Make sure you have go 1.8+ installed, if you dont have go please install it here.
Make sure your GOROOT and GOPATH are set correctly.
Run the following to get the code:
go get -u github.com/SaitTalhaNisanci/fatLama-backend-challenge
- Go
- gorilla mux router and dispatcher
- testify testing
- sqlite database engine
CodeReviewComments is followed during the development.
/search?searchTerm=camera&lat=51.948&lng=0.172943
Search engine splits the searchTerm to its words, and for each item
in the database it checks if there is any matched word. Even if there is
one match with an item's name it will be loaded. For example If
searchTerm is canon camera, items that have canon or camera
or both will be loaded. When calculating a score of an item, the number
of matched words with its name and searchTerm are linearly combined with
the distance.
score = textMatchCoeff * #matchedWords - distanceCoeff*distance(item, searchLocation)
Note that distance is subtracted as the further it is the less relevant it is.
These coefficients can be set better with a real data and some observations.
Default page size is 20, which means at most 20 items will be returned.
If there is no content for a search query, http.StatusNoContent is returned.
If search parameters are not valid, http.StatusBadRequest is returned.
Currently if the number of requests per second increase the latency will increase
but as far as tested in server_test.go the server wont fail to serve. Note that it is
only a smoke test not a soak test.
If an item has duplicate words in its name they will be counted as one. So
an item name camera camera camera will be same as camera.
The entry point for the server is server.go. Router and database are
initialized there. The router currently has only one endpoint which is /search.
There are 4 packages:
db contains database methods, process method processes incoming
search queries. A query channel sends search queries to this method to
be processed.
handler contains handler functions. Currently there is only one handler, search handler. Search parameter validation etc are done in this package.
model contains database models. Currently there is only one model which is Item.
Item contains Name, Lat, Lng, Url, ImageUrls.
search package communicates with db package to load items and sort them by a score. Score is calculated based on linear combination of distance and matched words.
Different functionalities are separated into difference packages for better readability and testing.
Currently each request hits the database. For scalability and better latency in memory cache such as `redis, hazelcast can be used.
The code is tested extensively,
To run tests:
go test ./...
To run tests with race:
go test -race ./...
Travis can be added for CI to the repository.
Also note that currently the same database is used for testing and production but they should be separated.