Client¶
ZipkinClient is the entry point. Create one per process (or per set of credentials), share
it between threads/tasks and close it on shutdown. Service functions hang off the client (see
Functions).
| Function | Returns | Notes |
|---|---|---|
constructor / create / new |
client | validates the config (CONFIG error) |
fromEnv() |
client | config from GROW_ZIPKIN_* and vendor variables |
health() |
HealthStatus { healthy, status, details } |
unauthenticated; never throws for an unhealthy or unreachable service, only for CANCELLED and CONFIG |
ping() |
latency | round trip through the query API and storage (GET /api/v2/services); stock Zipkin has no authentication, a gateway in front may answer AUTHENTICATION |
serverVersion() |
version string | from GET /info; UNSUPPORTED when the server exposes none |
serverInfo() |
version and commit | from GET /info; empty fields when the server or a gateway answers 404 |
query / reporter / createReporter() |
sub-clients | trace search (Query) and span reporting (Reporter) |
initTracing() |
tracing handle | OpenTelemetry setup that exports to this Zipkin (Tracing setup) |
close() |
— | idempotent; flushes batch processors and stops watchers/renewers |
status is ok, degraded, unhealthy, unreachable (network error) or the vendor's word.
Create, use and close¶
import { ZipkinClient, ZipkinConfig } from "@grow-codes/zipkin";
const client = new ZipkinClient(new ZipkinConfig({ url: "http://127.0.0.1:9411" }));
const health = await client.health(); // { healthy, status, details }
const latencyMs = await client.ping(); // number (ms)
const version = await client.serverVersion(); // string
await client.close(); // safe to call twice
// Cancellation and per-call deadline: every I/O method accepts { signal }
const status = await client.health({ signal: AbortSignal.timeout(2_000) });
use grow_zipkin::{ZipkinClient, ZipkinConfig};
let config = ZipkinConfig::builder().url("http://127.0.0.1:9411").build()?;
let client = ZipkinClient::new(config)?;
let health = client.health().await?; // HealthStatus
let latency = client.ping().await?; // Duration
let version = client.server_version().await?; // String
client.close().await; // safe to call twice
// Cancellation: drop the future, e.g. with a deadline
let status = tokio::time::timeout(std::time::Duration::from_secs(2), client.health()).await;
Without an async runtime, enable the blocking feature:
client, err := zipkin.NewClient(zipkin.Config{
CommonConfig: zipkin.CommonConfig{URL: "http://127.0.0.1:9411"},
})
if err != nil {
return err
}
defer client.Close() // safe to call twice
health, err := client.Health(ctx) // HealthStatus
latency, err := client.Ping(ctx) // time.Duration
version, err := client.ServerVersion(ctx) // string
// Cancellation and deadlines come from ctx
ctx2, cancel := context.WithTimeout(ctx, 2*time.Second)
defer cancel()
health, err = client.Health(ctx2)
from grow_zipkin import CancelToken, ZipkinClient, ZipkinConfig
with ZipkinClient(ZipkinConfig(url="http://127.0.0.1:9411")) as client:
health = client.health() # HealthStatus
latency_ms = client.ping() # float (ms)
version = client.server_version() # str
# Cancel from another thread
token = CancelToken()
health = client.health(cancel=token)
# close() ran; calling it again is safe
AsyncZipkinClient has the same methods as coroutines (await client.health(),
await client.aclose(), async with), and honours task cancellation.
ZipkinConfig config = ZipkinConfig.builder().url("http://127.0.0.1:9411").build();
try (ZipkinClient client = ZipkinClient.create(config)) {
HealthStatus health = client.health();
Duration latency = client.ping();
String version = client.serverVersion();
// Async variants return CompletableFuture; cancel(true) cancels the request
CompletableFuture<HealthStatus> pending = client.healthAsync();
pending.orTimeout(2, TimeUnit.SECONDS).join();
} // close() is idempotent
Lifecycle rules¶
- The client is safe for concurrent use. Reuse it: it keeps a connection pool.
close()flushes queued work (for example log batches) within the configured deadline and is safe to call more than once. Any call on a closed client fails withCANCELLED("<operation>: client is closed").- Each language also offers its idiomatic scope: TypeScript
try/finally, Rustclose().await(plus best-effortDrop), Godefer client.Close(), Pythonwith/async with, Java try-with-resources.
Zipkin notes¶
initTracingon a closed client fails withCONFIG(notCANCELLED): it sets up a new tracing pipeline instead of calling Zipkin, so a closed client is a configuration mistake.- Stock Zipkin has no authentication. Credentials for a gateway in front of it go in
headers(see Configuration).