FastAPI-Wrapper CLI & APIness Streamlit App
Arvindra Sehmi, Oxford Economics Ltd. | Website | LinkedIn (Updated: 21 April, 2021)
fastapi-wrapper
is motivated by the work of jrieke on fastapi-csv
.
Create APIs from data files within seconds, using FastAPI
This is a Python package to create APIs from data files (Excel XLSX and CSV only), using a lightweight & fully customizable wrapper around FastAPI. Endpoints and query parameters are auto-generated based on the column names and data types in the data file. Its contents is written to either a temporary in-memory or persistent SQLite
database, so the API can be blazing fast, even for huge files, especially for the in-memory case.
TOC
- Why did I implement this?
- Mods
- Streamlit App Demo
- CLI Demo
- How to use fastapi-wrapper from the command line (CSV Example)
- How to use fastapi-wrapper from Python (XLSX Example)
- Example using the API in Power BI
- My thoughts on additional requirements
- The APINESS Factor
- Resources
- Epilogue
Why did I implement this?
- To enhance and extend some internal Oxford Economics (OE) tools, e.g., Modelit project [OE clients only, sorry!] so that MDL-generated data exports can enjoy some apiness
:-)
- An easy way to expose Excel-Files-as-Databases with a REST API so they can be queried in
Streamlit
, and other apps such asPower BI
andJupyter Notebooks
- It can be useful for mocking data sources in CI/CD testing pipelines
- To experiment with FastAPI, which has garnered a lot of attention and enterprise adoption (e.g. Microsoft)
Value of SQL Model DBs to OE?
- Command line tooling lends itself to pipeline batch automation working with model bases
- Use a standard, performant and scaleable data format and query language
- Portable across multiple device form factors and operating systems
- Easily shareable locally, in the cloud and edge data networks
- Much better for data analytics and vizualization, e.g. direct connectors in Power BI & Tableau
- Ability to store multiple model bases in a single SQL database, e.g. multiple scenarios or model vintages
- Easy to export data to multiple target formats from a SQL DB using commonly available tools and programming languages
- Can be used to augment model data exports from OE's Online Global Economic Model (SkyMod) API
Mods
The changes I made to jrieke's original implementation are:
- Built a
Streamlit
application to:- Interactively upload one or more Excel data files
- Configure each file's API endpoint
- Generate a SQLite database, for all or each file
- Enable downloading of generated SQLite databases, and
- Launch FastAPI to serve the APIs
- Simplified naming of auto-generated query params added to the API
- Added
cols
,where
andcmd
query params for richer SQL queries of the endpoint (including defense against destructive SQL injections) - Error handling
- Rendering response data as JSON and HTML
- Restructured return json results to add some metadata useful for debugging, for example:
{ "metadata": {
"database": "macro",
"table": "custommacromodel_l_a",
"sql_query": "SELECT * FROM custommacromodel_l_a WHERE (Location IN (\"United Kingdom\")) LIMIT 2",
"full_count": 1278,
"results_count": 1278
},
"data": [
{
"id": 457027,
"Location": "United Kingdom",
"Indicator": "GDP, agriculture, real, LCU",
.
.
.
"Year": 1980,
"Value": 8210.16
},
.
.
.
]
}
Streamlit App Demo
In this demo:
- The user successively uploads one or more XLSX/CSV files
- The app displays a row of edit fields for each uploaded file allowing the user to configure the database name, table name, and update mode
- The configuration and files are submitted for processing, that is, databases are generated and populated with their file data and (Fast)API endpoints are created
- After the user has uploaded and processed all their files, the FastAPI server is started making the API live
- The user interacts with the live API via a browser
IMPORTANT: The API is launched via
uvicorn
on its own thread and can't be killed. The only way currently to stop uvicorn and Streamlit is to kill the Python host process.
CLI Demo
(This demo gives you a good idea of the command line system. Note, the latest version differs slightly in terms of the API base URL format used.)
How to use fastapi-wrapper from the command line (CSV Example)
Installing the CLI
# Use pip in the root folder
pip install .
# Or, running setup from the root folder
python setup.py install
# To uninstall
pip uninstall fastapi-wrapper
Running the CLI
Use the CSV data file (472718 rows) in this repo for testing (CustomMacroModel_L_A.csv). To start an API, run one of the following commands:
# From file
fastapi-wrapper CustomMacroModel_L_A.csv
# Directly from URL
fastapi-wrapper https://raw.githubusercontent.com/asehmi/fastapi-wrapper/main/CustomMacroModel_L_A.csv
# If you provide an xlsx file, then add XLSX as the second argument on the command line (see Python example below)
Either command should start a fastapi
instance in uvicorn
(a lightning-fast ASGI web server) on the default host
and port
.
(base) C:\Dev\apiness>fastapi-wrapper custommacromodel_l_a.csv --host localhost --port 8000 --database macro
fastapi_wrapper v0.4.2
Creating > Database: macro | From file: custommacromodel_l_a.csv | Type: CSV | Update mode: replace
Initializing FastAPI_Wrapper...
