Skip to content

Self-host Cache

The Tuist cache service can be self-hosted to provide a private binary cache for your team. This is most useful for organizations with large artifacts and frequent builds, where placing the cache closer to your CI infrastructure reduces latency and improves cache efficiency. By minimizing the distance between your build agents and the cache, you ensure that network overhead doesn't negate the speed benefits of caching.

Info

Self-hosting cache nodes requires an Enterprise plan.

You can connect self-hosted cache nodes to either the hosted Tuist server (https://tuist.dev) or a self-hosted Tuist server. Self-hosting the Tuist server itself requires a separate server license. See the server self-hosting guide.

Prerequisites#

  • Docker and Docker Compose
  • S3-compatible storage bucket
  • A running Tuist server instance (hosted or self-hosted)

Deployment#

The cache service is distributed as a Docker image at ghcr.io/tuist/cache. We provide reference configuration files in the cache directory.

Tip

We provide a Docker Compose setup because it's a convenient baseline for evaluation and small deployments. You can use it as a reference and adapt it to your preferred deployment model (Kubernetes, raw Docker, etc.).

Configuration files#

bash
curl -O https://raw.githubusercontent.com/tuist/tuist/main/cache/docker-compose.yml
mkdir -p docker
curl -o docker/nginx.conf https://raw.githubusercontent.com/tuist/tuist/main/cache/docker/nginx.conf

Environment variables#

Create a .env file with your configuration.

Tip

The service is built with Elixir/Phoenix, so some variables use the PHX_ prefix. You can treat these as standard service configuration.

bash
# Secret key used to sign and encrypt data. Minimum 64 characters.
# Generate with: openssl rand -base64 64
SECRET_KEY_BASE=YOUR_SECRET_KEY_BASE
# Public hostname or IP address where your cache service will be reachable.
PUBLIC_HOST=cache.example.com
# URL of the Tuist server used for authentication (REQUIRED).
# - Hosted: https://tuist.dev
# - Self-hosted: https://your-tuist-server.example.com
SERVER_URL=https://tuist.dev
# S3 Storage configuration
S3_BUCKET=your-cache-bucket
S3_HOST=s3.us-east-1.amazonaws.com
S3_REGION=us-east-1
# Optional: dedicated Xcode cache bucket.
# Useful when you want separate retention policies, storage classes,
# or cost tracking for Xcode cache vs. module/Gradle cache artifacts.
# S3_XCODE_CACHE_BUCKET=your-xcode-cache-bucket
# S3 authentication (Option 1: static credentials)
S3_ACCESS_KEY_ID=your-access-key
S3_SECRET_ACCESS_KEY=your-secret-key
# S3 authentication (Option 2: IAM role / IRSA)
# Omit S3_ACCESS_KEY_ID and S3_SECRET_ACCESS_KEY, and set
# AWS_WEB_IDENTITY_TOKEN_FILE and AWS_ROLE_ARN instead.
# These are typically injected automatically by EKS or similar platforms.
# CAS storage (required for non-compose deployments)
STORAGE_DIR=/storage
# Optional dedicated KV SQLite database path.
# Defaults to /data/key_value.sqlite.
KEY_VALUE_DATABASE_PATH=/data/key_value.sqlite
Variable Required Default Description
SECRET_KEY_BASE Yes Secret key used to sign and encrypt data (minimum 64 characters).
PUBLIC_HOST Yes Public hostname or IP address of your cache service. Used to generate absolute URLs.
SERVER_URL No https://tuist.dev URL of your Tuist server for authentication.
STORAGE_DIR Yes Directory where CAS artifacts are stored on disk. The provided Docker Compose setup uses /storage.
KEY_VALUE_DATABASE_PATH No /data/key_value.sqlite Path to the dedicated SQLite database used by the key-value store.
POOL_SIZE No 2 Connection pool size for the primary metadata SQLite database.
KEY_VALUE_POOL_SIZE No POOL_SIZE Connection pool size for the dedicated key-value SQLite database.
S3_BUCKET Yes S3 bucket for module and Gradle cache artifacts. Also used for Xcode cache artifacts when S3_XCODE_CACHE_BUCKET is unset.
S3_XCODE_CACHE_BUCKET No S3_BUCKET Optional dedicated bucket for Xcode cache artifacts. When set, Xcode cache reads and writes use this bucket directly. Useful when you want separate retention policies, storage classes, or cost tracking for Xcode cache artifacts.
S3_HOST Yes S3 endpoint hostname (e.g. s3.us-east-1.amazonaws.com).
S3_REGION Yes S3 region. Also accepted as AWS_REGION.
S3_ACCESS_KEY_ID Conditional S3 access key. Required when using static credentials. Also accepted as AWS_ACCESS_KEY_ID. See S3 authentication.
S3_SECRET_ACCESS_KEY Conditional S3 secret key. Required when using static credentials. Also accepted as AWS_SECRET_ACCESS_KEY. See S3 authentication.
S3_ENDPOINT No Full S3 endpoint URL. When set, overrides S3_HOST with the parsed host and scheme. Useful for S3-compatible providers.
AWS_WEB_IDENTITY_TOKEN_FILE No Path to a web identity token file for IAM role authentication. See S3 authentication.
AWS_ROLE_ARN No IAM role ARN to assume when using web identity token authentication.
DISK_HIGH_WATERMARK_PERCENT No 85 Disk usage percentage that triggers LRU eviction.
DISK_TARGET_PERCENT No 70 Target disk usage after eviction.
KEY_VALUE_MAX_DB_SIZE_BYTES No 26843545600 Maximum size of the dedicated key-value SQLite database before size-based KV eviction starts.
KEY_VALUE_EVICTION_MIN_RETENTION_DAYS No 1 Minimum age a key-value entry must reach before size-based KV eviction can remove it.
KEY_VALUE_EVICTION_MAX_DURATION_MS No 300000 Maximum runtime for a single KV eviction pass.
KEY_VALUE_EVICTION_HYSTERESIS_RELEASE_BYTES No 24696061952 Target size after KV eviction finishes, providing hysteresis so the worker does not thrash near the limit.
KEY_VALUE_READ_BUSY_TIMEOUT_MS No 2000 SQLite busy-timeout (in milliseconds) for KV read-through requests. If the database is locked longer than this, the read is treated as a cache miss and the value is fetched from S3.
KEY_VALUE_MAINTENANCE_BUSY_TIMEOUT_MS No 50 SQLite busy-timeout (in milliseconds) for background maintenance operations (PRAGMA queries, incremental vacuum). A low value prevents maintenance from blocking read traffic.
PHX_SOCKET_PATH No /run/cache/cache.sock Path where the service creates its Unix socket (when enabled).
PHX_SOCKET_LINK No /run/cache/current.sock Symlink path that Nginx uses to connect to the service.
OTEL_EXPORTER_OTLP_ENDPOINT No gRPC endpoint of an OpenTelemetry Collector for distributed tracing.
LOKI_URL No Base URL of a Loki-compatible endpoint for log forwarding.
DEPLOY_ENV No production Environment label used in traces and log labels (e.g. production, staging).

