upload

package
v0.0.0-...-c82e7e9 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Oct 31, 2026 License: MPL-2.0 Imports: 19 Imported by: 0

Documentation

Overview

Package upload implements the backend-neutral, instance-owned core file store.

Index

Constants

View Source
const (
	MaxChunkBytes                   = 256 << 10
	MaxFileBytes             uint64 = 500 << 20
	ActiveUploadsPerInstance        = 2
	ActiveUploadsPerHost            = 8
	BytesPerInstance         uint64 = 2 << 30
	BytesPerHost             uint64 = 8 << 30
	CompletionTTL                   = 30 * time.Minute
)

Variables

This section is empty.

Functions

This section is empty.

Types

type Config

type Config struct {
	ManagementDirectory string
	Limits              runtime.Limits
	// ModuleMountRoots lists every trusted host subtree exposed to modules.
	// Roots must already exist; lexical and resolved overlap both fail closed.
	ModuleMountRoots []string
}

Config deliberately has no defaults for the five operator-required upload limits. The supervisor provisions ManagementDirectory/uploads as a private, runtime-owned 0700 child outside every module mount. The management parent may remain root-owned and grant only trusted runtime-group traversal.

type UploadReadLease

type UploadReadLease struct {
	// contains filtered or unexported fields
}

UploadReadLease pins one immutable upload for a single admitted outbound operation. Deletion/expiry prevent new leases but do not cancel this lease. Cancellation closes readers; the owning backend joins its multipart/HTTP workers before calling Release. Instance cleanup joins forced releases.

func (*UploadReadLease) Context

func (l *UploadReadLease) Context() context.Context

Context is bounded by both the actual admission and outbound operation. Use it for HTTP and multipart workers so instance stop cancels the complete operation.

func (*UploadReadLease) Open

Open returns a fresh offset-zero reader. Every reader must be closed before Release can complete. A shorter per-attempt context also cancels that reader.

func (*UploadReadLease) Reference

func (l *UploadReadLease) Reference() *wire.UploadedFileRef

Reference returns a fresh validated metadata copy, never a filesystem path.

func (*UploadReadLease) Release

func (l *UploadReadLease) Release() error

Release prevents further Open calls, waits for all owned readers to close, and drops the storage pin exactly once. It does not cancel admitted readers. Cancellation/stop instead close readers through the lease's context callback.

type UploadStore

type UploadStore struct {
	// contains filtered or unexported fields
}

UploadStore owns all reservation and retirement state under a single lock. Byte charges and record entries are released independently, only after actual unlink and, for entries, the original deletion retention window have ended.

func New

func New(ctx context.Context, config Config) (*UploadStore, error)

New verifies and exclusively claims the existing supervisor-provisioned ManagementDirectory/uploads child, then reclaims its old internal spool files. It never recovers references from an earlier runtime.

func (*UploadStore) AcquireReadLease

func (s *UploadStore) AcquireReadLease(ctx context.Context, principal runtime.InstancePrincipal, ref *wire.UploadedFileRef) (*UploadReadLease, error)

AcquireReadLease requires an actual runtime operation admission in ctx and a finite outbound-operation deadline. Metadata is checked against the stored owner, exact size/name/MIME and original expiry; none is used as authority.

func (*UploadStore) CleanupInstance

func (s *UploadStore) CleanupInstance(ctx context.Context, p runtime.InstancePrincipal) error

CleanupInstance is runtime.ManagerOptions.FileCleanup. It denies new work, cancels and joins writers/readers, drops retention, and reclaims actual bytes. A deadline or unlink failure keeps the ownership lock and all remaining charges so the manager can fail closed and retry rather than declaring reclamation.

func (*UploadStore) Close

func (s *UploadStore) Close(ctx context.Context) error

Close cancels all instances and releases exclusive directory ownership only after every worker and actual spool file has been reclaimed. Retry on failure. The composition must stop/join backend operations before closing the store.

func (*UploadStore) Delete

Delete prevents new reads. Already admitted leases remain usable until their own operation deadline. A successful unlink retains an owner-checked tombstone until the original expiration, even for zero-byte files.

func (*UploadStore) LastCleanupError

func (s *UploadStore) LastCleanupError() error

LastCleanupError returns a fixed, structured diagnostic, never a filesystem path or underlying OS error. A failure remains observable after later retries.

func (*UploadStore) RegistrarFactory

func (s *UploadStore) RegistrarFactory(services runtime.InstanceServices) (core.HelperRegistrar, error)

RegistrarFactory is runtime.ManagerOptions.FileTransfer. It captures the real instance Authorizer, never a metadata identity. Construct one shared store per runtime; host quotas and the ownership lock must not be split among instances.

func (*UploadStore) Sweep

func (s *UploadStore) Sweep(ctx context.Context) error

Sweep reattempts failed cleanup and expires records without extending any TTL. Timer cleanup failures are also available through LastCleanupError.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL