Retry & circuit breaker
createRetryStore's exponential backoff and per-instance circuit breaker.
createRetryStore wraps any ObjectStore with exponential-backoff retries
and a circuit breaker. Place it directly above the network store, under the
cache — see the production stack guide
for why that order matters.
createRetryStorecreateRetryStore(store: ObjectStore, options?: RetryOptions): ObjectStore| Parameter | Type | Description |
|---|---|---|
store | ObjectStore | The store to wrap. |
options? | RetryOptions | See below. |
Backoff options
| Option | Default | What it does |
|---|---|---|
retries | 3 | Attempts after the first (total attempts = retries + 1). |
initialDelayMs | 100 | Base delay, doubled every attempt. |
maxDelayMs | 5000 | Upper bound on the backoff delay. |
jitter | 0.3 | Random jitter added to each delay, as a fraction of it. |
isRetryable | see below | Decide whether an error is worth retrying. |
The store contract already maps "not found" to null rather than throwing
(see the ObjectStore guide) — so any
thrown error here is a genuine failure. The default isRetryable retries
network faults, throttling, and HTTP 5xx/429 responses; anything else
(a 4xx that isn't throttling, a programming error) fails immediately.
The circuit breaker
createRetryStore(store, {
breaker: { threshold: 5, resetMs: 30_000 }, // defaults shown
});After threshold consecutive retryable failures, the store stops even
trying — every call fails fast with CircuitOpenError (code: "EUNAVAILABLE") for resetMs, then lets exactly one request through to
probe whether the backend has recovered (concurrent calls keep failing fast
until that probe settles). Any success resets the count, and errors your
isRetryable rejects (a 403, a malformed request) never count — they say
nothing about the backend's health.
Fail-fast during an outage matters as much as retrying during a blip — without a breaker, a downed backend means every single request in flight pays the full retry+backoff cost before failing, which can turn a backend outage into a self-inflicted request pile-up.
Set breaker: false to disable it entirely.
Observability
createRetryStore(store, {
onRetry: (info) => console.warn('retrying', info),
});onRetry fires on every retry attempt — wire it into your existing logging/
metrics pipeline rather than debugging retries blind.