RAGFlow on GKE Autopilot

RAGFlow is an open-source document intelligence and Retrieval-Augmented Generation (RAG) platform. It ingests PDFs, Word documents, HTML pages, and other formats, chunks and embeds them, stores vectors in Elasticsearch, exposes a REST API for question-answering, and provides a web UI for knowledge base management and enterprise search. This module deploys RAGFlow on GKE Autopilot on top of the App_GKE foundation, which provisions and manages the shared Google Cloud and Kubernetes infrastructure.

This guide focuses on the cloud services RAGFlow uses and how to explore and operate them from the Google Cloud Console and the command line. For the mechanics that are common to every GKE application — Workload Identity, ingress, autoscaling, CI/CD, Cloud Armor, IAP, Binary Authorization, VPC Service Controls, backups, and the deployment lifecycle — refer to the App_GKE foundation guide rather than repeating them here.

Deployment prerequisite: RAGFlow_GKE requires Elasticsearch_GKE to be deployed first. The elasticsearch_hosts variable is mandatory — the plan is rejected if it is empty when deploy_application = true.

---

1. Overview

RAGFlow runs as a containerised Python/Nginx web workload. The deployment wires together a focused set of Google Cloud services:

Capability	Google Cloud service	Notes
Compute	GKE Autopilot	Custom-built RAGFlow pods, 4 vCPU / 8 GiB by default, horizontally autoscaled
Database	Cloud SQL for MySQL 8.0	Required — RAGFlow does not support PostgreSQL
Vector search	Elasticsearch (Elasticsearch_GKE)	External dependency — must be deployed first; elasticsearch_hosts is mandatory
Task queue	Redis (Memorystore)	Required for document processing workers
Shared files	Filestore (NFS)	Enabled by default for shared document processing storage
Object storage	Cloud Storage	A dedicated ragflow-documents bucket
Secrets	Secret Manager	Auto-generated database password
Ingress	Cloud Load Balancing	External LoadBalancer service; optional custom domain + managed certificate

Sensible defaults worth knowing up front:

• MySQL 8.0 is mandatory. The database engine is fixed; selecting PostgreSQL or NONE breaks startup.
• elasticsearch_hosts is required. RAGFlow cannot index or search documents without a reachable Elasticsearch endpoint. Deploy Elasticsearch_GKE first.
• Redis is required for document processing. With enable_redis = true (the default), REDIS_HOST and REDIS_PORT are injected automatically. Without Redis, uploaded files remain unprocessed indefinitely.
• Scale-to-zero is disabled. min_instance_count is hard-capped at 1. RAGFlow loads embedding models at startup (2–3 minutes); scale-to-zero would cause requests to time out.
• A custom image is always built. Cloud Build extends infiniflow/ragflow using the Dockerfile in RAGFlow_Common/scripts/, with APP_VERSION set from application_version.
• service_conf.yaml is generated at startup. The custom entrypoint writes /ragflow/conf/service_conf.yaml from injected environment variables before starting the RAGFlow processes.
• Session affinity is ClientIP. This ensures upload sessions and multi-step document processing requests consistently reach the same pod.
• enable_custom_domain defaults to true. This routes the service through the Gateway API load balancer, which — combined with the default reserve_static_ip = true — provisions a managed certificate and HTTPS out of the box via a <ip>.nip.io hostname even when application_domains is left empty. Set application_domains to serve a real hostname instead.
• The database password is generated automatically and stored in Secret Manager.

---

2. Google Cloud Services & How to Explore Them

All commands assume you have run gcloud container clusters get-credentials <cluster> --region <region> --project <project> and that PROJECT, REGION, and NAMESPACE are set. The namespace and other identifiers are reported in the deployment Outputs.

A. GKE Autopilot — the RAGFlow workload

RAGFlow pods are scheduled on Autopilot, which bills for the CPU/memory the pods actually request. Horizontal Pod Autoscaling sizes the deployment between the minimum and maximum replica counts.

