# ScopeDB Rust SDK examples
Start with a quickstart, then choose a write pattern whose delivery tradeoffs match the workload. Every example uses only the public `scopedb-client` API. The shared client helper uses `Client::builder(...).api_key(...)`, so copied examples do not need a direct reqwest dependency. API keys belong only in a trusted process; never return one to an untrusted client.
These examples focus on SDK integration and assume valid ScopeQL. For language syntax, use the canonical [Quickstart](https://docs.scopedb.io/guides/quickstart), [query guide](https://docs.scopedb.io/guides/query-events), and [language reference](https://docs.scopedb.io/reference/).
The commands below run from the repository root.
## Read-only discovery
These examples can run against a reachable ScopeDB endpoint without modifying data.
| [`statement.rs`](statement.rs) | Query shorthand and object result rows | `cargo run --example statement` |
| [`catalog.rs`](catalog.rs) | Automatic REST catalog pagination and full resources | `cargo run --example catalog` |
| [`table.rs`](table.rs) | Quoted table identifiers and full table descriptions | `cargo run --example table` |
## Before running a write example
Every write example refuses to start unless `SCOPEDB_TABLE` names an existing, disposable table. Do not point these examples at production unless the writes are intentional.
The append, stream, telemetry, and transform examples can share this schema:
```sql
CREATE TABLE public.sdk_example_events (
id int,
event_id string,
occurred_at timestamp,
name string,
attributes object
);
```
Configuration comes from:
- `SCOPEDB_ENDPOINT` (defaults to `http://127.0.0.1:6543`)
- `SCOPEDB_API_KEY` (optional Bearer API key; preferred)
- `SCOPEDB_TOKEN` (fallback when `SCOPEDB_API_KEY` is unset or empty)
- `SCOPEDB_DATABASE` (defaults to `scopedb`)
- `SCOPEDB_SCHEMA` (defaults to `public`)
- `SCOPEDB_TABLE` (required for writes)
Set the disposable destination once before running a write example:
```sh
export SCOPEDB_TABLE=sdk_example_events
```
```powershell
$env:SCOPEDB_TABLE = "sdk_example_events"
```
## Choose a write journey
| [`append.rs`](append.rs) | The caller owns one exact NDJSON request boundary | `cargo run --example append` |
| [`append_stream.rs`](append_stream.rs) | The SDK should asynchronously batch rows with strict delivery | `cargo run --example append_stream` |
| [`bulk_append.rs`](bulk_append.rs) | A backfill needs bounded memory and concurrent strict batches | `cargo run --example bulk_append` |
| [`telemetry.rs`](telemetry.rs) | Logs or events need non-blocking, observable best-effort delivery | `cargo run --example telemetry` |
| [`ingest_transform.rs`](ingest_transform.rs) | JSON records need a SQL transform before insertion | `cargo run --example ingest_transform` |
`append.rs` sends exactly one NDJSON request. `append_stream.rs` uses the default `Stop` policy: `send()` and `send_all()` wait only for local admission, while a successful `flush()` or `shutdown()` is a remote commit barrier for the accepted prefix.
`bulk_append.rs` keeps producer memory bounded and sends multiple HTTP batches concurrently. It does not add durable resume, idempotency, transactionality, or whole-job rollback. Earlier concurrent batches may have committed even when a later batch fails.
`telemetry.rs` opts into `AppendFailurePolicy::Continue` and uses `try_send()` on the request path. It keeps working after a failed batch, but does not retain the batch for replay. The shutdown report makes rejected, ambiguous, and local loss observable.
When the continue-mode circuit is open, `try_send()` rejects immediately while the backpressured `send()` path can queue within the configured byte budget and wait for a later probe.
## Delivery contract
- The table append API accepts NDJSON only: one JSON row object per line, not a JSON array.
- `send()`, `send_all()`, and `Ok(())` from `try_send()` mean local admission; they do not confirm a remote commit.
- A successful strict barrier confirms that its accepted prefix committed.
- A continue-mode barrier is settlement. Always inspect its `AppendDeliveryReport`.
- The stream retries only an exact temporary HTTP batch explicitly reported as `rejected`.
- A timeout, transport failure, or invalid success response is `unknown`. The rows may already exist remotely, so never blindly replay that payload.
- `shutdown()` closes admission and settles accepted rows. It is not an abort or rollback; stop and join producer tasks before calling it.
- Dropping an enqueued `flush()` future does not cancel its remote settlement; keep the future alive to receive the interval report and use `stats()` for post-cancellation diagnostics.
- Concurrent batches do not have a defined commit order. Use `max_concurrent_batches(1)` when requests must be submitted serially.
- An in-memory stream is not a durable queue. Audit or billing writes need an application-owned outbox and a reconciliation path for unknown outcomes.
## Check the examples
Compile every example without running it:
```sh
cargo check --examples
```