endpoint service exposes its apps, loading flag,
counts, memory total, CF info and SSH access as signals. The Cloud
Foundry summary tab, its info card and the Users tab read them
directly. Pages should look and behave as before. - The Cloud Foundry organization service exposes its spaces, apps, app
instance count, memory total, loading flag, user count, the current
user's role and the quota link as signals. The organization summary
page and its summary card read them directly, and the page now
subscribes to the route, private domain, user-provided service and
quota requests itself. Pages should look and behave as before. - The services behind the Cloud Foundry, organization, space,
Kubernetes, endpoint list, shell (profile, notifications, tabs,
history, recent activity, favorites) and wizard pages expose their
state as signals, and the pages read the signals directly. The
Observable copies of that state are gone. Requests (counts, quotas,
dashboard status) stay streams that each page subscribes to. - Permission checks are signals end to end: the checkers are computed
from the current user's role, scope and feature flag signals, and
can()returns a signal that pages and directives read directly.
The stream versions of the checks and the role wrapper services are
gone. - Wizard steps submit with Promises: a step's next action is an async
function, and the stepper awaits it. The unused ObservableonNext
input on steps and theStepOnNextFunctiontype are gone. - Component state that was held in RxJS subjects or refreshed with
markForCheckis held in signals, across the stepper, the deploy and
add service instance wizards, the autoscaler policy editor, the login
page, the log and analysis report views and the custom form
controls. Subjects that nothing read are removed, as are the unused
global event configuration channel,SteppersServiceand the Helm
MenuService. - State that pages read is held in signals across Cloud Foundry, core,
Kubernetes, Helm and the autoscaler: Observable wrappers around
signals, and the template subscriptions that read them, are removed.
Timing workarounds that delayed values by a tick or half a second are
gone, as are unused services, members and watchers that outlived the
rows or steps that created them. - The application detail data service reads with
httpResource. SessionService,KubernetesNamespacesFilterServiceand the unused
Observable members of the metrics service are removed.
Chores
- The production build now fails on any warning that is not listed, with
a reason, inscripts/build-warnings-allowed.mjs. Listed warnings still
print on every build. This is what let the Monaco 0.57 bump push the
editor chunk past its size budget unnoticed; that budget is now 2.85 MB. make build release cfno longer warns that frontend and backend are
not valid modifiers for release; only modifiers you type are checked.- The project's own build, lint, release and CI scripts run with bun
instead of callingnodedirectly. Node remains the runtime for the
Angular CLI and vitest; the version check now reports the Node on PATH
rather than bun's built-in Node compatibility version. - The developer, testing, extension and Cloud Foundry deploy guides now
give themakeandbuncommands the repo uses today, and the
frontend tests guide lists the checks a pull request runs. - The Makefile, the tiered E2E runner and the STB workflow run their
tools throughbuninstead ofnpxornpm. - Removed the unused Code Climate configuration and script, and the
BrowserStack instructions, which stopped working with the Protractor
suite. make check coveragenow reports backend coverage as well: one Go
profile per module incoverage/backend/, with per-module and total
statement coverage.make check frontend coverageand
make check backend coveragerun one side.- Frontend coverage now counts every source file in the tested packages,
including files no test loads; before, those files were left out of the
report and the totals read high. The report moved tocoverage/frontend/. - Removed the unused devkit
applicationanddev-serverbuilders.