• Console: Kubernetes Engine → Workloads → select the RAGFlow workload to see pods, revisions, and events. Kubernetes Engine → Services & Ingress shows the external IP.
• CLI:

  kubectl get pods,svc,hpa -n "$NAMESPACE"
  kubectl logs -n "$NAMESPACE" deploy/<service-name> --tail=100
  kubectl describe hpa -n "$NAMESPACE"          # current vs target utilisation

See App_GKE for how Autopilot, scaling, and the workload type (Deployment vs StatefulSet) are managed.

B. Cloud SQL for MySQL 8.0

RAGFlow stores all application metadata (user accounts, knowledge bases, task state) in a managed Cloud SQL for MySQL 8.0 instance. Pods reach it privately through the Cloud SQL Auth Proxy sidecar over a Unix socket, so no public IP is exposed. On first deploy an initialization Job creates the application database and user.

• Console: SQL → select the instance for connections, backups, flags, and metrics.
• CLI:

  gcloud sql instances list --project "$PROJECT"
  gcloud sql instances describe <instance-name> --project "$PROJECT"
  # Open an interactive shell to inspect schema/data:
  gcloud sql connect <instance-name> --user=<db-user> --project "$PROJECT"

The instance name, database name, user, and the Secret Manager secret holding the password are all surfaced in the Outputs. For the connection model, automated backups, and password rotation, see App_GKE.

C. Elasticsearch — vector search

RAGFlow requires an external Elasticsearch instance for document indexing and vector search. It is not bundled in this module — deploy Elasticsearch_GKE separately and pass its elasticsearch_endpoint output as elasticsearch_hosts. The environment variables ELASTICSEARCH_HOSTS and ELASTICSEARCH_USERNAME are injected automatically.

• Console: Kubernetes Engine → Workloads → Elasticsearch namespace.
• CLI:

  # Confirm RAGFlow can reach Elasticsearch:
  kubectl exec -n "$NAMESPACE" deploy/<service-name> -- \
    curl -s "$ELASTICSEARCH_HOSTS/_cluster/health" | grep status
  # Check connectivity from within the cluster:
  kubectl run -it --rm curl --image=curlimages/curl --restart=Never -- \
    curl -s "http://<elasticsearch-ip>:9200/_cluster/health"

D. Redis — task queue

Redis is the backbone of RAGFlow's document processing pipeline. Workers poll Redis for tasks; without it, uploaded files are never parsed, chunked, or embedded.

• Console: Memorystore → Redis (if using a managed instance).
• CLI:

  redis-cli -h <redis-host> ping        # from a host with network access
  redis-cli -h <redis-host> info keyspace

E. Filestore (NFS) and Cloud Storage

By default, RAGFlow mounts a Filestore (NFS) share into every pod for shared document processing storage. A dedicated Cloud Storage bucket (ragflow-documents) is also provisioned for document ingestion; the workload service account is granted access automatically. Additional GCS Fuse volumes can be mounted via gcs_volumes.

• Console: Filestore → Instances; Cloud Storage → Buckets.
• CLI:

  gcloud filestore instances list --project "$PROJECT"
  gcloud storage buckets list --project "$PROJECT"
  gcloud storage ls gs://<documents-bucket>/     # bucket name is in the Outputs
  # Confirm the NFS share is mounted inside a pod:
  kubectl exec -n "$NAMESPACE" deploy/<service-name> -- df -h | grep -i nfs

See App_GKE for NFS provisioning, GCS Fuse, and CMEK options.

F. Secret Manager

The database password is stored as a Secret Manager secret and injected into pods at runtime; plaintext never appears in configuration.

• Console: Security → Secret Manager.
• CLI:

  gcloud secrets list --project "$PROJECT"
  gcloud secrets versions access latest --secret=<secret-name> --project "$PROJECT"

The database password secret name is in the Outputs. See App_GKE for the Secret Store CSI integration and rotation.

