Server and dashboard
Server and dashboard release notes
Release notes for the server image, tagged codemagic-patch-server-vX. The
dashboard is included in the image and has no separate version. Every entry
starts with Upgrade impact: migrations, configuration changes, and anything
that restarts or replaces a service. Review it before upgrading. Managed Patch
runs the latest server; self-host operators upgrade with
cmpatch selfhost upgrade. The
other surfaces' releases are on the
Release Notes page.
Reference: Configuration · Operations · GitHub releases
0.11.0
19th November 2026
Upgrade impact
- Runs migration
0023_webhooks. It adds three tables and one column onrelease_job, and leaves existing rows as they are. - No configuration changes are required.
- Do not roll back to an older image after the migration. Server 0.10.0 does not start on a migrated database. See Upgrades.
Webhooks
Adds webhooks for release events. Register a URL for your team, and the server
sends it a signed POST when a release goes live, changes, or fails.
cmpatch webhook create --url https://example.com/hooks/patch
- Sends
release.published,release.promoted,release.rolled_back,release.updated,release.disabled,release.enabled, andrelease.failed, pluspingfor a test. - Signs every request as described by Standard Webhooks, so you can verify it with an existing library.
- Retries a failed delivery for about a day. After that you can send it again yourself. Delivery history is kept for 30 days.
- Accepts public
httpsURLs only. On a self-hosted server, setWEBHOOK_ALLOW_PRIVATE_TARGETS=trueto allow private addresses andhttp.
Team admins and owners manage webhooks on the dashboard's Webhooks page or
with cmpatch webhook in CLI 0.11.0. See
Webhooks.
Other changes
- Adds
ownertoPOST /v1/teams. The request creates the team and invites its owner, and it works with an API token. A server older than 0.11.0 ignoresowner, so check the response forowner_invitation. See API tokens. - Makes the dashboard usable on a phone. Tables drop their secondary columns, and management actions are hidden on narrow screens.
- The Add member dialog now says that no email is sent to the person you add.
- Shortens page subtitles and dialog text across the dashboard.
See the server release.
0.10.0
17th November 2026
Upgrade impact
- Update
cmpatchto CLI 0.10.0 or later before you upgrade. CLI 0.9.0 and older leave the server's checkout on the default branch, and this release's upgrade script refuses a checkout that is not at a release tag. A refused upgrade changes nothing, so update the CLI and run it again. - The upgrade checks out the
codemagic-patch-server-v0.10.0tag and pullsghcr.io/codemagic-ci-cd/codemagic-patch-server:0.10.0instead of building, so the server must reachghcr.io.--build-from-sourcebuilds on the server as before. - Runs the stock
caddy:2-alpineimage for Caddy, pinned by digest, and serves the dashboard from the server image./worker/*answers 404 instead of the dashboard page, and a plain-HTTP install no longer sends an HSTS header. - Runs migration
0021_credential_generation, which adds one integer column each touser_account,api_token, andoauth_session. - Runs migration
0022_service_account_owner_team, which addsuser_account.owner_team_id. It marks amember provisionaccount as a service account of its team when it has never signed in and belongs to that team only. Existing tokens and sessions stay valid. - A session older than 90 days fails at its next refresh. Sign in again.
- CI that uses an API token to create tokens, add or invite members, change roles, or transfer apps stops working. See API tokens.
- Do not switch back to an older image alone once the database is migrated. If the upgrade fails after the stack is recreated, follow the steps the command prints, which end with restoring the pre-upgrade backup. See Upgrades.
Breaking: API tokens cannot create credentials or grant access
An API token (cm_pat_…) can no longer give access to anyone else, so a
leaked token leaves no access behind once it is deleted. These need a browser
sign-in (the dashboard, or cmpatch login without --token) and answer a
token with 403, reason session_required:
token create, and approving a CLI loginmember add,member invite,member updatemember provision,member rotate-tokenapp transfer
A token can still manage apps, deployments, and releases, and remove members, invitations, and tokens.
- The server logs each refused request with the token's ID.
- Roles and invitations a token granted before this release are not revoked. If a token may have leaked, review your team's members.
- An account that has never signed in through a provider loses its API
tokens at its first provider sign-in. This covers
member provisionaccounts, and accounts created before OAuth sign-in that are linked by email. Create new tokens after that sign-in.
See API tokens.
Sessions
- A sign-in ends 90 days after the provider sign-in, however often it refreshes. A CLI sign-in approved from the dashboard shares that dashboard sign-in's deadline.
