Skip to main content
The Content Addressable Storage (CAS) service implements the Remote Execution API v2 CAS interface for storing and retrieving blobs based on their content digest.

​
Overview

The CAS service provides content-addressable blob storage where each blob is identified by its digest (hash). This enables efficient deduplication and content verification. Key features:
  • Content-addressed blob storage
  • Batch read and write operations
  • Missing blob detection
  • Directory tree traversal

​
Configuration

​
instance_name
string
required
The instance name identifying this CAS endpoint. Must match client configuration.
​
cas_store
string
required
Reference to the store configured in the stores section.
{
  services: {
    cas: [
      {
        instance_name: "main",
        cas_store: "MAIN_CAS_STORE"
      }
    ]
  }
}

​
gRPC Methods

​
FindMissingBlobs

Determines which blobs are missing from the CAS. Request:
​
instance_name
string
required
The instance name
​
blob_digests
Digest[]
required
List of digests to check
Response:
​
missing_blob_digests
Digest[]
List of digests that are not present in the CAS
Example with grpcurl: 
grpcurl -plaintext \
  -d '{
    "instance_name": "main",
    "blob_digests": [
      {"hash": "abc123...", "size_bytes": 1024}
    ]
  }' \
  localhost:50051 build.bazel.remote.execution.v2.ContentAddressableStorage/FindMissingBlobs

​
BatchUpdateBlobs

Upload multiple blobs in a single request. Request:
​
instance_name
string
required
The instance name
​
requests
Request[]
required
Array of upload requests, each containing:
  • digest: Content digest
  • data: Blob data (bytes)
  • compressor: Optional compression (identity, zstd, deflate)
Response:
​
responses
Response[]
Array of responses for each blob:
  • digest: The blob digest
  • status: gRPC status (OK on success)
Batch operations are limited by max_decoding_message_size (default 4MB)

​
BatchReadBlobs

Download multiple blobs in a single request. Request:
​
instance_name
string
required
The instance name
​
digests
Digest[]
required
List of blob digests to read
​
acceptable_compressors
Compressor[]
List of acceptable compression algorithms (identity, zstd, deflate)
Response:
​
responses
Response[]
Array of responses for each blob:
  • digest: The blob digest
  • data: Blob data (bytes)
  • compressor: Compression used
  • status: gRPC status

​
GetTree

Recursively fetch a directory tree. Request:
​
instance_name
string
required
The instance name
​
root_digest
Digest
required
Digest of the root Directory message
​
page_size
int32
Maximum number of directories per response (default: all)
​
page_token
string
Token for pagination
Response (stream):
​
directories
Directory[]
List of Directory messages
​
next_page_token
string
Token for fetching next page

​
Error Codes

CodeDescription
NOT_FOUNDBlob or instance not found
INVALID_ARGUMENTInvalid digest format or instance name
RESOURCE_EXHAUSTEDRequest too large
INTERNALStorage backend error

​
Implementation Details

From nativelink-service/src/cas_server.rs: 
pub struct CasServer {
    stores: HashMap<String, Store>,
}
The CAS server maintains a map of instance names to store implementations, allowing multiple isolated CAS namespaces in a single deployment.
The CAS service follows the Remote Execution API v2 specification

Build docs developers (and LLMs) love

Get started for freeTalk to us