G. Networking & ingress

enable_custom_domain defaults to true, which routes the workload through the Gateway API load balancer with a Google-managed certificate — automatically serving HTTPS on a <ip>.nip.io hostname when application_domains is empty. Set application_domains to serve a real custom hostname instead. A static IP is reserved by default so the address survives redeploys.

• Console: Network services → Load balancing; VPC network → IP addresses.
• CLI:

  kubectl get ingress,svc -n "$NAMESPACE"
  gcloud compute addresses list --project "$PROJECT"

See App_GKE for custom domains, Cloud CDN, and static IP details.

H. Cloud Logging & Monitoring

Pod stdout/stderr flow to Cloud Logging; GKE and Cloud SQL metrics flow to Cloud Monitoring. Optional uptime checks and alert policies are available.

• Console: Logging → Logs Explorer; Monitoring → Dashboards / Alerting.
• CLI:

  gcloud logging read 'resource.type="k8s_container" AND resource.labels.namespace_name="'"$NAMESPACE"'"' \
    --project "$PROJECT" --limit 50

---

3. RAGFlow Application Behaviour

• First-deploy database setup. An initialization Job (mysql:8.0-debian) creates the RAGFlow database (rag_flow) and user (ragflow) and grants privileges before the application starts. It is idempotent and safe to re-run.
• service_conf.yaml generated at startup. The custom entrypoint generates /ragflow/conf/service_conf.yaml from injected environment variables — including MySQL host/user/database, Elasticsearch endpoint, Redis host/port, and optional credentials — before starting the RAGFlow processes (Nginx, RAGFlow server, task workers).
• Document processing pipeline. Uploaded documents are queued in Redis and processed asynchronously by RAGFlow's document workers: OCR, chunking, embedding, and indexing into Elasticsearch. Without a reachable Redis and Elasticsearch, files appear uploaded but are never processed.
• Embedding model loading on startup. RAGFlow downloads and loads embedding models during first boot. The startup probe targets /v1/health with a 120-second initial delay and 60 failure retries to allow ample time. Liveness transitions to /v1/health once startup succeeds.
• Health endpoints. The readiness and liveness probes use /v1/health and /v1/system/version. These return HTTP 200 only when the application is fully initialised and all services are reachable.
• Session affinity. session_affinity = "ClientIP" routes requests from a browser to the same pod, ensuring upload sessions and in-progress document workflows remain consistent.

---

4. Configuration Variables

Variables are grouped exactly as they appear on the deployment platform. Only settings specific to or notable for RAGFlow are listed; every other input is inherited from App_GKE with its standard behaviour and defaults.

Group 1 — Project & Identity

Variable	Default	Description
project_id	(required)	Target Google Cloud project.
region	us-central1	Region for the workload and regional resources.
elasticsearch_hosts	(required)	Elasticsearch HTTP endpoint (e.g. http://10.0.0.5:9200). Set to the elasticsearch_endpoint output from Elasticsearch_GKE.

Group 2 — Deployment Environment

Variable	Default	Description
tenant_deployment_id	demo	Short suffix that makes resource names unique per environment.
support_users	[]	Emails granted project access and monitoring alerts.
resource_labels	{}	Labels applied to all resources for cost/ownership tracking.

Group 3 — Application Identity

Variable	Default	Description
application_name	ragflow	Base name for resources. Do not change after first deploy.
application_display_name	RAGFlow	Friendly name shown in the Console.
application_description / description	(set)	Workload description annotation.
application_version	v0.13.0	RAGFlow image version tag passed as APP_VERSION to Cloud Build; increment to roll out a new version.