- Adds sign out everywhere to the dashboard's API tokens page. It ends every session on your account, including the current browser and every CLI, and deletes all your API tokens. See Sessions and signing out everywhere.
Service accounts
- A
member provisionaccount is a service account of the team that created it. It has access in that team only, even if it is added to another team, and the Members page marks it with a Service account badge. - An admin of the owning team replaces its token with Rotate token on the
Members page or
cmpatch member rotate-token. The old token stops working immediately.
Self-hosting
- Each server release publishes one image for x86 and ARM, with a build provenance attestation. CLI 0.10.0 installs and upgrades from it instead of building on the server.
- The server refuses to start on a database that has applied a migration the
release does not know.
cmpatch selfhost upgradechecks for that before it changes anything.
Metrics API
Metrics responses now include values the dashboard used to calculate itself. Existing fields are unchanged.
- Adds
success_rateto every metrics object. It is applied over applied plus failed, from 0 to 1, ornullwhen both are 0. - Adds
summaryto deployment metrics and timeseries responses. It holds the totals of the returned page or range, and the active devices of the last complete UTC day. - Adds
cumulative_successto each timeseries series, the running Applied total per day.
Dashboard
- Removes the Metrics pages. A deployment page shows adoption over time and
the totals of the selected range, and a release page shows its counters and
the failure drill-down. Old
/metricslinks answer 404. - The release page shows who created and last updated the release, what they did, and when. Job attempts and failure detail appear only while a job runs or after it fails.
- The deployment page refreshes its release history while a release is processing.
- Each provider's sign-in button shows its logo.
See the server release.
0.9.0
3rd November 2026
Upgrade impact
- Runs migration
0020_server_instance, which creates the emptyserver_instancetable for the usage report's instance id. No existing table is touched and no lock of note is taken. - Sends Codemagic an anonymous usage report of aggregate counts once a day,
on by default. Codemagic uses it only for public statistics, never to
enforce the licence's 1M MAU limit. Set
SERVER_TELEMETRY=false(orDO_NOT_TRACK=1) in.env.selfhostto turn it off. docker-compose.selfhost.ymlnow passesSERVER_TELEMETRY,DO_NOT_TRACK, andSERVER_UPDATE_CHECKthrough to the server. ASERVER_UPDATE_CHECK=falseline in.env.selfhostwas silently dropped before, and now turns the GitHub release lookup off.- Rollback is safe. An older image ignores the
server_instancetable.
Dashboard
- The deployment charts use muted series colours instead of neon ones, in light and dark mode.
- The Status page names the bundled object store Silo, which is what runs. The health check path is unchanged.
See the server release.
0.8.0
28th October 2026
Upgrade impact
- Runs migration
0019_metricEventDailyAggregates, which adds the daily aggregate tables and the fold watermark, and builds an index onmetric_event (emitted_at). Metrics ingest waits for the index build, which takes longer on a large history. Reads are not blocked. - Shortly after startup, the server folds the existing history outside the raw retention window and deletes those raw events, including individual failure details. On a large history, the first run takes one long transaction per folded day.
- Postgres keeps the space the fold frees for new rows instead of returning it to the operating system, so the volume stops growing but does not shrink. To shrink it, see Disk size.
- A server older than 0.8.0 reads raw rows only, so the days already folded disappear from its metrics. To roll back, restore the pre-upgrade backup instead of only switching back to an older image.
- No configuration changes are required.
Metrics: 8-day raw retention
- Keeps raw metric events for 8 UTC days. Older events are folded once a day into daily aggregates, which are kept indefinitely. Counters, adoption charts, timeseries, and failure reasons are unchanged across the fold.
- The per-event failure list (
failure-eventsin the CLIs, the failure dialog's event list in the dashboard) covers the retained days only. A laterAppliedclears a device's earlierFailedevents only within that window. - Acknowledges but does not store an event dated 8 or more days before the day the server receives it, or more than 24 hours ahead. SDKs already drop events older than 7 days, so devices with a correct clock are unaffected.
Self-hosting
- The self-host scripts support a local stack reached at an ngrok
https://address throughcmpatch selfhost install --ngrok. - The Status page's upgrade hint now says to run
cmpatch selfhost upgradefrom the machine you installed from, which covers Fly installs as well as SSH ones.
See the server release.
