Documentation
¶
Overview ¶
Package blob is the platform's object-storage seam: opaque bytes at string keys, behind the one interface every backend must satisfy (CLAUDE.md: backend variability lives behind an interface with one shared contract suite — internal/blob/blobtest). The first consumer is the skills registry (docs/plan/06_skills.md), which stores canonical skill-version archives at `skills/{skill_id}/{version}.zip`; the key namespace deliberately leaves room for later surfaces (the deferred Files API) to share the store.
Index ¶
Constants ¶
const ( MetricOpDuration = "blob.op.duration" MetricOpBytes = "blob.op.bytes" )
MetricOpDuration and MetricOpBytes are platform-native names (dotted, lowercase, unit in the Unit field) recorded at the Store seam, so every backend and every consumer shares one view of object-storage health. Exported so the telemetry contract test can assert they reach a collector.
Variables ¶
var ErrNotFound = errors.New("blob: object not found")
ErrNotFound reports a Get of a key that has no object. Implementations wrap it so callers can errors.Is across backends.
Functions ¶
func FilesKey ¶
FilesKey is the object-storage key a new Files-API file's bytes are written at — the `files/{file_id}` namespace this package's doc reserves. It lives here, not in a feature package (unlike skills' own BlobKey), because it has no home package: the api registry, the executor's outputs harvest and the dream runner all write files objects.
It names where a writer puts bytes, never where a reader finds them. Since #578 a session's copy of an upload is a files row of its own that names the upload's key in object_key, so every reader reads store.FileObjectKeySQL, which spells this same layout in SQL for every row that owns its object.
func SessionCheckpointKey ¶
SessionCheckpointKey is the object-storage key for a session's workspace checkpoint — the `workspace/{session_id}/checkpoint.tar.gz` layout plan 24 fixes. It lives here for FilesKey's reason: the executor writes it (the idle-TTL capture), the executor's reaper and the API's session delete both remove it, and one definition keeps the writer and its removers from drifting.
Types ¶
type Store ¶
type Store interface {
// Put stores exactly size bytes from r at key, overwriting any existing
// object: a reader with fewer bytes than size is an error, and bytes
// beyond size are never read. contentType is stored as object metadata
// for HTTP consumers.
Put(ctx context.Context, key string, r io.Reader, size int64, contentType string) error
// Get returns the object's bytes and size. A missing key is ErrNotFound
// from Get itself, never deferred to the first Read. The caller closes
// the reader.
Get(ctx context.Context, key string) (io.ReadCloser, int64, error)
// Delete removes the object at key. Deleting a missing key is not an
// error: a crashed-and-retried delete must converge, not flap.
Delete(ctx context.Context, key string) error
}
Store is the object-storage contract. Keys are opaque non-empty strings; "/" separators are conventional namespacing, not directories.
func WithMetrics ¶
WithMetrics wraps a Store so every operation records its duration (by op and outcome) and payload size (by op). Get's duration covers the call that opens the object, not the caller's subsequent reads.
Source Files
Directories
¶
| Path | Synopsis |
|---|---|
|
Package backend selects an object-storage backend by name, so every binary that reaches object storage constructs it from the same config point instead of each mapping the environment its own way.
|
Package backend selects an object-storage backend by name, so every binary that reaches object storage constructs it from the same config point instead of each mapping the environment its own way. |
|
Package blobtest is test support for the blob.Store seam: it starts one Dockerized Silo (a MinIO fork, see Image) per test binary and hands out per-test targets (endpoint, credentials, fresh bucket name) for backends to construct stores against.
|
Package blobtest is test support for the blob.Store seam: it starts one Dockerized Silo (a MinIO fork, see Image) per test binary and hands out per-test targets (endpoint, credentials, fresh bucket name) for backends to construct stores against. |
|
Package gcs is the Google Cloud Storage backend, on the native cloud.google.com/go/storage client rather than GCS's S3-interop XML API.
|
Package gcs is the Google Cloud Storage backend, on the native cloud.google.com/go/storage client rather than GCS's S3-interop XML API. |
|
gcstest
Package gcstest is test support for the GCS backend: one Dockerized fake-gcs-server per test binary, per-test fresh buckets, and the gate for the opt-in tier that calls real Cloud Storage.
|
Package gcstest is test support for the GCS backend: one Dockerized fake-gcs-server per test binary, per-test fresh buckets, and the gate for the opt-in tier that calls real Cloud Storage. |
|
Package s3 is the S3-compatible blob.Store backend, on minio-go so one implementation speaks to MinIO, AWS S3, Ceph RGW, or anything else with the S3 wire protocol — never a MinIO-specific API (an operator must be able to swap the vendor without touching this package).
|
Package s3 is the S3-compatible blob.Store backend, on minio-go so one implementation speaks to MinIO, AWS S3, Ceph RGW, or anything else with the S3 wire protocol — never a MinIO-specific API (an operator must be able to swap the vendor without touching this package). |