Production usage
Keep account state up to date
Use REST to submit orders and private WebSocket subscriptions to follow what happens next. Order, fill, balance and position events update your application's account state as changes occur.
- Connect the private streams and watch for subscription acknowledgements or errors before submitting orders. Load any starting state the streams do not provide, and account for events received while that data loads.
- Submit the order through REST. Save the returned IDs and inspect the response for errors or immediate results. An acknowledgement alone does not establish the order's later state.
- Apply matching WebSocket events to local state. Use order and fill events to track execution, and balance or position updates for the account changes your application needs. These events can arrive separately.
- Use REST to fill a specific gap: startup data, updates missed during a disconnect, conflicting events, or an outcome that neither the response nor the stream has resolved. Once account state is current, continue applying WebSocket updates.
Do not query REST after every submission or poll it while the required streams are healthy and complete. A quiet stream can mean nothing has changed. Check connection and subscription health separately from account activity.
Trigger follow-up work from position changes
Where the exchange and market provide position updates, they are useful triggers for follow-up logic. For example, when a position opens or increases, use its updated size to place or adjust stop-loss and take-profit (SL/TP) orders. Compare the previous and current position state; receiving another position message does not necessarily mean the position changed.
For signed position sizes, compare absolute exposure within the same direction and handle a direction change separately. Also resize or cancel obsolete protection when exposure shrinks or the position closes.
Debounce bursts of partial-fill updates to reduce API calls, and set a maximum wait so continuous updates cannot postpone adjustments indefinitely. Choose the debounce delay and maximum wait to fit the exchange's rate limits and the delay your strategy can tolerate.
When applying an adjustment:
- Check stream health and use the latest reconciled position and order state.
- Process changes for the same position sequentially and skip adjustments that are already satisfied.
- Pause dependent adjustments while recovering account state.
- Include the account, market and position side in state keys where applicable.
Recover after a connection gap
Use the SDK's reconnect lifecycle events to mark affected state as stale when connection recovery starts and to trigger REST reconciliation when the socket reopens. The event names differ by SDK; see the recovery section below. A reopened socket does not prove that authentication, subscriptions or account-state recovery are complete.
Confirm the required streams are healthy, load any missed order, fill, balance or position state, and account for updates received during those reads. Then compare current positions and open orders with the intended state and create, amend or cancel orders only where needed, including SL/TP orders. Resolve uncertain requests before submitting replacements so recovery does not create duplicates. Resume dependent actions once their required state is current.
See Account State for the full workflow. The private stream guide shows how to receive updates; the examples print events rather than maintain a complete account-state store.
The Advanced Trade user feed supplies order state. Load balances, positions and any other fields not covered by the selected feed separately. For decisions based on positions, use a source that covers your product.
Set timeouts and bound concurrency
Set an HTTP timeout for each operation and limit concurrent requests across workers. See Configuration for client options.
Retry reads deliberately
Retry transient read failures with a maximum attempt count or deadline, exponential backoff and jitter. Respect server retry guidance. Correct invalid parameters, permissions and signatures before retrying.
Resolve uncertain order outcomes
Persist the client order ID before submission and the exchange order ID when returned. Inspect the response for business errors. Then let the private stream track order changes. If a timeout or disconnect leaves the outcome unresolved, reconcile the original request with getOrder(), getOrders() and getFills() as appropriate; do not blindly recreate it.
Follow pagination. An empty first page or a delayed stream message does not prove that an order was never accepted.
Recover streaming state
reconnect signals recovery starting; mark affected state stale. reconnected signals a reopened socket and can trigger reconciliation, but does not establish that subscriptions, authentication or account recovery have completed.
The SDK reconnects and resubscribes cached topics automatically; no manual resubscription is needed. Process incoming snapshots and updates, repair missing state with REST, and account for events received while that repair runs. Resume actions only after the state they depend on is current. Never assume missed updates are replayed in full.
Use channel-specific sequence and snapshot rules for market data. Heartbeats show activity, not the correctness of a reconstructed book. A quiet order feed may simply mean no orders changed.
Account and rate-limit boundaries
Apply request budgets across workers sharing an account or key. Coinbase limits vary by API family and endpoint. Keep system clocks synchronized for signing, validate permissions on startup, and separate credentials for different products and environments.
Logging
Record operation names, response codes, duration, retries and reconnects. Select fields using the SDK's error format; whole request and error objects can contain credentials.
Shut down cleanly
Stop new work, settle or reconcile outstanding requests, call WebsocketClient.closeAll() and clear application timers. The stream example includes shutdown handlers.
Upgrading
Review release notes and run integration checks before upgrading. Commit the lockfile to reproduce deployed versions.