Group 4 — Runtime & Scaling

Variable	Default	Description
deploy_application	true	Set false to provision infrastructure only.
cpu_limit	4000m	CPU per pod; 4 vCPU recommended for document parsing workloads.
memory_limit	8Gi	Memory per pod; 8 GiB recommended (embedding models are large).
min_instance_count	1	Minimum replicas. Hard-capped at 1 — scale-to-zero is not supported.
max_instance_count	5	Maximum replicas (autoscaler ceiling).
container_port	80	RAGFlow's Nginx frontend listens on port 80.
enable_cloudsql_volume	true	Cloud SQL Auth Proxy sidecar for socket connections.
termination_grace_period_seconds	60	Increase for in-flight document processing jobs.
deployment_timeout	1800	Seconds Terraform waits for the rollout — generous to accommodate model loading.
enable_vertical_pod_autoscaling	false	Let Autopilot tune resource requests automatically.

Group 5 — Environment Variables & Secrets

Variable	Default	Description
environment_variables	{}	Extra non-secret settings. Do not override MYSQL_*, ELASTICSEARCH_*, or REDIS_* — these are injected automatically.
secret_environment_variables	{}	Map of env var → Secret Manager secret name.

Group 6 — GKE Backend & Cluster

Variable	Default	Description
service_type	LoadBalancer	How the Service is exposed.
session_affinity	ClientIP	Sticky routing — required for consistent upload sessions and document processing.
workload_type	null	Auto-resolves to StatefulSet when per-pod storage is enabled.
network_tags	["nfsserver"]	Node/pod tags; nfsserver is required for NFS connectivity.

Group 7 — StatefulSet Configuration

These settings apply only when workload_type = "StatefulSet" or stateful_pvc_enabled = true.

Variable	Default	Description
stateful_pvc_enabled	null	Provision a PVC for local model storage. Recommended for production.
stateful_pvc_size	10Gi	PVC size; provision at least 50 GiB for production document workloads.
stateful_pvc_mount_path	/data	Container mount path for the PVC.
stateful_pvc_storage_class	standard-rwo	Kubernetes StorageClass for PVCs.
stateful_fs_group	0	fsGroup GID for PVC ownership; set if the container process requires a specific GID.

Group 8 — Resource Quota

Variable	Default	Description
enable_resource_quota	false	Cap namespace CPU/memory/object counts.
quota_memory_requests / quota_memory_limits	""	Must use binary units (8Gi, 16384Mi) — bare integers are read as bytes and block scheduling.

Group 9 — Reliability Policies

Variable	Default	Description
enable_pod_disruption_budget	true	Protect availability during node upgrades.
pdb_min_available	1	Raise min_instance_count above 1 if you need eviction headroom.
enable_topology_spread	false	Spread pods across zones.

Group 10 — Observability & Health

Variable	Default	Description
startup_probe / startup_probe_config	/v1/health, 120s delay, 60 retries	HTTP probe allowing ample time for embedding model loading.
liveness_probe / health_check_config	/v1/health	HTTP liveness probe once startup succeeds.
uptime_check_config	{ enabled=false, path="/v1/health" }	Optional Cloud Monitoring uptime check; disabled by default.
alert_policies	[]	Optional metric alert policies.

Group 11 — Jobs & Scheduled Tasks

Variable	Default	Description
initialization_jobs	[]	Leave empty to use the built-in db-init job.
cron_jobs	[]	Recurring scheduled tasks deployed as Kubernetes CronJobs.
additional_services	[]	Additional companion Kubernetes Deployments (e.g. inline Elasticsearch for dev).

Group 12 — CI/CD & GitHub Integration

Standard App_GKE Cloud Build / Cloud Deploy integration — see App_GKE. Key inputs: enable_cicd_trigger, github_repository_url, github_token, enable_cloud_deploy.

Group 13 — Filesystem (NFS)

Variable	Default	Description
enable_nfs	true	Shared Filestore volume for document processing storage (keep enabled for multi-replica).
nfs_mount_path	/mnt/nfs	Mount path inside the container.
nfs_instance_name	""	Name of an existing NFS GCE VM; auto-discovered when empty.

