NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules, #1072 by repository stars
Latest stable
v0.136.0
Last release today
21 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Nearly every release is documented
notes for 60 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
2 years old
482 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
Note Compose services that use environment-backed secrets or configs, or inline config content, are recreated once after the upgrade because they now
Note
Compose services that use environment-backed secrets or configs, or inline config content, are recreated once after the upgrade because they now carry a content digest label.
Warning
The OpenAPI operation IDs for project and stack actions changed. If you use a generated API client, regenerate it.
Standalone Compose did not notice when the value of an environment-backed secret or config, or the content of an inline config, changed. The container kept running with the old value.
doco-cd now stores a per-service SHA-256 digest label (cd.doco.deployment.content.sha). When a value changes, doco-cd detects the change and recreates the services that use it.
recreate.ignore.docker inspect, so low-entropy values can be guessed offline.A long-running scheduled job could hold a stack lock and stop later polls from running, even for unrelated stacks.
Compose projects and Swarm stacks were each described by one generic OpenAPI path, even though they support several actions. Each action now has its own route with a description, operation ID, and parameters. For example, replicas is documented as required for scale.
404 Not Found.Full Changelog: v0.135.0...v0.136.0
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Note Stacks with job services scaled to zero (for example Swarm jobs with deploy.replicas: 0 ) redeploy once after the upgrade, because their project
Note
Stacks with job services scaled to zero (for example Swarm jobs with deploy.replicas: 0) redeploy once after the upgrade, because their project hash now includes the job labels.
When doco-cd deployed the Swarm stack that contains itself, the rolling update stopped the old instance while it still waited for convergence. The update worked, but the deployment was recorded as failed, counted as a deployment error, and the next deployment of the stack started with a retry.
doco-cd now updates its own service last and treats the expected shutdown as a handover:
Confirmation requires a compatible doco-cd instance with Swarm manager access. Until one runs, the report waits. Compose handovers now also retry the final status report without blocking startup.
See Self-Updating -> Failure handling.
A deployment of the stack that contains doco-cd always replaced the doco-cd container, even when the commit only changed another service. doco-cd now hands over only when its own service must be recreated:
1.Custom labels, such as a commit SHA, are not part of the config hash and do not cause a handover on their own.
A deployment that collides with an active handover of the same stack also no longer opens a second GitHub Check that hides the result of the new instance. A check that this run already opened is closed as skipped.
With the containerd image store (the default for new Docker 29 installations), Swarm pulls service images by digest. These images keep a repo@digest name and are never dangling, so prune_images left every replaced image on the nodes.
The prune job now also removes unused images whose only name is a digest reference. It keeps images that:
Swarm keeps replaced tasks as task history for rollbacks, and those tasks keep their images in use. To free these images on the next prune, lower the task history limit on a manager node:
docker swarm update --task-history-limit 1See Swarm Mode -> Image pruning.
Label-only changes to cd.doco.job.* on a job service scaled to zero were skipped with deployment skipped. The scheduler kept the old job configuration, so a new setting such as cd.doco.job.timeout was never applied. These changes now trigger a redeployment that updates the scheduler without starting the job. Services scaled to zero without job labels are still ignored for change detection.
Full Changelog: v0.134.0...v0.135.0
Nothing published for this version
Caution Breaking change API_SECRET and API_SECRET_FILE are no longer supported. Migrate to API_KEYS or API_KEYS_FILE (see below), assigning full to a…
Caution
Breaking change
API_SECRET and API_SECRET_FILE are no longer supported. Migrate to API_KEYS or API_KEYS_FILE (see below), assigning full to a key that needs the former all-access behavior (See the wiki for migration steps).
Remove the legacy variables entirely; their presence prevents startup.
API keys can now be assigned cumulative role presets or explicit permissions. Authorization is enforced across both the REST API and MCP tools: REST requests without the required permission are denied, and MCP exposes only the tools allowed by the key’s grant.
See REST API -> authentication to see how to set it up:
The role presets are cumulative:
read grants read access to health, jobs, runs, and sync windowsinspect adds Compose project and Swarm stack inspectionwrite adds workload control, job and poll triggers, and mirror compactionfull adds project and stack deletion.For more precise access, configure a key with an explicit permissions list instead of a role. See the wiki for available permissions.