S3 authentication#

The cache service supports multiple methods for authenticating with S3. The method is determined automatically based on which environment variables are set.

Static credentials#

Set S3_ACCESS_KEY_ID and S3_SECRET_ACCESS_KEY (or AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY). This is the simplest method and works with any S3-compatible provider.

bash
S3_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
S3_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY

IAM role with web identity (IRSA)#

If you run the cache service on Kubernetes with EKS, you can authenticate using IAM Roles for Service Accounts (IRSA) or EKS Pod Identity. Omit S3_ACCESS_KEY_ID and S3_SECRET_ACCESS_KEY, and ensure the following environment variables are available to the container:

  • AWS_WEB_IDENTITY_TOKEN_FILE — path to the projected service account token (injected by EKS)
  • AWS_ROLE_ARN — the IAM role to assume

EKS injects these variables automatically when a service account is annotated with the IAM role.

EC2 instance profile#

When neither static credentials nor a web identity token file are present, the service falls back to the default AWS credential chain. This means it can authenticate using an EC2 instance profile or ECS task role without any additional configuration — just ensure the instance or task has an IAM role with S3 access.

Start the service#

bash
docker compose up -d

Verify the deployment#

bash
curl http://localhost/up

Configure the cache endpoint#

After deploying the cache service, register it with your Tuist server:

  • Hosted Tuist server (https://tuist.dev):

    1. Navigate to your organization's Settings page.
    2. Find the Custom cache endpoints section.
    3. Add your cache service URL (for example, https://cache.example.com).