0.7.0
19th October 2026
Upgrade impact
- Runs migration
0018_drop_metric_event_sdk_version, which drops the unusedsdk_versioncolumn from themetric_eventtable. The lock is brief and the table is not rewritten. Older SDKs that still sendsdk_versionkeep working, and the server ignores the field. - A server older than 0.7.0 still writes
sdk_version, so its metrics ingest fails after this migration. To roll back, restore the pre-upgrade backup instead of only switching back to an older image. - No configuration changes are required.
Fixes
- The deployment and release pages fetch only the selected adoption chart range instead of all four. This reduces database load, and the pages no longer stall under heavy usage.
- A backup restarts the exact server container it stopped. Before, the
restart could start a server with a different image or configuration if
either changed during the backup, as happens during
cmpatch selfhost upgrade.
See the server release.
0.6.0
6th October 2026
Upgrade impact
- Runs migration
0017_app_framework, which adds aframeworkcolumn to theapptable with the defaultreact-native. No index is rebuilt. - Problem details include a machine-readable
reason, anddetailno longer names acmpatchcommand. CLI 0.6.0 andcmpatch-capacitorprint the remedy themselves. Update scripts that matched the olddetailtext. - No configuration changes are required.
Capacitor apps
The server stores each app's framework (react-native or capacitor), and
cmpatch-capacitor sets it on app create. GET /v1/teams/:id/apps accepts
framework= and returns only apps with that framework, or all apps without
it. Apps created before this release are treated as React Native.
The dashboard shows commands for the app's framework. For a Capacitor app,
the New release dialog takes the built web assets (the webDir of
capacitor.config, or a ZIP of its contents), requires a target binary
version, and shows the cmpatch-capacitor release command instead of
cmpatch release-react.
Binary version filter
Adds a Binary version menu to the deployment page. It filters the
Adoption card and the release history table, and lists the versions that
reported in the selected range, plus group rows such as 1.x and 1.0.x
when a group covers two or more versions. The Metrics page has the same menu,
and its Active devices card counts the selected version or group. Release
detail has a By binary version table with Downloaded, Ready, Applied,
Failed, and success rate per reported binary version, shown when two or more
versions reported.
GET /v1/metrics/deployments/:id/timeseriesandGET /v1/deployments/:id/releasesacceptbinary_version(exact) orbinary_version_prefix(dot-separated digits;1matches1.0.0and1.10.2, not10.0.0). Both together return400 invalid_combination. The timeseries response includesbinary_versions, andGET /v1/metrics/releases/:releaseIdnow returns{ binary_versions, release }.- In filtered release history, a release matches when its explicit target binary version matches or any version in its current resolved target set matches. Fingerprint-expanded targets are included and superseded generations are not. Rollback gating and the suggested target for a new release use the unfiltered history.
Full and diff download metrics
Adds a card to release detail with the size and download count of the full bundle and of the diff from the previous release. The diff size is measured against the previous release only, so it is approximate when devices skip releases.
See the configuration reference and server release.
0.5.1
2nd October 2026
Upgrade impact
- Replaces the bundled MinIO object storage images, which are no longer available to pull, with PGSTY Silo images. The upgrade replaces the storage container. Existing storage volumes stay compatible and need no data migration.
- No database migration or configuration change.
Fixes
- A backup restarts only the server service. It no longer recreates dependent services such as object storage or the database before the backup is confirmed.
See the server release.
0.5.0
23rd September 2026
Upgrade impact
- Runs migration
0016_metric_event_lifecycle_names, which rebuilds an index on themetric_eventtable while holding an exclusive lock on it. On servers with a large metrics history, the upgrade takes longer. - Stores events from SDK 0.4.x and earlier under the new
ReadyandAppliedlifecycle names. Existing data is kept as is, and the metrics API response keys (installed,success) are unchanged. SDK 0.5.0 needs this server; see compatibility. - Adds the optional setting
GITLAB_OAUTH_BASE_URL, for GitLab sign-in on a self-managed instance. Leaving it unset changes nothing.
GitLab sign-in
Adds GitLab sign-in for self-hosted servers, on gitlab.com or a self-managed
instance set with GITLAB_OAUTH_BASE_URL. GitLab accounts that have only a
confirmed primary email can sign in, and the sign-in button shows the GitLab
name and icon.
Command palette and navigation
Adds a command palette (⌘K / Ctrl+K) that opens apps, releases, metrics, and
teams, and runs common actions. The deployment page shows update adoption and
links to the Metrics screens, and the sidebar lists apps with their
deployments. Release details include a cumulative Applied chart of update
adoption over time.
- Fixes dashboard page scrolling and removes blank space that a hidden chart table added at the bottom of the page. Cleans up how the deployment and release detail pages show status, metrics, charts, and history.