Their directory moved the compiled devkit underdist-devkit/src/,
sobun run prepare-backendand the setup step that generates the
backend plugin list could not finddist-devkit/backend.js. Both run
again. - Dependency updates:
| Package | From | To |
|---|---|---|
| eslint | 10.11.0 | 10.12.0 |
| @types/node | 26.6.3 | 26.6.4 |
| go-cfenv | 1.24.3 | 1.24.4 |
| code.cloudfoundry.org/clock | 1.89.0 | 1.90.0 |
| postcss (docs site) | 8.5.28 | 8.5.29 |
| lucide-react (docs site) | 1.48.0 | 1.52.0 |
| baseline-browser-mapping (docs site) | 2.11.26 | 2.11.27 |
| caniuse-lite (docs site) | 1.0.30001812 | 1.0.30001814 |
- The Go update also moves the backend's indirect Kubernetes client
libraries from 0.36.4 to 0.37.1, and thegolang.org/xpackages. - Removed the unused
browserstack-localdevelopment dependency. - Backend dependency updates, from moving to the current CF CLI v8
branch (untagged; the next v8 release will replace the
pseudo-version):
| Package | From | To |
|---|---|---|
| code.cloudfoundry.org/cli/v8 | 8.19.0 | 8.19.1-0.20261006035801-8f3eae3b52da |
| github.com/golang-jwt/jwt/v5 | (new) | 5.3.1 |
| google.golang.org/grpc | 1.83.2 | 1.84.0 |
| code.cloudfoundry.org/tlsconfig | 0.65.0 | 0.68.0 |
| code.cloudfoundry.org/bytefmt | 0.86.0 | 0.92.0 |
| github.com/sirupsen/logrus | 1.10.1 | 1.10.2 |
| github.com/mattn/go-runewidth | 0.0.28 | 0.0.30 |
- The backend linter moves from golangci-lint 2.13.1 to 2.14.0. Go
1.27.2 writes a newer compiler export-data format that 2.13.1 cannot
read, somake check lintandmake check gatefailed on any
machine with Go 1.27.2 installed. - Service instance, binding and service key writes are now tested with
Cloud Controller responses recorded on a live foundation, from a
broker answering synchronously and one answering asynchronously,
including broker failures. The backend replays them through its
write handlers, and the frontend specs read what the backend
returned. Each write is checked on the fast path and the polled
path, and a broker's sync or async answer must give the user the
same outcome. Broker failure messages that differ between the two
paths are marked as expected failures (#6002). - A new E2E spec runs the create-service-instance wizard against a
broker switched between synchronous and asynchronous responses
(services.testBrokerin the secrets profile; see
e2e/secrets.yaml.template). make dev backendprints which database the backend will open before
it starts: for SQLite the file's full path, whether it is kept between
restarts or reset at startup, and the endpoints registered in it. Each
checkout has its ownconsole-database.db, so its endpoints, tokens
and favourites change with the directory the backend runs from.- Pull requests that change only
docs/now build the documentation
website, as the deploy does, and the docs lint rejects relative links
that leavedocs/. A link to../CONTRIBUTING.mdhad passed both and
stopped the site from publishing; it now uses its GitHub URL. - The CF API client (capi) is taken from a fork commit carrying the
user-provided delete fix until a capi release includes it. - The marketplace and service e2e specs that could not fail (calls to
helpers the fixture never provided,expect(true), dialogs closed
before acting) are replaced by tests that change real instances and
check the result in CF: slow updates and deletes, user-provided
create, edit and delete, binding, managed edit, and the marketplace
and services list filters. - Release-notes fragments can name a subcategory,
[Features: Performance];
bullets that share one are grouped under a heading inside their
section.changelog.d/README.mdlists the suggested names. - Dependency updates:
| Package | From | To |
|---|---|---|
| Angular (framework, CDK, CLI, build) | 22.2.1 | 22.2.3 |
| angular-eslint | 22.5.0 | 22.6.0 |
| marked | 18.0.14 | 18.1.0 |
| js-yaml | 5.4.2 | 5.4.3 |
| jsdom | 30.1.1 | 30.1.3 |
| @types/node | 26.6.4 | 26.6.5 |
| code.cloudfoundry.org/cli/v8 | 8.19.1 snapshot | 8.19.1 |
| code.cloudfoundry.org/clock | 1.90.0 | 1.91.0 |
| github.com/aws/aws-sdk-go-v2 | 1.47.1 | 1.47.3 |
| github.com/prometheus/client_golang | 1.24.1 | 1.25.0 |
| golang.org/x/crypto | 0.57.0 | 0.58.0 |
| golang.org/x/sync | 0.23.0 | 0.24.0 |
| helm.sh/helm/v3 | 3.22.0 | 3.22.1 |
| sigs.k8s.io/aws-iam-authenticator | 0.7.20 | 0.7.22 |
| @radix-ui/react-avatar (docs site) | 1.2.6 | 1.3.1 |
| @radix-ui/react-slot (docs site) | 1.3.3 | 1.5.0 |
| lucide-react (docs site) | 1.52.0 | 1.59.0 |
| react, react-dom (docs site) | 19.3.0 | 19.3.1 |
| postcss (docs site) | 8.5.29 | 8.5.30 |
| prettier (docs site) | 3.9.9 | 3.9.10 |
| baseline-browser-mapping (docs site) | 2.11.27 | 2.11.29 |
| caniuse-lite (docs site) | 1.0.30001814 | 1.0.30001816 |
- aws-iam-authenticator 0.7.22 needs Go 1.27.2: four backend modules
now declare it, and the.tool-versionsGo pin moves to 1.27.3. - Dependency updates:
| Package | From | To |
|---|---|---|
| marked | 18.1.0 | 18.1.1 |
| js-yaml | 5.4.3 | 5.5.0 |
| code.cloudfoundry.org/clock | 1.91.0 | 1.92.0 |
- js-yaml 5.5.0 follows the YAML spec more strictly: a block plain
scalar that starts with,,]or}is now a parse error, which
Kubernetes kube-config registration and the Helm chart values editor
report as invalid YAML. - Aligned the angular-eslint package with its plugins (22.6.0). The
version mismatch made the template label rule report every label
that wraps its control, 22 false warnings. Dependabot now bumps the
package together with its plugins, and lint fails on any ESLint
warning, so warnings can no longer pass the gate unnoticed. make bump set TO=<version>sets an exact version, the way the
other bumps run, and honoursDRYRUN=yes. Version bumps now stamp
this repository's commit SHA even when run from another directory;
before, they stamped whatever repository the shell was in.