Group 14 — Cloud Storage & Artifact Registry

Variable	Default	Description
create_cloud_storage	true	Provision the GCS document bucket.
storage_buckets	[{ name_suffix="data" }]	Additional GCS buckets beyond ragflow-documents.
gcs_volumes	[]	GCS Fuse volumes mounted into the RAGFlow container.
manage_storage_kms_iam / enable_artifact_registry_cmek	false	CMEK encryption options.
max_images_to_retain	7	Maximum recent Artifact Registry images to keep.

Group 15 — Elasticsearch & Redis

Variable	Default	Description
enable_redis	true	Use Redis for the document processing task queue.
redis_host	""	Redis host. When empty and NFS is enabled, the NFS server IP is used. Set explicitly for Memorystore.
redis_port	6379	Redis port.
redis_auth	""	Optional Redis auth password (sensitive).
elasticsearch_username	""	Elasticsearch username. Leave empty when xpack.security.enabled = false.
enable_inline_elasticsearch	false	Deploy a single-node Elasticsearch sidecar. Development only — data is lost on pod restart.
elasticsearch_version	8.13.0	Elasticsearch image tag for the inline instance.

Group 16 — Database Backend

Variable	Default	Description
database_type	MYSQL_8_0	Fixed — do not change.
db_name	rag_flow	Database name. Immutable after first deploy.
db_user	ragflow	Application user. Immutable after first deploy.
database_password_length	32	Generated password length (16–64).
enable_auto_password_rotation	false	Zero-downtime DB password rotation.

Group 17 — Backup & Maintenance

Variable	Default	Description
backup_schedule	0 2 * * *	Automated backup cron (UTC).
backup_retention_days	7	Retention; raise to 30–90 for production/compliance.
enable_backup_import / backup_source / backup_uri	restore options	Restore from a backup on deploy.

Group 18 — Custom SQL Scripts

enable_custom_sql_scripts, custom_sql_scripts_bucket, custom_sql_scripts_path, custom_sql_scripts_use_root — run SQL from a GCS bucket after provisioning. See App_GKE.

Group 19 — Custom Domain, Static IP & Networking

Variable	Default	Description
enable_custom_domain	true	Provision Ingress for custom hostnames + managed certificate.
application_domains	[]	Hostnames to serve.
reserve_static_ip	true	Stable external IP across redeploys.
network_tags	["nfsserver"]	GCP network tags applied to pods.

Group 20 — Identity-Aware Proxy (IAP)

Variable	Default	Description
enable_iap	false	Require Google sign-in in front of RAGFlow.
iap_authorized_users / iap_authorized_groups	[]	Who may access.
iap_oauth_client_id / iap_oauth_client_secret	""	Required when IAP is enabled (sensitive).
iap_support_email	""	Shown on the OAuth consent screen.

Group 21 — Cloud Armor

Variable	Default	Description
enable_cloud_armor	false	Attach a Cloud Armor (WAF) policy to the Ingress backend.
admin_ip_ranges	[]	CIDRs allowed privileged access.
cloud_armor_policy_name	default-waf-policy	Policy name.

Group 22 — VPC Service Controls & Audit Logging

Variable	Default	Description
enable_vpc_sc	false	Enforce a VPC-SC perimeter (requires organization_id).
vpc_cidr_ranges / vpc_sc_dry_run	(set)	Access level CIDRs / dry-run mode.
enable_audit_logging	false	Detailed Cloud Audit Logs.

---

5. Outputs

These values are returned on a successful deployment and are the quickest way to locate and explore the running resources.

