0
688
VistasHow to improve GET query performance in FastAPI?
I'm currently working on developing API with FastAPI.
The API is for getting certain amount of data on a specific date from MariaDB.
I implemented this using FastAPI Pagination, but the it took 9 seconds to get a response.
Would there be other ways to improve performance?
I'm attaching a little background below:
- The current GET API URL
http://localhost:8080/images/{date}?{page=N}&{size=M} (e.g. http://localhost:8080/images/20211223?page=1&size=50)
- Table information
The table has about 10,000,000 rows, and here are a few lines of aggregated data. There is no index in this table, as of now. If needed, I'd add it!
- The current implementation
# get function part
@router.get('/{stnd_ymd}', response_model=Page[ResponseImage])
async def get_images(stnd_ymd: str):
image_list = Images.get_all(stnd_ymd=stnd_ymd, cnts_ty_cd="CT_IMAGE")
return paginate(image_list)
# about `get_all` function
def get_all(cls, session: Session = None, **kwargs):
sess = next(db.session()) if not session else session
query = sess.query(cls)
for key, val in kwargs.items():
col = getattr(cls, key)
query = query.filter(col == val)
result = query.all()
if not session:
sess.close()
return result
If there is any other information needed, please tell me!
Thanks,
2 Respuestas
Responde la pregunta0
As some people pointed out in the comments, the slow response time could be because of badly optimized query. For example in your get_all method when applying filters for query you always fetch all columns which might not be the best practice when fetching such big amount of data so try fetching only those columns that are needed for your frontend/client in specific view. Also I think that you need to write your own custom pagination done at the query level (which will be faster because it's done by database itself) because I have suspicion that fastapi-pagination only paginates already pre-fetched data. Example of pagination using SQLAlchemy:
offset = (page_number * items_count) - items_count
filters = [
getattr(cls, column_name) == value
for column_name, value in kwargs.items()
]
query = query.where(*filters)
result = query.offset(offset).limit(items_count).all()
Try using fastapi-profiler when looking for performance bottlenecks in your code. Here is simple configuration I used in one of my projects:
app.add_middleware(
CProfileMiddleware,
enable=True,
print_each_request=True,
strip_dirs=False,
sort_by="cumtime"
)
0
Which ORM framework do you use? Tortoise? SQLAlchemy? Fastapi-pagination integrates with bunch of frameworks
Then you want to understand what SQL query is produced. For this, I would debug the request and dive deep into ORM framework code to find a place when SQL query is generated. Usually, simple step into debugging should work.
Some ORM frameworks allow you to just see SQL queries logged. Here is a question about SQLAlchemy query log.
After you found your query you want to test it out manually and see what you can improve. ANALYZE command can be handy to get information about used table indexes and stuff.