Explicit permissions grant only the listed capabilities.
- key: read-only-key
role: read
- key: deploy-key
permissions:
- projects:inspect
- projects:controlPrevent hung jobs from blocking future runs with the new cd.doco.job.timeout label:
labels:
cd.doco.job.enabled: "true"
cd.doco.job.schedule: "0 2 * * *"
cd.doco.job.execution_mode: "one_off"
cd.doco.job.timeout: "1h"When the deadline expires, doco-cd gracefully stops the workload, restores any stop_services targets, marks the run as failed, and releases skip_running. Timeouts use the existing failure notifications and error metrics, and also apply to manual executions.
restart and one_off, plus Swarm one_off. Swarm restart jobs cannot use a timeout.one_off jobs retain their original deadline rather than receiving a fresh timeout. Restart-mode deadlines do not survive a doco-cd restart.stop_services does not. Graceful shutdown can extend beyond the deadline.Accepts positive Go durations such as 90s, 30m, or 1h30m. Unset or empty means unlimited runtime, preserving existing behavior.
Full Changelog: v0.133.2...v0.134.0
Fixed Compose deployments ignoring force_recreate: true . Explicitly forced deployments now recreate all services, while reconciliation preserves unch
Fixed Compose deployments ignoring force_recreate: true. Explicitly forced deployments now recreate all services, while reconciliation preserves unchanged containers.
Full Changelog: v0.133.1...v0.133.2
fix(encryption): reduce warnings for unavailable SOPS keys by @kimdre in #2155
fix(docker): preserve healthy services on readiness failure by @kimdre in #2139
Full Changelog: v0.132.0...v0.133.0
Nothing published for this version
OpenBao pki-role certificates are now issued only when a deployment runs. Before this fix, every poll or webhook check issued a new certificate, even
OpenBao pki-role certificates are now issued only when a deployment runs. Before this fix, every poll or webhook check issued a new certificate, even when nothing changed or the deployment was paused or blocked by a sync window. Doco-CD now uses placeholders for pki-role refs and their _KEY/_FULL companions while it checks for changes. It issues the real certificates only after the deployment is admitted. This cuts unneeded certificate issuance and PKI load.
Note
with INTERPOLATE_RESOLVED_SECRETS=true, certificates are still issued on every check because doco-cd needs the real values before it can check for changes.
Full Changelog: v0.131.0...v0.132.0
Nothing published for this version
Added Prometheus gauges for the state of Swarm services deployed by doco-cd. All gauges have the labels context, stack and service.
Added Prometheus gauges for the state of Swarm services deployed by doco-cd. All gauges have the labels context, stack and service.
doco_cd_service_desired_tasksdoco_cd_service_running_tasksdoco_cd_service_update_stateFull Changelog: v0.130.0...v0.131.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Security: Git credentials are now host-scoped
Important
User action required!
See also GHSA-ggcr-c46c-q2gj
Previously a global Git access token or GitHub App credential could be sent to any HTTP(S) host selected through a repository URL. Global credentials now require GIT_AUTH_DOMAINS, otherwise doco-cd will not start.
Add every Git host that should receive the global credential:
environment:
GIT_ACCESS_TOKEN: ${GIT_ACCESS_TOKEN}
GIT_AUTH_DOMAINS: |
- domains: [github.com, git.example.com]This also applies to global GitHub App credentials.
If you deliberately accept the risk of sending global credentials to unmatched hosts, you can set GIT_AUTH_ALLOW_ANY_HOST: "true".
Warning
This is insecure and not recommended.
If a stack needs many secrets, you do not have to list each one in external_secrets anymore. Set external_secrets_paths to a list of provider paths instead. Doco-CD fetches every secret below each path and makes it available as a variable:
name: myapp
external_secrets_paths:
- path: 0db45926-c97c-40d4-a3aa-fefd5d5fb492:prod:/myapp/ # Provider specific path, see the provider pages
prefix: MYAPP_ # Optional, added in front of every variable name
recursive: true # Optional, also fetch secrets in sub foldersMore info in the wiki.