See the configuration reference and server release.
0.4.0
14th September 2026
Upgrade impact
- Adds the optional setting
S3_INTERNAL_BUCKET, described below. Leaving it unset keeps single-bucket routing. - No database migration.
Changes
- S3-compatible storage can use
S3_INTERNAL_BUCKETto keep internal uploads and worker artifacts in a separate private bucket while release artifacts are served publicly. Leaving it unset keeps single-bucket routing. - Adds light, dark, and system themes to the dashboard, refreshes branding, and improves accessibility. Fixes translucent callouts in dark mode.
- Release details no longer show Active counts in delivery metrics. Deployment summaries show successful installs, and release history includes downloads. Success counts are installations, not users currently running a version.
- Run
scripts/selfhost/install.sh --allow-httpfrom a local clone to test the self-hosted stack on localhost over HTTP, without TLS certificates or public DNS. Service ports bind to loopback, and OAuth setup is still required.
See the configuration reference and server release.
0.3.0
7th September 2026
Upgrade impact
- Runs migration
0013_metric_event_failure_payload, which adds the nullablefailure_payloadcolumn tometric_eventwithout backfilling existing events. - Runs migrations
0014_metric_event_failure_feedand0015_metric_event_device_outcome, which create three indexes on that table for failure details and device outcomes. Index creation is not concurrent and blocks writes to the table, so allow extra upgrade time on servers with a large metrics history. - If you use Cloudflare, delete the two-hour Edge TTL Cache Rule for
.jsonfrom the previous setup guide after you upgrade, and remove anyMANIFEST_CACHE_CONTROLline from.env.selfhost. The Cache Rule overrides the origins-maxageand extends the stale window from five minutes to two hours. A leftoverMANIFEST_CACHE_CONTROLpins the oldno-cachevalue, so manifests are never cached at the edge. See CDN settings. - No configuration change beyond the Cloudflare cleanup.
CloudFront delivery
Devices can fetch artifacts from Amazon CloudFront, in addition to Cloudflare or the origin host, for bundled MinIO or external S3/GCS. The install wizard prints the AWS Console values, generates the origin secret, and verifies the distribution before you move DNS. It does not create AWS resources or ask for AWS credentials beyond the scoped purge key. See CDN settings.
Failure details
Failed installs reported by SDK 0.3.0
include diagnostic details: HTTP status, the origin or platform error
message, and on Android 11+ the previous process exit reason. Before, the
dashboard showed only a network count. On the metrics view, open a failure
reason to see its codes, then open a code to see the values behind it. See
Analytics and metrics.
Failure counts exclude later successes
Failure counts include only devices that never got the update, not network
retries that later succeeded. When a device later installs the same package
successfully, its earlier Failed events for that package are dropped. A
late Failed after an Applied is acknowledged and discarded.
Status page
Adds a Status page to the dashboard. It shows API and database health, whether the download URL is reachable, disk use on a bundled single-host install, and the running server version compared with the latest GitHub release. The disk and version cards are hidden on topologies where they do not apply.
See the server release.
0.2.0
25th August 2026
Upgrade impact
- Rejects uploads from CLI 0.2.0 and the dashboard with
409when the fingerprint disagrees with the one recorded for the deployment and binary version, unless the client overrides it. Older clients keep the warning-only behavior. - Adds optional setup for Compose to target an external PostgreSQL database and S3-compatible bucket (see below). The default install is unchanged.
Fingerprint mismatch check
Uploads can send block_on_fingerprint_mismatch: true. When the release
fingerprint differs from the value already recorded for that deployment and
binary version, the server returns 409 and creates no release. The
dashboard compares fingerprints the same way and shows the details before you
proceed. CLI 0.2.0 and the dashboard send the flag by default. Older clients
and explicit overrides keep the previous warning-only behavior. See
Fingerprinting.
Failure reasons
Adds failure_reasons to each release in the metrics APIs and the
dashboard. It breaks failures down by category: integrity, signature,
rollout, or another category. The counts sum to failed. See
Analytics and metrics.
External Postgres and object storage
Compose can use a managed PostgreSQL database and an S3-compatible bucket
instead of the bundled Postgres and MinIO containers. The default
scripts/selfhost/install.sh run still creates the bundled stack. For an
external database or storage, pass the install script's database and storage
flags, then configure adapters as described in
Infrastructure and
Configuration reference.
The default backup script does not copy external databases or buckets. Use provider-level backups for those components. See Operations and the server release.