src/jetstream/VERSIONnow gets the version exactly as
package.jsondoes, with or without av. The version-bump tests
pass again, and the duplicate copy undertest/is gone.- Dependency updates:
| Package | From | To |
|---|---|---|
| Analog (vite-plugin-angular, vitest-angular) | 2.8.0 | 2.9.2 |
| vitest, @vitest/coverage-v8, @vitest/ui | 5.0.3 | 5.0.5 |
| acorn | 8.18.0 | 8.19.0 |
| @oxc-project/runtime | 0.152.0 | 0.155.0 |
| baseline-browser-mapping | 2.11.27 | 2.11.31 |
| code.cloudfoundry.org/clock | 1.92.0 | 1.93.0 |
| sigs.k8s.io/aws-iam-authenticator | 0.7.22 | 0.7.23 |
| baseline-browser-mapping (docs site) | 2.11.29 | 2.11.30 |
- Dependabot now groups the Analog packages so they move together.
- Removed
deploy/proxy.env, an unused docker-compose file left over
from 2017 whose sample database password tripped secret scanning. - The build and packaging guide now lists the
make deploy website
commands and the one-time fork setup for Pages previews. Before, they
were only inwebsite/README.md, which the docs site doesn't publish. - Dependency updates:
| Package | From | To |
|---|---|---|
| sass | 1.105.1 | 1.106.1 |
| pixelmatch | 7.2.0 | 8.0.0 |
| caniuse-lite (docs site) | 1.0.30001816 | 1.0.30001817 |
| lucide-react (docs site) | 1.59.0 | 1.61.0 |
- pixelmatch 8 compares colors in OKLab, so screenshot diff counts
change; the threshold scale does not. - Dependency updates:
| Package | From | To |
|---|---|---|
| Angular CLI, build, devkit and cdk | 22.2.3 | 22.2.5 |
| typescript-eslint | 8.71.0 | 8.72.1 |
| chart.js | 4.5.1 | 4.6.1 |
| @playwright/test | 1.63.0 | 1.64.1 |
| baseline-browser-mapping | 2.11.31 | 2.11.34 |
| sigs.k8s.io/aws-iam-authenticator | 0.7.23 | 0.7.24 |
| github.com/aws/aws-sdk-go-v2 | 1.47.3 | 1.47.4 |
| baseline-browser-mapping (docs site) | 2.11.30 | 2.11.34 |
| caniuse-lite (docs site) | 1.0.30001818 | 1.0.30001820 |
- aws-iam-authenticator 0.7.24 needs Go 1.27.3, so building the backend
now needs Go 1.27.3 or later. make check gateand CI now type-check the frontend spec files.
Vitest strips types when it runs a spec, so a spec could call a
method that does not exist and still pass. Three specs that did not
test what they claimed now do.- Dependency updates:
| Package | From | To |
|---|---|---|
| Angular | 22.2.3 | 22.2.4 |
| @analogjs/vite-plugin-angular, vitest-angular | 2.9.2 | 2.9.6 |
| eslint | 10.12.0 | 10.14.0 |
| sass | 1.106.1 | 1.107.1 |
| @types/node | 26.6.5 | 26.6.6 |
| github.com/cloudfoundry-community/go-cfenv | 1.24.4 | 1.25.0 |
| github.com/labstack/echo/v5 | 5.4.2 | 5.4.4 |
| github.com/ncruces/go-sqlite3 | 0.35.6 | 0.36.0 |
| go.yaml.in/yaml/v4 | 4.0.0-rc.6 | 4.0.0-rc.7 |
| lucide-react (docs site) | 1.61.0 | 1.68.0 |
| postcss (docs site) | 8.5.31 | 8.5.32 |
- go-sqlite3 0.36.0 brings SQLite 3.54.1 for the SQLite datastore.
