service/vendor/go.mongodb.org/mongo-driver/mongo/options/gridfsoptions.go

330 lines
9.7 KiB
Go
Raw Normal View History

2023-12-21 22:17:40 +08:00
// Copyright (C) MongoDB, Inc. 2017-present.
//
// Licensed under the Apache License, Version 2.0 (the "License"); you may
// not use this file except in compliance with the License. You may obtain
// a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
package options
import (
"time"
"go.mongodb.org/mongo-driver/bson"
"go.mongodb.org/mongo-driver/bson/bsoncodec"
"go.mongodb.org/mongo-driver/mongo/readconcern"
"go.mongodb.org/mongo-driver/mongo/readpref"
"go.mongodb.org/mongo-driver/mongo/writeconcern"
)
// DefaultName is the default name for a GridFS bucket.
var DefaultName = "fs"
// DefaultChunkSize is the default size of each file chunk in bytes (255 KiB).
var DefaultChunkSize int32 = 255 * 1024
// DefaultRevision is the default revision number for a download by name operation.
var DefaultRevision int32 = -1
// BucketOptions represents options that can be used to configure GridFS bucket.
type BucketOptions struct {
// The name of the bucket. The default value is "fs".
Name *string
// The number of bytes in each chunk in the bucket. The default value is 255 KiB.
ChunkSizeBytes *int32
// The write concern for the bucket. The default value is the write concern of the database from which the bucket
// is created.
WriteConcern *writeconcern.WriteConcern
// The read concern for the bucket. The default value is the read concern of the database from which the bucket
// is created.
ReadConcern *readconcern.ReadConcern
// The read preference for the bucket. The default value is the read preference of the database from which the
// bucket is created.
ReadPreference *readpref.ReadPref
}
// GridFSBucket creates a new BucketOptions instance.
func GridFSBucket() *BucketOptions {
return &BucketOptions{
Name: &DefaultName,
ChunkSizeBytes: &DefaultChunkSize,
}
}
// SetName sets the value for the Name field.
func (b *BucketOptions) SetName(name string) *BucketOptions {
b.Name = &name
return b
}
// SetChunkSizeBytes sets the value for the ChunkSize field.
func (b *BucketOptions) SetChunkSizeBytes(i int32) *BucketOptions {
b.ChunkSizeBytes = &i
return b
}
// SetWriteConcern sets the value for the WriteConcern field.
func (b *BucketOptions) SetWriteConcern(wc *writeconcern.WriteConcern) *BucketOptions {
b.WriteConcern = wc
return b
}
// SetReadConcern sets the value for the ReadConcern field.
func (b *BucketOptions) SetReadConcern(rc *readconcern.ReadConcern) *BucketOptions {
b.ReadConcern = rc
return b
}
// SetReadPreference sets the value for the ReadPreference field.
func (b *BucketOptions) SetReadPreference(rp *readpref.ReadPref) *BucketOptions {
b.ReadPreference = rp
return b
}
// MergeBucketOptions combines the given BucketOptions instances into a single BucketOptions in a last-one-wins fashion.
func MergeBucketOptions(opts ...*BucketOptions) *BucketOptions {
b := GridFSBucket()
for _, opt := range opts {
if opt == nil {
continue
}
if opt.Name != nil {
b.Name = opt.Name
}
if opt.ChunkSizeBytes != nil {
b.ChunkSizeBytes = opt.ChunkSizeBytes
}
if opt.WriteConcern != nil {
b.WriteConcern = opt.WriteConcern
}
if opt.ReadConcern != nil {
b.ReadConcern = opt.ReadConcern
}
if opt.ReadPreference != nil {
b.ReadPreference = opt.ReadPreference
}
}
return b
}
// UploadOptions represents options that can be used to configure a GridFS upload operation.
type UploadOptions struct {
// The number of bytes in each chunk in the bucket. The default value is DefaultChunkSize (255 KiB).
ChunkSizeBytes *int32
// Additional application data that will be stored in the "metadata" field of the document in the files collection.
// The default value is nil, which means that the document in the files collection will not contain a "metadata"
// field.
Metadata interface{}
// The BSON registry to use for converting filters to BSON documents. The default value is bson.DefaultRegistry.
Registry *bsoncodec.Registry
}
// GridFSUpload creates a new UploadOptions instance.
func GridFSUpload() *UploadOptions {
return &UploadOptions{Registry: bson.DefaultRegistry}
}
// SetChunkSizeBytes sets the value for the ChunkSize field.
func (u *UploadOptions) SetChunkSizeBytes(i int32) *UploadOptions {
u.ChunkSizeBytes = &i
return u
}
// SetMetadata sets the value for the Metadata field.
func (u *UploadOptions) SetMetadata(doc interface{}) *UploadOptions {
u.Metadata = doc
return u
}
// MergeUploadOptions combines the given UploadOptions instances into a single UploadOptions in a last-one-wins fashion.
func MergeUploadOptions(opts ...*UploadOptions) *UploadOptions {
u := GridFSUpload()
for _, opt := range opts {
if opt == nil {
continue
}
if opt.ChunkSizeBytes != nil {
u.ChunkSizeBytes = opt.ChunkSizeBytes
}
if opt.Metadata != nil {
u.Metadata = opt.Metadata
}
if opt.Registry != nil {
u.Registry = opt.Registry
}
}
return u
}
// NameOptions represents options that can be used to configure a GridFS DownloadByName operation.
type NameOptions struct {
// Specifies the revision of the file to retrieve. Revision numbers are defined as follows:
//
// * 0 = the original stored file
// * 1 = the first revision
// * 2 = the second revision
// * etc..
