Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit

Automatically generated by python-semantic-release

Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Gracefully manage your API interactions

Actions Status PyPI Downloads GitHub Code style: black try/except style: tryceratops Types: mypy Follow guilatrova Sponsor guilatrova

Gracy helps you handle failures, logging, retries, throttling, and tracking for all your HTTP interactions. Gracy uses httpx under the hood.

"Let Gracy do the boring stuff while you focus on your application"


🧑‍💻 Get started


pip install gracy


poetry add gracy


Examples will be shown using the PokeAPI.

Simple example

# 0. Import
import asyncio
import typing as t
from gracy import BaseEndpoint, Gracy, GracyConfig, LogEvent, LogLevel

# 1. Define your endpoints
class PokeApiEndpoint(BaseEndpoint):
    GET_POKEMON = "/pokemon/{NAME}" # 👈 Put placeholders as needed

# 2. Define your Graceful API
class GracefulPokeAPI(Gracy[str]):
    class Config:  # type: ignore
        BASE_URL = "" # 👈 Optional BASE_URL
        # 👇 Define settings to apply for every request
        SETTINGS = GracyConfig(
          log_response=LogEvent(LogLevel.INFO, "{URL} took {ELAPSED}"),
            "default": lambda r: r.json()

    async def get_pokemon(self, name: str) -> t.Awaitable[dict]:
        return await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})

pokeapi = GracefulPokeAPI()

async def main():
      pokemon = await pokeapi.get_pokemon("pikachu")


More examples


Strict/Allowed status code

By default Gracy considers any successful status code (200-299) as successful.


You can modify this behavior by defining a strict status code or increase the range of allowed status codes:

from http import HTTPStatus


or a list of values:

from http import HTTPStatus

  strict_status_code={HTTPStatus.OK, HTTPStatus.CREATED}

Using strict_status_code means that any other code not specified will raise an error regardless of being successful or not.


You can also keep the behavior, but extend the range of allowed codes.

from http import HTTPStatus


or a list of values

from http import HTTPStatus

  allowed_status_code={HTTPStatus.NOT_FOUND, HTTPStatus.FORBIDDEN}

Using allowed_status_code means that all successful codes plus your defined codes will be considered successful.

This is quite useful for parsing as you'll see soon.

⚠️ Note that strict_status_code takes precedence over allowed_status_code, probably you don't want to combine those. Prefer one or the other.

Custom Validators

You can implement your own custom validator to do further checks on the response and decide whether to consider the request failed (and as consequence trigger retries if they're set).

from gracy import GracefulValidator

class MyException(Exception):

class MyCustomValidator(GracefulValidator):
    def check(self, response: httpx.Response) -> None:
        jsonified = response.json()
        if jsonified.get('error', None):
          raise MyException("Error is not expected")

        return None


class Config:
  SETTINGS = GracyConfig(
    retry=GracefulRetry(retry_on=MyException, ...),  # Set up retry to work whenever our validator fails
    validators=MyCustomValidator(),  # Set up validator


Parsing allows you to handle the request based on the status code returned.

The basic example is parsing json:

    "default": lambda r: r.json()

In this example all successful requests will automatically return the json() result.

You can also narrow it down to handle specific status codes.

class Config:
  SETTINGS = GracyConfig(
      "default": lambda r: r.json()
      HTTPStatusCode.NOT_FOUND: None

async def get_pokemon(self, name: str) -> dict| None:
  # 👇 Returns either dict or None
  return await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})

Or even customize exceptions to improve your code readability:

class PokemonNotFound(GracyUserDefinedException):
  ... # More on exceptions below

class Config:
      "default": lambda r: r.json()
      HTTPStatusCode.NOT_FOUND: PokemonNotFound

async def get_pokemon(self, name: str) -> Awaitable[dict]:
  # 👇 Returns either dict or raises PokemonNotFound
  return await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})


Who doesn't hate flaky APIs? 🙋

Yet there're many of them.

Using tenacity, backoff, retry, aiohttp_retry, and any other retry libs is NOT easy enough. 🙅

You still would need to code the implementation for each request which is annoying.