Output	Description
service_name	Kubernetes Service name.
namespace	Namespace the workload runs in.
service_cluster_ip	In-cluster ClusterIP.
stage_service_cluster_ips	Map of ClusterIPs for stage-specific services (Cloud Deploy).
service_external_ip	External LoadBalancer IP (when a static IP is reserved).
service_url	URL to reach RAGFlow.
database_instance_name	Cloud SQL instance name.
database_name	Application database name.
database_user	Application database user.
database_password_secret	Secret Manager secret holding the DB password.
database_host / database_port	DB endpoint (127.0.0.1 via the Auth Proxy) / port.
storage_buckets	Created Cloud Storage buckets.
network_name / network_exists / regions	VPC network, presence, available regions.
container_image / container_registry	Deployed image and Artifact Registry repo.
monitoring_enabled / monitoring_notification_channels	Monitoring status and channels.
initialization_jobs / db_import_job	Names of the setup and (optional) import jobs.
deployment_id / tenant_id / resource_prefix	Naming identifiers.
project_id / project_number	Project identifiers.
cicd_enabled / cicd_configuration	CI/CD status and details.
github_repository_url / github_repository_owner / github_repository_name	Connected repository details.
artifact_registry_repository / cloudbuild_trigger_name / cloudbuild_trigger_id	Registry and build trigger.
kubernetes_ready	Whether the cluster/workload is ready. false on first apply of a new inline cluster — the CI/CD pipeline must re-run apply to complete deployment.
vpc_sc_enabled / vpc_sc_perimeter_name / vpc_sc_dry_run_mode	VPC-SC status.
audit_logging_enabled / artifact_registry_cmek_enabled	Audit logging and CMEK status.

---

6. Configuration Pitfalls & Sensible Defaults

Risk: Critical (data loss / outage / security) — High (service degraded) — Medium (cost or partial degradation) — Low (minor).

Setting	Sensible value	Risk	Consequence if wrong
elasticsearch_hosts	required — set from Elasticsearch_GKE	Critical	RAGFlow cannot index or search; all ingestion and retrieval operations fail. Plan is rejected when empty and deploy_application = true.
enable_redis	true	Critical	Without Redis the document processing task queue never runs; uploaded files remain unprocessed indefinitely.
database_type	MYSQL_8_0	Critical	RAGFlow requires MySQL; PostgreSQL/NONE breaks startup.
enable_cloudsql_volume	true	Critical	RAGFlow connects via Unix socket; disabling the proxy sidecar causes a database connection failure on startup.
db_name / db_user	set once	Critical	Immutable after first deploy; renaming recreates the database/user and destroys data.
enable_backup_import	false unless restoring	Critical	Enabling without a valid backup_uri fails the import job.
quota_memory_requests / _limits	binary units	Critical	Bare integers are bytes and block all scheduling.
redis_host	explicit Memorystore IP or "" (NFS fallback)	High	An unreachable or wrong Redis host silently breaks all async document workers.
min_instance_count	1	High	0 causes scale-to-zero; cold starts take 2–3 minutes and requests time out.
memory_limit	8Gi	High	Embedding models plus the application server typically require 4–8 GiB; too little causes OOM kills during document processing.
stateful_pvc_enabled	true for production	High	Without a PVC, pod restarts lose all in-progress processing state.
session_affinity	ClientIP	High	Without stickiness, multi-replica upload sessions may split across pods.
elasticsearch_username	"" or correct user	High	If Elasticsearch security is enabled, leaving this blank causes HTTP 401 and breaks all indexing.
enable_nfs	true	High	Multi-replica deployments without shared storage see inconsistent document views across pods.
enable_iap / enable_cloud_armor	enable for production	Medium	RAGFlow's web UI is otherwise publicly reachable.
backup_retention_days	7 (raise for prod)	Medium	Too short for compliance retention.
pdb_min_available vs min_instance_count	leave headroom	Medium	1/1 can stall node upgrades (single pod cannot be evicted).
application_version	v0.13.0	Medium	Incrementing triggers an image rebuild and rolling restart; verify MySQL schema compatibility for major version jumps.

---

For the foundation behaviour referenced throughout — IAM and Workload Identity, autoscaling, ingress and certificates, CI/CD, Cloud Armor, IAP, Binary Authorization, VPC-SC, backups, and image mirroring — see App_GKE. RAGFlow-specific application configuration shared with the Cloud Run variant is described in RAGFlow_Common.