shardd

Official Python client for shardd — a globally distributed credit ledger with a sub-10ms write path in every region.

• Zero config — pass an API key; the SDK picks the closest healthy edge.
• Automatic failover — transient 5xx/timeouts fall over to the next region, reusing the idempotency nonce so retries collapse.
• Sync + async — Shardd and AsyncShardd share the same method surface.
• Fully typed — dataclasses, no runtime introspection magic.

Install

pip install shardd

30-second quickstart

import os
from shardd import Shardd

shardd = Shardd(os.environ["SHARDD_API_KEY"])

# Credit 500 units to user:alice in the `my-app` bucket.
result = shardd.create_event("my-app", "user:alice", 500)
print("new balance =", result.balance)

# Read back the whole bucket.
balances = shardd.get_balances("my-app")
for row in balances.accounts:
    print(f"{row.account} = {row.balance}")

Get an API key at https://app.shardd.xyz → Keys.

Async

import asyncio
from shardd import AsyncShardd

async def main():
    async with AsyncShardd(api_key) as shardd:
        result = await shardd.create_event("my-app", "user:alice", -100)
        print(result.balance)

asyncio.run(main())

API

Idempotency

Every create_event carries an idempotency_nonce. If you don't supply one, the SDK generates a UUID v4 for you. For safe retries, capture the nonce client-side and reuse it:

import uuid

nonce = str(uuid.uuid4())
result = shardd.create_event(
    "my-app", "user:alice", -100,
    note="order #9821",
    idempotency_nonce=nonce,
)
# A retry with the same `nonce` returns the original event and
# `result.deduplicated is True` — no double charge.

Failover behavior

The three prod regions (use1.api.shardd.xyz, euc1.api.shardd.xyz, ape1.api.shardd.xyz) are baked in as defaults. On the first request the client parallel-probes /gateway/health on all three, picks the lowest-latency healthy one, and pins it. If that edge returns 503/ 504/timeouts/connect-errors, the SDK marks it unavailable for 60s and retries the request once against the next-best candidate. Non-retryable errors (400, 401, 403, 404, 422) surface immediately — no retry, no failover.

Override the edges for local or self-hosted clusters:

shardd = Shardd(
    api_key,
    edges=[
        "http://localhost:8081",
        "http://localhost:8082",
        "http://localhost:8083",
    ],
)

Error handling

from shardd import Shardd, InsufficientFundsError, ShardError

try:
    shardd.create_event("my-app", "user:alice", -1000)
except InsufficientFundsError as err:
    print(f"short {1000 - err.available_balance} credits")
except ShardError as err:
    if err.retryable:
        # queue for retry
        ...
    else:
        raise

License

MIT © shardd