package driver import ( "errors" "go.mongodb.org/mongo-driver/x/bsonx/bsoncore" ) // this is the amount of reserved buffer space in a message that the // driver reserves for command overhead. const reservedCommandBufferBytes = 16 * 10 * 10 * 10 // ErrDocumentTooLarge occurs when a document that is larger than the maximum size accepted by a // server is passed to an insert command. var ErrDocumentTooLarge = errors.New("an inserted document is too large") // Batches contains the necessary information to batch split an operation. This is only used for write // oeprations. type Batches struct { Identifier string Documents []bsoncore.Document Current []bsoncore.Document Ordered *bool } // Valid returns true if Batches contains both an identifier and the length of Documents is greater // than zero. func (b *Batches) Valid() bool { return b != nil && b.Identifier != "" && len(b.Documents) > 0 } // ClearBatch clears the Current batch. This must be called before AdvanceBatch will advance to the // next batch. func (b *Batches) ClearBatch() { b.Current = b.Current[:0] } // AdvanceBatch splits the next batch using maxCount and targetBatchSize. This method will do nothing if // the current batch has not been cleared. We do this so that when this is called during execute we // can call it without first needing to check if we already have a batch, which makes the code // simpler and makes retrying easier. // The maxDocSize parameter is used to check that any one document is not too large. If the first document is bigger // than targetBatchSize but smaller than maxDocSize, a batch of size 1 containing that document will be created. func (b *Batches) AdvanceBatch(maxCount, targetBatchSize, maxDocSize int) error { if len(b.Current) > 0 { return nil } if maxCount <= 0 { maxCount = 1 } splitAfter := 0 size := 0 for i, doc := range b.Documents { if i == maxCount { break } if len(doc) > maxDocSize { return ErrDocumentTooLarge } if size+len(doc) > targetBatchSize { break } size += len(doc) splitAfter++ } // if there are no documents, take the first one. // this can happen if there is a document that is smaller than maxDocSize but greater than targetBatchSize. if splitAfter == 0 { splitAfter = 1 } b.Current, b.Documents = b.Documents[:splitAfter], b.Documents[splitAfter:] return nil }