Here's how Gracy allows you to implement your retry logic:

class Config:
Parameter Description Example
delay How many seconds to wait between retries 2 would wait 2 seconds, 1.5 would wait 1.5 seconds, and so on
max_attempts How many times should Gracy retry the request? 10 means 1 regular request with additional 10 retries in case they keep failing. 1 should be the minimum
delay_modifier Allows you to specify increasing delay times by multiplying this value to delay Setting 1 means no delay change. Setting 2 means delay will be doubled every retry
retry_on Should we retry for which status codes/exceptions? None means for any non successful status code or exception HTTPStatus.BAD_REQUEST, or {HTTPStatus.BAD_REQUEST, HTTPStatus.FORBIDDEN}, or Exception or {Exception, HTTPStatus.NOT_FOUND}
log_before Specify log level. None means don't log More on logging later
log_after Specify log level. None means don't log More on logging later
log_exhausted Specify log level. None means don't log More on logging later
behavior Allows you to define how to deal if the retry fails. pass will accept any retry failure pass or break (default)


Rate limiting issues? No more.

Gracy helps you proactively deal with it before any API throws 429 in your face.

Creating rules

You can define rules per endpoint using regex:

SIMPLE_RULE = ThrottleRule(
# Output: "2 requests per second for URLs matching re.compile('.*')"

COMPLEX_RULE = ThrottleRule(
  per_time=timedelta(minutes=1, seconds=30),
# Output: 10 requests per 90 seconds for URLs matching re.compile('.*\\/pokemon\\/.*')

Setting throttling

You can set up logging and assign rules as:

class Config:
        rules=ThrottleRule(r".*", 2), # 2 reqs/s for any endpoint


You can define and customize logs for events by using LogEvent and LogLevel:

verbose_log = LogEvent(LogLevel.CRITICAL)
custom_warn_log = LogEvent(LogLevel.WARNING, custom_message="{METHOD} {URL} is quite slow and flaky")
custom_error_log = LogEvent(LogLevel.INFO, custom_message="{URL} returned a bad status code {STATUS}, but that's fine")

Note that placeholders are formatted and replaced later on by Gracy based on the event type, like:

Placeholders per event

Placeholder Description Example Supported Events
{URL} Full url being targetted All
{UURL} Full Unformatted url being targetted{NAME} All
{ENDPOINT} Endpoint being targetted /pokemon/pikachu All
{UENDPOINT} Unformatted endpoint being targetted /pokemon/{NAME} All
{METHOD} HTTP Request being used GET, POST All
{STATUS} Status code returned by the response 200, 404, 501 After Request, On request errors
{ELAPSED} Amount of seconds taken for the request to complete Numeric After Request, On request errors
{RETRY_DELAY} How long Gracy will wait before repeating the request Numeric Any Retry event
{CUR_ATTEMPT} Current attempt count for the current request Numeric Any Retry event
{MAX_ATTEMPT} Max attempt defined for the current request Numeric Any Retry event
{THROTTLE_LIMIT} How many reqs/s is defined for the current request Numeric Any Throttle event
{THROTTLE_TIME} How long Gracy will wait before calling the request Numeric Any Throttle event
{THROTTLE_TIME_RANGE} Time range defined by the throttling rule second, 90 seconds Any Throttle event

and you can set up the log events as follows:


  1. Before request
  2. After response
  3. Response has non successful errors


  1. Before retry
  2. After retry
  3. When retry exhausted


  1. When reqs/s limit is reached
  2. When limit decreases again

Dynamic Customization

You can customize it even further by passing a lambda:

    lambda r: "Request failed with {STATUS}" f" and it was {'redirected' if r.is_redirect else 'NOT redirected'}"
    if r
    else "",

Consider that:

  • Not all log events have the response available, so you need to guard yourself against it
  • Placeholders still works (e.g. {STATUS})
  • You need to watch out for some attrs that might break the formatting logic (e.g. r.headers)

Custom Exceptions

You can define custom exceptions for more fine grained control over your exception messages/types.

The simplest you can do is:

from gracy import Gracy, GracyConfig
from gracy.exceptions import GracyUserDefinedException

class MyCustomException(GracyUserDefinedException):

class MyApi(Gracy[str]):
  class Config:
    SETTINGS = GracyConfig(
        HTTPStatus.BAD_REQUEST: MyCustomException

This will raise your custom exception under the conditions defined in your parser.