globis 14.0.3, was 13.0.6, for GHSA-3g3m-2fjr-6pw4. Stratos passes
no abort signal to it, so it was not exposed.- Dependency updates:
| Package | From | To |
|---|---|---|
| Angular | 22.2.4 | 22.2.6 |
| Angular CLI and build | 22.2.5 | 22.2.6 |
| @angular/cdk | 22.2.5 | 22.2.7 |
| @analogjs/vite-plugin-angular, vitest-angular | 2.9.6 | 2.9.7 |
| vitest, @vitest/coverage-v8, @vitest/ui | 5.0.5 | 5.1.0 |
| tailwindcss, @tailwindcss/postcss | 4.3.3 | 4.3.5 |
| ng-packagr | 22.2.4 | 22.2.6 |
| baseline-browser-mapping | 2.11.34 | 2.11.35 |
| github.com/ncruces/go-sqlite3 | 0.36.0 | 0.37.0 |
| go.yaml.in/yaml/v4 | 4.0.0-rc.7 | 4.0.0 |
| golang.org/x/sync | 0.24.0 | 0.25.0 |
| code.cloudfoundry.org/clock | 1.94.0 | 1.95.0 |
| github.com/aws/aws-sdk-go-v2 | 1.47.5 | 1.47.8 |
| lucide-react (docs site) | 1.68.0 | 1.70.0 |
| tailwindcss (docs site) | 4.3.3 | 4.3.5 |
- go-sqlite3 0.37.0 starts
database/sqltransactions with
BEGIN IMMEDIATE. In this code base only the schema migrations at
startup open transactions. - go.yaml.in/yaml/v4 moves from a release candidate to 4.0.0.
- The user endpoints guide now explains what the feature is for and why
endpoints may share an address, with the full rules on a new details
page. The Cloud Foundry endpoint guide describes the Register and
Connect dialog, CA certificate checks and application labels, and the
API keys guide covers key expiry.
Security Updates
source-map-jsmoved from 1.2.1 to 1.2.2 in the frontend build, the
devkit, the docs site and STB:
| Advisory | What it fixes |
|---|---|
| GHSA-68fv-2mgg-jv7q | Event-loop denial of service through indexed source-map section offsets |
launch-editormoved from 2.14.1 to 2.14.2 in the devkit, and its
vitefrom 8.3.0 to 8.3.5, whose bundled copy carried the same code:
| Advisory | What it fixes |
|---|---|
| GHSA-587h-7wmj-h5pm | fs.existsSync() existence check can be bypassed on Windows |
| GHSA-r825-w8mm-5pxw | Editor CLI flag misinterpretation causing command execution |
| GHSA-698v-4r3x-3vcv | escapeCmdArgs() omits % escaping on Windows |
github.com/klauspost/compressmoved from 1.18.0 to 1.20.1 in the
cfapppush plugin's module. The backend binary already built with
1.19.1, which is not affected:
| Advisory | What it fixes |
|---|---|
| GHSA-259r-337f-4rfw | Integer overflow and out-of-bounds write in the s2 decoder |
vitestand its browser packages moved from 4.1.10 to 4.1.12 in STB's
test tooling. Nothing in the shipped console uses them:
| Advisory | What it fixes |
|---|---|
| GHSA-82fw-gwwq-j7x9 | File read through the @vitest/mocker redirect mock |
github.com/labstack/echo/v5moved from 5.4.0 to 5.4.2 in the backend
and all six of its plugin and API modules. 5.4.1 fixes these; 5.4.2
adds hardening with no advisory:
| Advisory | What it fixes |
|---|---|
| GHSA-cxf3-9h37-7rr8 | Rejected WebSocket upgrades can bypass route-level access controls |
| GHSA-9969-6676-49h5 | WebSocket proxy keeps connections and handlers after client disconnect |
| GHSA-h53f-rm2w-w7rg | X-Real-IP extractor trusts the header instead of the sending peer |
| GHSA-x9ff-22gv-65f4 | BodyDump can discard BodyLimit overflow errors |
happy-dommoved from 20.14.5 to 21.0.4 in the unit-test tooling.
Nothing in the shipped console uses it:
| Advisory | What it fixes |
|---|---|
| GHSA-834v-8h66-3h79 | Incomplete fix for CVE-2026-33943: an import alias runs host code |
body-parser2.3.0 → 2.4.0,fast-uri3.1.8 → 3.1.9 and
serialize-javascript7.0.5 → 7.1.4 in the devkit:
| Advisory | What it fixes |
|---|---|
| GHSA-27r3-jxmh-p828 | body-parser: denial of service via trailing bytes after a compressed body |
| GHSA-w74x-v6f9-j3xg | fast-uri: URI scheme spoofing via percent-encoded scheme decoding |
| GHSA-wjr9-j795-jxcr | fast-uri: host confusion via unchecked WHATWG host canonicalization |
| GHSA-333x-w4r6-6crp | fast-uri: scheme injection via an unescaped first-segment colon |
| GHSA-xhq7-xmqj-769h | fast-uri: improper path equivalence in relative reference normalization |
| GHSA-m5mm-p8h7-m29v | serialize-javascript: XSS via an unescaped </script in a serialized function |