// * -2 = the second most recent revision
// * -1 = the most recent revision.
//
// The default value is -1
Revision *int32
}
// GridFSName creates a new NameOptions instance.
func GridFSName() *NameOptions {
return &NameOptions{}
}
// SetRevision sets the value for the Revision field.
func (n *NameOptions) SetRevision(r int32) *NameOptions {
n.Revision = &r
return n
}
// MergeNameOptions combines the given NameOptions instances into a single *NameOptions in a last-one-wins fashion.
func MergeNameOptions(opts ...*NameOptions) *NameOptions {
n := GridFSName()
n.Revision = &DefaultRevision
for _, opt := range opts {
if opt == nil {
continue
}
if opt.Revision != nil {
n.Revision = opt.Revision
}
}
return n
}
// GridFSFindOptions represents options that can be used to configure a GridFS Find operation.
type GridFSFindOptions struct {
// If true, the server can write temporary data to disk while executing the find operation. The default value
// is false. This option is only valid for MongoDB versions >= 4.4. For previous server versions, the server will
// return an error if this option is used.
AllowDiskUse *bool
// The maximum number of documents to be included in each batch returned by the server.
BatchSize *int32
// The maximum number of documents to return. The default value is 0, which means that all documents matching the
// filter will be returned. A negative limit specifies that the resulting documents should be returned in a single
// batch. The default value is 0.
Limit *int32
// The maximum amount of time that the query can run on the server. The default value is nil, meaning that there
// is no time limit for query execution.
//
// NOTE(benjirewis): MaxTime will be deprecated in a future release. The more general Timeout option may be used
// in its place to control the amount of time that a single operation can run before returning an error. MaxTime
// is ignored if Timeout is set on the client.
MaxTime *time.Duration
// If true, the cursor created by the operation will not timeout after a period of inactivity. The default value
// is false.
NoCursorTimeout *bool
// The number of documents to skip before adding documents to the result. The default value is 0.
Skip *int32
// A document specifying the order in which documents should be returned. The driver will return an error if the
// sort parameter is a multi-key map.
Sort interface{}
}
// GridFSFind creates a new GridFSFindOptions instance.
func GridFSFind() *GridFSFindOptions {
return &GridFSFindOptions{}
}
// SetAllowDiskUse sets the value for the AllowDiskUse field.
func (f *GridFSFindOptions) SetAllowDiskUse(b bool) *GridFSFindOptions {
f.AllowDiskUse = &b
return f
}
// SetBatchSize sets the value for the BatchSize field.
func (f *GridFSFindOptions) SetBatchSize(i int32) *GridFSFindOptions {
f.BatchSize = &i
return f
}
// SetLimit sets the value for the Limit field.
func (f *GridFSFindOptions) SetLimit(i int32) *GridFSFindOptions {
f.Limit = &i
return f
}
// SetMaxTime sets the value for the MaxTime field.
//
// NOTE(benjirewis): MaxTime will be deprecated in a future release. The more general Timeout
// option may be used in its place to control the amount of time that a single operation can
// run before returning an error. MaxTime is ignored if Timeout is set on the client.
func (f *GridFSFindOptions) SetMaxTime(d time.Duration) *GridFSFindOptions {
f.MaxTime = &d
return f
}
// SetNoCursorTimeout sets the value for the NoCursorTimeout field.
func (f *GridFSFindOptions) SetNoCursorTimeout(b bool) *GridFSFindOptions {
f.NoCursorTimeout = &b
return f
}
// SetSkip sets the value for the Skip field.
func (f *GridFSFindOptions) SetSkip(i int32) *GridFSFindOptions {
f.Skip = &i
return f
}
// SetSort sets the value for the Sort field.
func (f *GridFSFindOptions) SetSort(sort interface{}) *GridFSFindOptions {
f.Sort = sort
return f
}
// MergeGridFSFindOptions combines the given GridFSFindOptions instances into a single GridFSFindOptions in a
// last-one-wins fashion.
func MergeGridFSFindOptions(opts ...*GridFSFindOptions) *GridFSFindOptions {
fo := GridFSFind()
for _, opt := range opts {
if opt == nil {
continue
}
if opt.AllowDiskUse != nil {
fo.AllowDiskUse = opt.AllowDiskUse
}
if opt.BatchSize != nil {
fo.BatchSize = opt.BatchSize
}
if opt.Limit != nil {
fo.Limit = opt.Limit
}
if opt.MaxTime != nil {
fo.MaxTime = opt.MaxTime
}
if opt.NoCursorTimeout != nil {
fo.NoCursorTimeout = opt.NoCursorTimeout
}
if opt.Skip != nil {
fo.Skip = opt.Skip
}
if opt.Sort != nil {
fo.Sort = opt.Sort
}
}
return fo
}