You can improve it even further by customizing your message:

class PokemonNotFound(GracyUserDefinedException):
    BASE_MESSAGE = "Unable to find a pokemon with the name [{NAME}] at {URL} due to {STATUS} status"

    def _format_message(self, request_context: GracyRequestContext, response: httpx.Response) -> str:
        format_args = self._build_default_args()
        name = request_context.endpoint_args.get("NAME", "Unknown")
        return self.BASE_MESSAGE.format(NAME=name, **format_args)



Recommended for production environments.

Gracy reports a short summary using

pokeapi = GracefulPokeAPI()
# do stuff with your API

# OUTPUTGracy tracked that '{NAME}' was hit 1 time(s) with a success rate of 100.00%, avg latency of 0.45s, and a rate of 1.0 reqs/s.
❯ Gracy tracked a total of 2 requests with a success rate of 100.00%, avg latency of 0.24s, and a rate of 1.0 reqs/s.


Uses print to generate a short list with all attributes:

pokeapi = GracefulPokeAPI()
# do stuff with your API

  / ___|_ __ __ _  ___ _   _
 | |  _| '__/ _` |/ __| | | |
 | |_| | | | (_| | (__| |_| |
  \____|_|  \__,_|\___|\__, |
                       |___/  Requests Summary Report

    Total Reqs (#): 1
       Success (%): 100.00%
          Fail (%): 0.00%
   Avg Latency (s): 0.39
   Max Latency (s): 0.39
         2xx Resps: 1
         3xx Resps: 0
         4xx Resps: 0
         5xx Resps: 0
      Avg Reqs/sec: 1.0 reqs/s

    Total Reqs (#): 1
       Success (%): 100.00%
          Fail (%): 0.00%
   Avg Latency (s): 0.04
   Max Latency (s): 0.04
         2xx Resps: 1
         3xx Resps: 0
         4xx Resps: 0
         5xx Resps: 0
      Avg Reqs/sec: 1.0 reqs/s

    Total Reqs (#): 2
       Success (%): 100.00%
          Fail (%): 0.00%
   Avg Latency (s): 0.21
   Max Latency (s): 0.00
         2xx Resps: 2
         3xx Resps: 0
         4xx Resps: 0
         5xx Resps: 0
      Avg Reqs/sec: 1.0 reqs/s


It requires you to install Rich.

pokeapi = GracefulPokeAPI()
# do stuff with your API

Here's an example of how it looks:


Replay requests

Gracy allows you to replay requests and responses from previous interactions.

This is powerful because it allows you to test APIs without latency or consuming your rate limit. Now writing unit tests that relies on third-party APIs is doable.

It works in two steps:

Step Description Hits the API?
1. Recording Stores all requests/responses to be later replayed Yes
2. Replay Returns all previously generated responses based on your request as a "replay" No


The effort to record requests/responses is ZERO. You just need to pass a recording config to your Graceful API:

from gracy import GracyReplay
from gracy.replays.storages.sqlite import SQLiteReplayStorage

record_mode = GracyReplay("record", SQLiteReplayStorage("pokeapi.sqlite3"))
pokeapi = GracefulPokeAPI(record_mode)

Every request will be recorded to the defined data source.


Once you have recorded all your requests you can enable the replay mode:

from gracy import GracyReplay
from gracy.replays.storages.sqlite import SQLiteReplayStorage

replay_mode = GracyReplay("replay", SQLiteReplayStorage("pokeapi.sqlite3"))
pokeapi = GracefulPokeAPI(replay_mode)

Every request will be routed to the defined data source resulting in faster responses.

⚠️ Note that parsers, retries, throttling, and similar configs will work as usual.

Advanced Usage

Customizing/Overriding configs per method

APIs may return different responses/conditions/payloads based on the endpoint.

You can override any GracyConfig on a per method basis by using the @graceful decorator.