Full Changelog: v0.129.1...v0.130.0
Nothing published for this version
Nothing published for this version
This release fixes some security vulnerabilities.
This release fixes some security vulnerabilities.
Full Changelog: v0.129.0...v0.129.1
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
doco-cd/<target>/deploy when a configuration target is set to keep reports for different targets separate on the same commit.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.128.0...v0.129.0
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
pre_start container lifecycle hooks. Because of this, scheduled jobs that combine cd.doco.job.execution_mode: one_off with pre_start hooks are now rejected with an error. You need to use restart execution_mode instead. See also the note in the wiki.pre_start hooks on every execution, not only on the first one.VERSION_CHECK_ENABLED (default: true) to disable the lookup and notification for new doco-cd versions on app startup.retry budgets and delays for Compose projects and Swarm stacks, see retry settings in the wiki.pause deployment setting that freezes changes without stopping existing workloads, see pause settings in the wiki.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.127.0...v0.128.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
The deprecated destroy.remove_dir deploy config option has been removed.
<!-- Release notes generated using configuration in .github/release.yml at main -->
DEPLOY_CONFIG_BASE_DIR now gets honored when selecting deployment configurations in OCI artifacts, with no root fallback.destroy.remove_dir deploy config option has been removed.one_off jobs reporting status: "created" after a successful manual API run, even though last_run.status is succeeded.component="mcp" field.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.126.0...v0.127.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
git://) URLs, allowing deployments from repositories served by simple git mirror servers like git daemon without requiring a Git hosting platform. See the wiki./v1/api/jobs) now returns more information about the latest job runs.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.125.1...v0.126.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.125.0...v0.125.1
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
Doco-CD provides a JSON Schema for deployment YAML files. Compatible editors can suggest settings and values, display descriptions and defaults on hover, and flag unknown settings or incorrect value types. More info in the wiki. At the moment the descriptions get generated from code comments in the doco-cd source code. <img width="373" height="285" alt="image" src="https://github.com/user-attachments/assets/079ad873-ca56-4ee7-8236-aec4aaeeb600" />
All settings in the deployment configuration, that previously had to be set in seconds, can now be set as duration strings like 10s, 5m or 1h45m. For example timeout: 300 can now be set as timeout: 5m. The previous numeric values in seconds remain valid.
Already decrypted and unchanged sops files get shared/reused across artifacts to reduce unecesarry decryption work (for example when the new artifact revision is only a image version bump). See the wiki for more info how doco-cd detects and decrypts encrypted files.
Successful native GitHub commit/status checks show more useful information in their summary.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.124.0...v0.125.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
> destroy.remove_dir is now deprecated and ignored. See destroy settings docs.
<!-- Release notes generated using configuration in .github/release.yml at main -->
This release brings clearer deployment status reporting, optional cleanup for unused source caches, and faster Git mirror operations. Please review the upgrade notes below for changes that may affect custom health checks, GitHub App status reporting, Prometheus queries, or stack cleanup.
live/ data. Enable it in Compose:environment:
SOURCE_GC_ENABLED: "true"
See source GC settings and what it removes.The image now uses a dedicated, smaller /healthcheck binary for each health checks. A custom Compose healthcheck.test overrides the image default.
Action: If you set it to test: ["CMD", "/doco-cd", "healthcheck"], remove the override or change it to test: ["CMD", "/healthcheck"].
See the health check docs.
Destroying a stack now preserves its shared source directory so it cannot remove data used by another stack.
[!IMPORTANT]
destroy.remove_diris now deprecated and ignored. See destroy settings docs.
Action: If you relied on that setting to reclaim unused source caches, enable source GC (disabled by default). See Destroy settings and Source GC settings
When deployment status reporting is enabled with GitHub App authentication, doco-cd now reports through GitHub Checks instead of commit statuses. Checks are separate from commit statuses in branch protection and require the App permission.
Action: Grant Checks: Read and write, approve the installation permission update, and review any required branch-protection checks. Token-based GitHub authentication continues to use commit statuses.
To reduce histogram series and scrape overhead, deployment_stage_duration_seconds no longer includes the deployment or context labels. Per-deployment timing remains available as deployment_duration_seconds.
