These manifests are the inspectable, low-level self-serve Kubernetes starting point. The published Durable Workflow Helm chart is the recommended path for most operators; the raw manifests stay supported for teams that intentionally do not want Helm in the rollout.
Both paths share the same external-persistence contract, the same singleton
scheduler invariant, and the same /api/ready-based readiness contract.
Pick one or the other per environment, not both.
The checked-in manifests are synchronized with the repository's stable source release and pin its Docker Hub tag:
durableworkflow/server:2.11.8
The raw ConfigMap sets LOG_CHANNEL=stderr for the API, worker and scheduler.
Read application logs with kubectl logs or your cluster's log collector. This
also works with existing published images when the setting is supplied explicitly.
Set LOG_LEVEL to select the minimum severity. For newline-delimited JSON, add
this ConfigMap entry:
LOG_STDERR_FORMATTER: 'Monolog\Formatter\JsonFormatter'An explicit LOG_CHANNEL override selects another Laravel channel. File channels
need their own persistent storage and collection; Kubernetes does not collect
storage/logs/laravel.log automatically. Restart persistent roles after changing
logging settings because configuration is cached at startup. DW_LOG_FORMAT is
not a supported setting.
Before production use, patch every workload image to the exact published tag or digest you intend to run:
kubectl set image -n durable-workflow deploy/durable-workflow-server \
server=durableworkflow/server:2.11.8
kubectl set image -n durable-workflow deploy/durable-workflow-worker \
worker=durableworkflow/server:2.11.8
kubectl set image -n durable-workflow cronjob/durable-workflow-scheduler \
scheduler=durableworkflow/server:2.11.8GitHub Container Registry publishes the same release line at
ghcr.io/durable-workflow/server:2.11.8. Digest pinning is preferred for strict
change control.
The manifests expect you to provide:
- an external MySQL or PostgreSQL database;
- external Redis or another supported lock-capable cache backend;
- real database, Redis, worker, operator, admin, and optional named principal token secrets;
- an ingress, gateway, or load balancer owned by your cluster platform;
- backup, restore, monitoring, and rollout procedures for your environment.
The included contract is deliberately bounded:
k8s/migration-job.yamlrunsserver-bootstrapbefore workloads start;k8s/server-deployment.yamlexposes/api/healthfor liveness and/api/readyfor usable readiness;k8s/worker-deployment.yamlruns the queue worker;k8s/scheduler-cronjob.yamlruns recurring schedule, unscoped task repair, timeout, and retention maintenance;k8s/secret.yamlseparates public config from app-level secrets and refers to externally managed database and Redis credentials.
Helm charts are now self-serve via helm/durable-workflow/;
managed-Kubernetes provider validation, advanced HA, active/active multi-region,
custom operators, storage classes, network policies, and environment-specific
security hardening are support-led or tracked separately.
Active/passive multi-region with operator-driven regional failover follows the
contract in docs/multi-region-validation.md;
each region still runs the documented single-region or small-cluster shape, and
this manifest contract does not add automatic cross-region orchestration. Use
overlays or direct
patches for namespace, image, resource, replica, ingress, and secret-manager
integration choices.