NOTE: Use @graceful_generator if your function uses yield.

from gracy import Gracy, GracyConfig, GracefulRetry, graceful, graceful_generator

retry = GracefulRetry(...)

class GracefulPokeAPI(Gracy[PokeApiEndpoint]):
    class Config:  # type: ignore
        BASE_URL = ""
        SETTINGS = GracyConfig(
                LogLevel.ERROR, "How can I become a pokemon master if {URL} keeps failing with {STATUS}"

        retry=None, # 👈 Disables retry set in Config
        log_errors=None, # 👈 Disables log_errors set in Config
            "default": lambda r: r.json()["order"],
            HTTPStatus.NOT_FOUND: None,
    async def maybe_get_pokemon_order(self, name: str):
        val: str | None = await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})
        return val

    @graceful( # 👈 Retry and log_errors are still set for this one
      parser={"default": lambda r: r.json()["order"]},
    async def get_pokemon_order(self, name: str):
      val: str = await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})
      return val

    @graceful_generator( # 👈 Retry and log_errors are still set for this one
      parser={"default": lambda r: r.json()["order"]},
    async def get_2_pokemons(self):
      names = ["charmander", "pikachu"]

      for name in names:
          r = await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})
          yield r

Customizing HTTPx client

You might want to modify the HTTPx client settings, do so by:

class YourAPIClient(Gracy[str]):
    class Config:  # type: ignore

    def __init__(self, token: token) -> None:
        self._token = token

    # 👇 Implement your logic here
    def _create_client(self) -> httpx.AsyncClient:
        client = super()._create_client()
        client.headers = {"Authorization": f"token {self._token}"}  # type: ignore
        return client

Overriding default request timeout

As default Gracy won't enforce a request timeout.

You can define your own by setting it on Config as:

class GracefulAPI(GracyApi[str]):
  class Config:
    BASE_URL = ""
    REQUEST_TIMEOUT = 10.2  # 👈 Here

Creating a custom Replay data source

Gracy was built with extensibility in mind.

You can create your own storage to store/load anywhere (e.g. SQL Database), here's an example:

import httpx
from gracy import GracyReplayStorage

class MyCustomStorage(GracyReplayStorage):
  def prepare(self) -> None: # (Optional) Executed upon API instance creation.

  async def record(self, response: httpx.Response) -> None:
    ... # REQUIRED. Your logic to store the response object. Note the httpx.Response has request data.

  async def load(self, request: httpx.Request) -> httpx.Response:
    ... # REQUIRED. Your logic to load a response object based on the request.

# Usage
record_mode = GracyReplay("record", MyCustomStorage())
replay_mode = GracyReplay("replay", MyCustomStorage())

pokeapi = GracefulPokeAPI(record_mode)

Hooks before/after request

You can set up hooks simply by defining async def before and async def after methods.

⚠️ NOTE: Gracy configs are disabled within these methods which means that retries/parsers/throttling won't take effect inside it.

class GracefulPokeAPI(Gracy[PokeApiEndpoint]):
    class Config:  # type: ignore
        BASE_URL = ""
        SETTINGS = GracyConfig(
            parser={HTTPStatus.NOT_FOUND: None},

    def __init__(self, *args: t.Any, **kwargs: t.Any) -> None:
        self.before_count = 0

        self.after_status_counter = defaultdict[HTTPStatus, int](int)
        self.after_aborts = 0
        self.after_retries_counter = 0

        super().__init__(*args, **kwargs)

    async def before(self, context: GracyRequestContext):
        self.before_count += 1

    async def after(
        context: GracyRequestContext, # Current request context
        response_or_exc: httpx.Response | Exception,  # Either the request or an error
        retry_state: GracefulRetryState | None,  # Set when this is generated from a retry
        if retry_state:
            self.after_retries_counter += 1

        if isinstance(response_or_exc, httpx.Response):
            self.after_status_counter[HTTPStatus(response_or_exc.status_code)] += 1
            self.after_aborts += 1

    async def get_pokemon(self, name: str):
        return await self.get(PokeApiEndpoint.GET_POKEMON, {"NAME": name})

In the example above invoking get_pokemon() will trigger before()/after() hooks in sequence.

📚 Extra Resources

Some good practices I learned over the past years guided Gracy's philosophy, you might benefit by reading:

Change log





Thanks to the last three startups I worked which forced me to do the same things and resolve the same problems over and over again. I got sick of it and built this lib.

Most importantly: Thanks to God, who allowed me (a random 🇧🇷 guy) to work for many different 🇺🇸 startups. This is ironic since due to God's grace, I was able to build Gracy. 🙌

Also, thanks to the httpx and rich projects for the beautiful and simple APIs that powers Gracy.