Starting API server (uvicorn)...
Check out the API docs at http://localhost:8000/docs | http://localhost:8000/redoc
--------------------------------------------------------------------------------
INFO: Started server process [18828]
INFO:uvicorn.error:Started server process [18828]
INFO: Waiting for application startup.
INFO:uvicorn.error:Waiting for application startup.
INFO: Application startup complete.
INFO:uvicorn.error:Application startup complete.
INFO: Uvicorn running on http://localhost:8000 (Press CTRL+C to quit)
INFO:uvicorn.error:Uvicorn running on http://localhost:8000 (Press CTRL+C to quit)
Command line switches
fastapi-wrapper --help
command line switch prints some useful info:
(base) C:\Dev\apiness>fastapi-wrapper --help
fastapi_wrapper v0.4.2
Usage: fastapi-wrapper [OPTIONS] DATA_PATH [DATA_FORMAT]:[CSV|XLSX]
Create APIs from CSV or XLSX data files within seconds, using fastapi.
Just pass along a data file and this command will start a fastapi instance
with auto-generated endpoints & query parameters to access the data.
Arguments:
DATA_PATH Path to the data file [required]
[DATA_FORMAT]:[CSV|XLSX] Format of data file [default: CSV]
Options:
--database TEXT Sqlite DB name. Defaults to in-memory DB.
[default: :memory:]
--if-exists [replace|append|fail]
Defines treatment of database if it exists
[default: replace]
--start-server / --no-start-server
Start server. [default: True]
--host TEXT IP to run the API on [default: 127.0.0.1]
--port INTEGER Port to run the API on [default: 8000]
--help Show this message and exit.
For example:
fastapi-wrapper custommacromodel_l_a.csv --host localhost --port 8000 --database macro
fastapi-wrapper
auto-generates endpoints and query parameters based on the CSV file- Here, the API will have an endpoint
/macro/custommacromodel_l_a
- The URL fragments are
/<database name | :memory:>/<database table name>
- The table name is the same as the file name, but lowercased
- The endpoint can be queried using the imported file's column names
- And finally, a
SQLite
database namedmacro.db
, will also be created
IMPORTANT: The endpoint fragments and column names used as query parameters are lowercased. Spaces in their names will be replaced with underscore ('_').
API documentation
The auto-generated API documentation is available here:
- Interactive: http://localhost:8000/docs
- Reference: http://localhost:8000/redoc
Querying the API
In a browser window enter the following parameterised URLs:
(All URL fragments below are prefixed with http://localhost:8000
)
/macro/custommacromodel_l_a?location=United Kingdom
/macro/custommacromodel_l_a?location=United Kingdom&indicatorcode=GVA
Additionally, fastapi-wrapper
creates some convenience query parameters for specific data types, e.g.
/macro/custommacromodel_l_a?year_gt=2020
(for int/float)/macro/custommacromodel_l_a?location_in=Kingdom
(for string, watch out: this one is case sensitive!)/macro/custommacromodel_l_a?year_lte=2020&Location_in=United Kingdom&indicator_in=GDPAGR
Numerical values can be quoted or not. Strings should not be quoted in query values.
You can explicitly specify SQL "where" read-only clauses using the where
parameter. Destructive SQL commands and clauses cause an exception.
/macro/custommacromodel_l_a?where=Year>="2029" AND Year<="2031" AND Indicator LIKE "%GDP%" AND LocationCode IN ("JAPAN","HK")
SQL command modifiers can be added to using the cmd
parameter.
/macro/custommacromodel_l_a?where=Year>=2030 AND Year<=2031 AND Location="United Kingdom" AND Indicator LIKE "%GDP%"&cmd=LIMIT 5
Note, strings in where
and cmd
values are quoted and must obey SQL syntax rules.
The columns returned by a query can be specified (including aliases) with the cols
parameter.
/macro/custommacromodel_l_a?where=Year>="2029" AND Year<="2031" AND LocationCode IN ("JAPAN","HK")&cols=Location, Indicator, Year, Value&cmd=LIMIT 10
/macro/custommacromodel_l_a?where=LocationCode IN ("JAPAN","HK")&cols=LocationCode as LOCCODE, Indicator as VAR, Year as YR, Value as VAL&cmd=LIMIT 10
IMPORTANT: Where column names are referenced as query parameters (not parameter values), they must be lowercased. When they appear as values for
where
andcols
parameters, case does not matter.
Downloading generated SQLite database
/download/macro.db
or/download/macro
Rendering results as HTML
By default results are rendered as JSON where tabluar data is available on the data
key. To render the data as an HTML table, simply add a tohtml
parameter to the query.
