How to Optimize JSON Response Times for Large Datasets in FastAPI?

Optimizing JSON Response Times for Large Datasets in FastAPI

Issue:

Retrieving a significant amount of JSON data from a FastAPI endpoint is noticeably slow, requiring approximately a minute. The data is initially loaded from a parquet file using json.loads() and filtered before being returned. Seeking a swifter approach to deliver the data.

Resolution:

The slow response time stems from multiple JSON conversions within the parse_parquet() function. FastAPI automatically encodes the returned value using jsonable_encoder before serializing it with json.dumps(), a time-consuming process. External JSON encoders like orjson or ujson offer potential speed enhancements.

However, the most efficient solution is to avoid unnecessary JSON conversions. The following code utilizes a custom APIRoute class to enable direct JSON responses from pandas DataFrames:

<code class="python">from fastapi import APIRoute

class TimedRoute(APIRoute):
    # Custom handler for capturing response time
    def get_route_handler(self):
        original_route_handler = super().get_route_handler()
        
        async def custom_route_handler(request):
            before = time.time()
            response = await original_route_handler(request)
            duration = time.time() - before
            response.headers["Response-Time"] = str(duration)
            print(f"route duration: {duration}")
            return response

        return custom_route_handler</code>

This code allows you to compare the response times of different data conversion methods. Using a sample parquet file with 160,000 rows and 45 columns, the following results were obtained:

• Default FastAPI Encoder (json.dumps()): Slowest
• orjson: Comparable to default encoder
• ujson: Slightly faster than orjson
• PandasJSON (df.to_json()): Most significantly faster

To improve user experience, consider setting the Content-Disposition header with the attachment parameter and a filename to initiate a download rather than displaying the data within the browser. This approach bypasses browser constraints and speeds up the process.

Additionally, Dask provides optimized handling of large datasets, offering alternatives to pandas. Streaming or asynchronous responses may also be considered to avoid memory issues when dealing with massive data volumes.

The above is the detailed content of How to Optimize JSON Response Times for Large Datasets in FastAPI?. For more information, please follow other related articles on the PHP Chinese website!