Action: Remove deployment/context label filters or groupings from queries on the stage-duration histogram, or use the per-deployment metric.
See metrics docs.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.123.0...v0.124.0
Nothing published for this version
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
CERT_FULL) similar to CERT_KEY, see the wiki.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.122.1...v0.123.0
Nothing published for this version
Nothing published for this version
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
This release fixes an issue with self-updating where the deploying instance posted a "Pending - In Progress" commit status and then handed over, while the instance finishing the handover never updated the commit status.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.122.0...v0.122.1
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
A long requested feature is now fully supported; Doco-CD is now able to update itself, even without a secondary updater instance. See the Self-Updating wiki page for more information and a migration guide.
This release also fixes a bunch of bugs that were introduced after the artifact storage update in v0.120.0.
[!NOTE] After upgrading, every Swarm service gets one rolling update. Services that use
recreate.ignoreare recreated once when they move to the new live copy. Standalone commits that only changerecreate.ignorefiles now run a deployment instead of being skipped. Nothing is recreated, but it sends the "Deployment completed" notification if notifications are configured.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.121.0...v0.122.0
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.120.1...v0.121.0
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
Fixed a crash that could occur when concurrent webhook deployments fetched updates into the same Git repository mirror. Git reads now use fresh, lock-scoped repository handles, keeping parallel deployments safe and ensuring each deployment reports the revision it actually deployed.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.120.0...v0.120.1
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
This update improves deployment safety, concurrency, and performance, especially for repositories with multiple auto-discovered stacks.
Git repositories and OCI sources are now prepared as immutable, content-addressed artifacts. Deployments use a dedicated artifact for each revision instead of a shared mutable working tree. This allows independent stacks to deploy concurrently, while deployments for the same stack remain serialized. Older webhook events are also prevented from overwriting newer deployments.
Source preparation and Git checks are more efficient:
git_depth match.Read-only pre-deployment work, such as source initialization and change detection, is now handled separately from Docker-mutating deployment work. This allows more preparation tasks to run concurrently without consuming deployment slots.
You can find more information in the artifact storage documentation.
| Variable | Default | Purpose |
|---|---|---|
MAX_CONCURRENT_PREDEPLOYMENTS |
8 |
Maximum number of concurrent read-only pre-deployment operations, including initialization and change detection. |
ARTIFACT_GC_ENABLED |
true |
Enables cleanup of unreferenced Git revision and OCI digest artifacts. |
ARTIFACT_GC_RETENTION_RECORDS |
2 |
Number of recent unreferenced artifacts to keep per repository or artifact. |
ARTIFACT_GC_RETENTION_TTL |
1m |
Minimum time older unreferenced artifacts are retained before removal. |
ARTIFACT_GC_INTERVAL |
10m |
How often artifact cleanup runs. A sweep also runs at startup. |
MAX_CONCURRENT_DEPLOYMENTS continues to control the number of deployments that may perform Docker mutations. MAX_CONCURRENT_PREDEPLOYMENTS controls preparation separately and can be adjusted according to available CPU, network, and storage capacity.
ARTIFACT_GC_ENABLED if older artifacts must remain available.[!CAUTION] Services that write inside the cloned repository with relative bind mounts (inside the doco-cd data directory/volume) will lose their data when the service is re-/deployed from a new artifact revision. Each artifact revision/version is like a clean, new Git worktree: When doco-cd deploys from a new Git commit or OCI artifact version, it first creates a new
artifacts/<revision>worktree directory to deploy from. If you use Pre- / Post-Deployment Scripts to generate configuration or data, it might be recommended to use named volumes or absolute host paths depending on your use case.
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.119.0...v0.120.0
Nothing published for this version
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
${...} placeholder unchanged instead of returning an error. Previously, doco-cd passed that unresolved value through to the deployment. Now the Infisical provider fails fast when the returned secret still contains an Infisical reference expression.stop_services now stay stopped correctly while the scheduled job is still running.Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.118.1...v0.119.0
<!-- Release notes generated using configuration in .github/release.yml at main -->
<!-- Release notes generated using configuration in .github/release.yml at main -->
Full Changelog: https://github.com/kimdre/doco-cd/compare/v0.118.0...v0.118.1
One column per month.