/macro/custommacromodel_l_a?where=Year>="2029" AND Year<="2031" AND Indicator LIKE "%GDP%" AND LocationCode IN ("JAPAN","HK")&tohtml
How to use fastapi-wrapper from Python (XLSX Example)
Ensure the required packages are installed:
pip install -r requirements.txt
Create a file app.py
:
from fastapi_wrapper import FastAPI_Wrapper
app = FastAPI_Wrapper().create_database(database='gcfs', data_path='GCFS Countries.xlsx', data_format='XLSX', if_exists='replace')
Start from terminal just like a normal fastapi app:
uvicorn app:app
OR, create a file main.py
:
from fastapi_wrapper import FastAPI_Wrapper
import uvicorn
app = FastAPI_Wrapper().create_database(database='gcfs', data_path='GCFS Countries.xlsx', data_format='XLSX', if_exists='replace')
uvicorn.run(app, host='localhost', port='8000')
Start from terminal just like a normal python app:
python main.py
The queries are similar to the CSV case above, except the endpoint is different:
/gcfs/gcfs_countries?cmd=LIMIT 1000
/gcfs/gcfs_countries?where=Location LIKE "%Kingdom%" OR Location LIKE "%States%"
This shows use of cols
query parameter:
/gcfs/gcfs_countries?cols=location, indicator, year, value&location_in=United Kingdom
This shows use of _ina
and _inz
query parameters to match at the beginning and end of values respectively, and column name aliases:
/gcfs/gcfs_countries?location_in=United Kingdom&indicator_code_ina=GDP&indicator_code_inz=USC&cols=Location as LOC,Indicator as VAR,Indicator_Code as VARCODE,Value as VAL,Year as YR
Extending the API
The cool thing: FastAPI_Wrapper
only extends FastAPI
. Therefore, you can do all the stuff you can do with a normal fastapi instance, e.g. add a new endpoint:
# Add a new endpoint, just like in normal fastapi
@app.get("/hello")
def hello(self):
return {"Hello:", "World"}
This way you can easily modify endpoints previously generated from the CSV file.
Updating data
If your CSV file changes, you can update the API data with:
app.update_database()
Note, this will only update the data, not the API endpoints or query parameters. To do that, you need to create a new FastAPI_Wrapper
instance or re-start uvicorn
.
Example using the API in Power BI
See the file TestReport.pbix
(requires the free PBI Desktop Application on Windows)
M-Language data transformation script
Power BI report
My thoughts on additional requirements
FastAPI_Wrapper
is a neat idea, but "What is the real value prop?"
-
Needs a decent SQL DB, Redis, or a cloud DB. SQLite is not slow, apparently, SQLite is faster than nearly every other database at read-heavy use cases, and it's super versatile, but I think we need a streaming query API to make it easier to consume in BI tools.
-
This program relies on the built-in SQLite Python binding, and may come with limits we don't want. Need to investigate:
-
Would it be better supporting GraphQL?
-
Extend to handle TXT files, different table layouts, etc.
-
Have the ability to import from Excel named ranges / data tables.
-
[DONE] Have the ability to download databases.
-
Break tight coupling between the api endpoint and the Excel filename
- [DONE] User should interact with an "upload & generate api" workflow
- Provide a namespace to isolate APIs from one another (could be the organisation name, user name, etc.) - details can come from a login profile - useful for online deployments
- [DONE implicitly] Specify the endpoint name
- [DONE except for namespace] Upload file and provision API using file contents type information, namespace, and endpoint name
- Access API via:
/namespace/endpoint_name?....
-
During provisioning workflow, allow an API access key to be generated for the endpoint (or for all endpoints in the login account)
- To access API protected endpoints, the user should send an
apikey
header variable - (Study FastAPI docs re: protecting endpoints)
- To access API protected endpoints, the user should send an
-
Allow users to clone an API endpoint
-
Provide integrations (which will require system
upload
andmanagement
APIs)- Dropbox / OneDrive / GDrive listeners
- Zapier, IFTTT, etc.
-
Customized client-side error handling hooks (400s, 500s, etc.)
-
The autogenerated API docs should be specific to each
namespace + endpoint
combination, rather than thehost domain
http://localhost:8000/namespace/endpoint/docs
, nothttp://localhost:8000/docs
The APINESS Factor
Resources
There are tons of resources on the web. These are some I have looked at:
- FastAPI
- Quickly Develop Highly Performant APIs with FastAPI & Python
- Deploying a Python FastAPI on Azure App Service
- How to capture arbitrary paths at one route in FastAPI
- Sharing python objects across multiple workers
- Why You Should Try FastAPI
- Video: How to Deploy FastAPI on Azure App Service in just 30 minutes
- Turn your Excel Workbook into a SQLite Database
- Uvicorn + FastAPI
- Dynamic/runtime method creation (code generation) in Python
- FastAPI as primary ASGI app
- Python Threading
- DbVisualizer
- SQLite command line shell
Datasette
This is an excellent, very complete solution of Excel files >> SQLite with vizualisations.
(Found after I started my project; so is there any need to develop this solution further??)
Epilogue
Looks like some folks are making a living from this sort of thing!