  • Self-hosted Tuist server:

    1. Set TUIST_CACHE_ENDPOINTS to a comma-separated list of cache node URLs (for example, https://cache-1.example.com,https://cache-2.example.com).
    2. Restart the Tuist server to apply the configuration.
graph TD
  A[Deploy cache service] --> B[Register cache endpoint]
  B --> C[Tuist CLI uses your endpoint]

Once configured, the Tuist CLI will use your self-hosted cache.

Volumes#

The Docker Compose configuration uses three volumes:

Volume Purpose
storage Binary artifact storage
sqlite_data SQLite metadata storage. By default this holds both /data/repo.sqlite (artifact metadata, orphan cleanup state, Oban) and /data/key_value.sqlite (KV metadata).
cache_socket Unix socket for Nginx-service communication

Background maintenance#

The cache service runs several background maintenance loops that keep disk and database usage within bounds. You can tune their behavior through the environment variables listed in the configuration table above.

  • CAS disk eviction — When disk usage exceeds DISK_HIGH_WATERMARK_PERCENT (default 85%), the service removes the least-recently-used local artifacts until usage drops to DISK_TARGET_PERCENT (default 70%). Evicted artifacts remain in S3.
  • KV eviction — Entries not accessed within 30 days are always removed regardless of database size. Additionally, when the KV database exceeds KEY_VALUE_MAX_DB_SIZE_BYTES (default 25 GiB), the service removes entries older than KEY_VALUE_EVICTION_MIN_RETENTION_DAYS until the database shrinks to KEY_VALUE_EVICTION_HYSTERESIS_RELEASE_BYTES. Each pass is capped at KEY_VALUE_EVICTION_MAX_DURATION_MS.
  • Orphan cleanup — Scans the disk storage tree for files without a matching cache_artifacts row and removes them. This depends on the primary metadata database, not the KV database.

For a detailed explanation of how each process works internally, see the architecture guide.

Health checks#

  • GET /up — Returns 200 when healthy
  • GET /metrics — Prometheus metrics

Monitoring#

Prometheus metrics#

The cache service exposes Prometheus-compatible metrics at /metrics.

Database pool metrics exposed by the cache service include:

  • cache_repo_pool_checkout_queue_length
  • cache_repo_pool_ready_conn_count
  • cache_repo_pool_size
  • cache_repo_pool_db_connection_connected_total
  • cache_repo_pool_db_connection_disconnected_total

The pool last_value metrics are labeled with:

  • repo — cache or key_value
  • database — sqlite

If you use Grafana, you can import the reference dashboard.

Distributed tracing#

Set OTEL_EXPORTER_OTLP_ENDPOINT to enable OpenTelemetry traces. The cache service instruments Bandit (HTTP server), Phoenix (request lifecycle), Ecto (database queries), Finch (outgoing HTTP), and Broadway (message processing). Traces are exported via gRPC to the configured collector.

Log forwarding#

Set LOKI_URL to forward application logs to a Loki-compatible endpoint. Logs are pushed via the Loki HTTP API with app=tuist-cache, env, and level labels.

Upgrading#

bash
docker compose pull
docker compose up -d

The service runs database migrations automatically on startup. After upgrading, expect a brief warm-up period while the KV cache repopulates from new traffic.

Troubleshooting#

Cache not being used#

If you expect caching but are seeing consistent cache misses (for example, the CLI is repeatedly uploading the same artifacts, or downloads never happen), follow these steps:

  1. Verify the custom cache endpoint is correctly configured in your organization settings.
  2. Ensure your Tuist CLI is authenticated by running tuist auth login.
  3. Check the cache service logs for any errors: docker compose logs cache.

Do I need the repo.sqlite file?#

repo.sqlite is the primary metadata database. It stores artifact metadata, orphan cleanup state, and background job data. It is required for normal operation.

If you upgraded from an older version that also stored KV metadata in repo.sqlite, KV data has moved to the dedicated key_value.sqlite file. The legacy KV tables (key_value_entries, key_value_entry_hashes) in repo.sqlite are no longer used and can be removed during a maintenance window to reclaim space.

Socket path mismatch#

If you see connection refused errors:

  • Ensure PHX_SOCKET_LINK points to the socket path configured in nginx.conf (default: /run/cache/current.sock)
  • Verify PHX_SOCKET_PATH and PHX_SOCKET_LINK are both set correctly in docker-compose.yml
  • Verify the cache_socket volume is mounted in both containers