microsoft/aspire · error · InvalidOperationException

The Microsoft Foundry model catalog returned the same…

Error message

The Microsoft Foundry model catalog returned the same continuation token twice. Refusing to overwrite the generated model descriptors with a partial catalog. Response: {GetResponseSnippet(pageResponse)}

What it means

While paging the Foundry catalog, GetAllModelsAsync tracks the previous continuation token and throws if the service returns the identical token twice, which would loop forever and produce a partial catalog. The guard exists so generated descriptors are never overwritten from an incomplete page walk.

Solutions

Retry the tool run after some time — repeated identical tokens from a degraded service are often transient.
Inspect the response snippet to see the token values and confirm the server is not advancing pagination.
Check whether a proxy/cache is serving stale pages; bypass caching or add no-cache headers in ModelClient.
Verify the requested API version supports continuation tokens as the tool sends them; adjust paging parameters if the contract changed.

Defensive patterns

Strategy: retry

Try / catch

try
{
    await RunGenModelAsync();
}
catch (InvalidOperationException ex) when (ex.Message.Contains("same continuation token twice"))
{
    // service-side paging stall: back off and retry the whole run
}

Prevention

Bypass caches/proxies for catalog requests to avoid stale pages
Retry the full pagination run with backoff on repeated tokens
Monitor Foundry API version changes affecting continuation tokens
Cap total pages walked as an additional loop guard

When it happens

Trigger: The Foundry catalog's next-page continuation token does not advance — consecutive pages return the same token — inside the do/while pagination loop of GetAllModelsAsync.

Common situations: Foundry service bug or degraded state where the next-page link is stale; cached/proxied responses replaying the same page; API version mismatch causing the server to ignore paging parameters; index larger than the server can page correctly.

Understand the failure class

Background: "invalid response format", "malformed payload", "missing data field": when an API returns 200 but the response shape is wrong — this error's family across 23 libraries.

Related errors

AI-assisted analysis of microsoft/aspire@25830f84bd (2026-09-16). Data as JSON: /api/errors/23446f4f83a7b7a5. Report an issue: GitHub.

Appendix: source

Thrown at src/Aspire.Hosting.Foundry/tools/GenModel.cs:783

            foreach (var model in pageModels)
            {
                allModels.Add(model);
            }

            Console.WriteLine($"Fetched page with {pageModels.Count} models. Total so far: {allModels.Count}");

            // Check if there's a continuation token for the next page
            previousToken = continuationToken;
            continuationToken = apiResponse?.IndexEntitiesResponse?.ContinuationToken;

            if (!string.IsNullOrEmpty(continuationToken))
            {
                Console.WriteLine($"Found continuation token for next page: {continuationToken.Substring(0, Math.Min(50, continuationToken.Length))}...");

                // Check for infinite loop (same token repeated)
                if (continuationToken == previousToken)
                {
                    throw new InvalidOperationException($"The Microsoft Foundry model catalog returned the same continuation token twice. Refusing to overwrite the generated model descriptors with a partial catalog. Response: {GetResponseSnippet(pageResponse)}");
                }
            }

        } while (!string.IsNullOrEmpty(continuationToken));

        Console.WriteLine($"\n=== Pagination Complete ===");
        Console.WriteLine($"Total pages fetched: {pageCount}");
        Console.WriteLine($"Total models collected: {allModels.Count}");

        RunFixups(allModels);

        // Return the consolidated response using our model
        return new ConsolidatedResponse
        {
            TotalCount = allModels.Count,
            PagesCombined = "All pages fetched",
            Entities = allModels
        };

View on GitHub (pinned to 25830f84bd)