Bug with Open XML SDK Disposing SpreadsheetDocument

[JoseAngelSalmeronMorote]

Describe the bug

Reading metadata properties of an Excel file leads to a huge increase of memory while disposing SpreadsheetDocument.

Screenshots

Memory issue on Spreadsheetdocument disposal:

To Reproduce

Minimal repro:

var openSettings = new OpenSettings
{
    RelationshipErrorHandlerFactory = package =>
    {
        return new UriRelationshipErrorHandler();
    }
};
using (FileStream fs = new FileStream(filePath, FileMode.Open, FileAccess.ReadWrite, FileShare.ReadWrite))
{
    using (SpreadsheetDocument myDoc = SpreadsheetDocument.Open(fs, true, openSettings))
    {
        //Access main part of document 
        CustomFilePropertiesPart FileProperties = myDoc.CustomFilePropertiesPart;

        if (FileProperties == null)
        {
            output = MetadataError.ThereAreNoProperties;
            return output;
        }

        // Bug here: Executing this line leads to later issue while disposing 'myDoc'.
        DocumentFormat.OpenXml.CustomProperties.Properties props = FileProperties.Properties;
        output = MetadataError.Ok;
    } // Issue here: this line never returns.
    fs.Close(); // Line never reached.
}

Steps to reproduce the behavior:

1. Get an Excel file of 100Mb.
2. Add any metadata attribute: 'Example' of type 'Text' with value 'MyExampleValue'.
3. Execute above code.
4. See how memory increases until several Gb and the process hang.

Observed behavior

Memory increases until its limit. Dispose line never returns.

Expected behavior
Memory should not increase and .Disposing the SpreadsheetDocument should not hang the application.

Desktop (please complete the following information):

• OS: [Windows Server 2022 Datacenter Evaluation]
• Office version [N/A]
• .NET Target: reproduced with .NETStandard 2.0 and .NETFramework 4.0
• DocumentFormat.OpenXml Version: Reproduced with 2.12.0 and 2.20.0.

Additional context

Bug also reproducible with 3.3.0 version and next snipet:

using (FileStream fs = new FileStream(filePath, FileMode.Open, FileAccess.ReadWrite, FileShare.ReadWrite))
{
    using (SpreadsheetDocument myDoc = SpreadsheetDocument.Open(fs, true))
    {
        //Access main part of document 
        CustomFilePropertiesPart FileProperties = myDoc.CustomFilePropertiesPart;

        if (FileProperties == null)
        {
            output = MetadataError.ThereAreNoProperties;
            return output;
        }

        // Bug here: Executing this line leads to later issue while disposing 'myDoc'.
        DocumentFormat.OpenXml.CustomProperties.Properties props = FileProperties.Properties;
    } // Issue here: this line never returns.
    fs.Close(); // Line never reached.
}

The root issue seems to be when opening the document with isEditable = true.

[twsouthwick]

This doesn't really look reproducible. Can you provide a standalone sample? This could create a sample doc, then do whatever you need to reproduce it (does it need to be in a loop?)

[JoseAngelSalmeronMorote]

Hi, did you test it with a large Excel file? With small ones the issue is not reproducible.

I've created this table when looking for a treshold:

Type	Size	Memory (Mb)	Status
Word	20Mb	negligible	✅
Excel	50Mb	4.000	❌
Excel	25Mb	1.900	❌
Excel	8Mb	negligible	✅
Excel	12Mb	negligible	✅
Excel	19Mb	1.800	❌
Excel	16Mb	1.500	❌

Also i've observed some diference between versions: 3.3.0 doesnt need to execute any code after opening the document with isEditable = true, while version 2.12.0 needs to execute this line: DocumentFormat.OpenXml.CustomProperties.Properties props = FileProperties.Properties;

This means issue is only reproducible with large files containing at least 1 custom property for 2.12.0 while for 3.3.0 it's reproducible for all large files.

You may download a full executing sample using minimal repro with a test file that reproduce the issue for both, 2.12.0 (ConsoleApp1) and 3.3.0 (ConsoleApp2) from here.

Please let me know if you need anything else to reproduce the issue.