Configuration¶
LokiConfig holds every setting of a client. It is immutable after construction,
validated when it is built (invalid values raise a CONFIG error) and its string/debug form
redacts every credential. The Grafana Loki fields (tenant, Basic/Bearer credentials and the
cardinality guard) are listed in Grafana Loki settings below.
Common settings¶
| Setting | Default | Environment variable |
|---|---|---|
url — endpoint base URL |
http://127.0.0.1:3100 |
GROW_GRAFANA_LOKI_URL |
connectTimeout — connection setup |
10 s | GROW_GRAFANA_LOKI_CONNECT_TIMEOUT |
requestTimeout — per attempt (headers + body) |
30 s | GROW_GRAFANA_LOKI_REQUEST_TIMEOUT |
totalTimeout — whole call incl. retries and backoff |
120 s | GROW_GRAFANA_LOKI_TOTAL_TIMEOUT |
retry — retry policy (below) |
3 retries | GROW_GRAFANA_LOKI_MAX_RETRIES |
tls.caCertPath / tls.caCertPem — extra CA |
none | GROW_GRAFANA_LOKI_CA_CERT (path) |
tls.clientCertPath + tls.clientKeyPath — mutual TLS |
none | GROW_GRAFANA_LOKI_CLIENT_CERT, GROW_GRAFANA_LOKI_CLIENT_KEY |
tls.insecureSkipVerify — tests only, logs a warning once |
false |
GROW_GRAFANA_LOKI_INSECURE_SKIP_VERIFY |
proxy — proxy URL |
none (environment proxies are ignored) | GROW_GRAFANA_LOKI_PROXY |
headers — extra request headers |
{} |
— |
userAgentSuffix — appended to the User-Agent |
none | — |
hooks — request/response callbacks (Hooks) |
none | — |
logger — diagnostic logger (Hooks) |
silent | — |
Every request sends User-Agent: grow-codes-grafana-loki/<version> (<language>).
Grafana Loki settings¶
| Setting | Default | Environment variables (precedence order) |
|---|---|---|
url |
http://127.0.0.1:3100 |
GROW_GRAFANA_LOKI_URL, LOKI_URL |
tenantId — sent as X-Scope-OrgID on every request and on the tail handshake |
none (header not sent) | GROW_GRAFANA_LOKI_TENANT_ID, LOKI_TENANT_ID |
username — HTTP Basic user (gateways, Grafana Cloud instance id) |
none | GROW_GRAFANA_LOKI_USERNAME, LOKI_USERNAME, LOKI_BASIC_AUTH_USER |
password — HTTP Basic password / access token (secret) |
none | GROW_GRAFANA_LOKI_PASSWORD, LOKI_PASSWORD, LOKI_BASIC_AUTH_PASSWORD |
bearerToken — Authorization: Bearer … (secret) |
none | GROW_GRAFANA_LOKI_BEARER_TOKEN, LOKI_BEARER_TOKEN |
multiTenantQueries — allow tenant ids joined with \| |
false |
— |
allowedLabels — cardinality guard allowlist |
the default list below | — |
disableLabelGuard — push sends every stream label as an index label |
false |
— |
Field names follow each language: tenant_id, bearer_token, allowed_labels, … in Python and
Rust, TenantID, BearerToken, AllowedLabels, … in Go.
Validation (CONFIG error, the message never contains a value):
usernameandpasswordare given together; Basic and Bearer are mutually exclusive; a blank bearer token is rejected.- Tenant id: not empty, at most 150 bytes, not
.or.., onlyA-Z a-z 0-9 ! - _ . * ' ( ).a|b(a multi-tenant query) is accepted only withmultiTenantQueries(the server must enablemulti_tenant_queries_enabled); every part must then be valid.withTenant(id)applies the same rules (VALIDATIONerror). - Every
allowedLabelsentry is a valid label name ([a-zA-Z_][a-zA-Z0-9_]*, not starting with__).
Loki itself has no authentication: auth_enabled: true only makes the tenant header
mandatory. Basic and Bearer credentials are for gateways in front of Loki and for hosted Loki
(Grafana Cloud: user = numeric instance id, password = access token). The aliases
LOKI_BASIC_AUTH_USER / LOKI_BASIC_AUTH_PASSWORD are the names the Grow Codes platform already
uses.
Cardinality guard¶
Every label value creates a new Loki stream, so high-cardinality labels (request ids, user ids,
trace ids) make Loki slow and expensive. push and the log exporter therefore keep only
allowlisted names as index labels and move every other label or attribute to the entry's
structured metadata (an entry's own metadata wins). Structured metadata is stored next to the
line and can still be filtered: {service_name="api"} | request_id="r-1".
Default allowlist: service_name, service_namespace, deployment_environment, host_name,
level, job, instance, namespace, pod, container. Set allowedLabels to change it, or
disableLabelGuard to send push streams unchanged (the LogRecord exporter always applies the
guard). Attribute keys are normalized the way Loki normalizes OTLP attributes (service.name →
service_name).
From the environment¶
fromEnv resolves each setting with this precedence:
- the value passed in code (overrides),
GROW_GRAFANA_LOKI_<NAME>,- the vendor variables of Grafana Loki (see Grafana Loki settings), in their documented order,
- the default.
Parsing rules:
- Durations accept integer milliseconds (
1500) or Go-style units:500ms,10s,2m,1h, combinations such as1m30s. - Booleans accept
true/false,1/0,yes/no,on/off(any case). - An empty variable counts as unset.
- A present but invalid value raises a
CONFIGerror that names the variable, never its value.
import { LokiClient, LokiConfig } from "@grow-codes/grafana-loki";
// Environment only
const client = LokiClient.fromEnv();
// Environment with explicit overrides (explicit values win)
const config = LokiConfig.fromEnv({ requestTimeoutMs: 5_000 });
// Everything in code
const explicit = new LokiConfig({
url: "https://grafana-loki.example.com",
tenantId: "team-a",
bearerToken: process.env.MY_LOKI_TOKEN,
connectTimeoutMs: 5_000,
requestTimeoutMs: 15_000,
totalTimeoutMs: 60_000,
retry: { maxRetries: 5, initialBackoffMs: 200 },
tls: { caCertPath: "/etc/ssl/internal-ca.pem" },
proxy: "http://proxy.internal:3128",
headers: { "x-team": "payments" },
userAgentSuffix: "my-app/1.4",
});
const custom = new LokiClient(explicit);
console.log(String(explicit)); // bearerToken shown as [REDACTED]
use std::time::Duration;
use grow_grafana_loki::{LokiClient, LokiConfig, RetryPolicy};
// Environment only
let client = LokiClient::from_env()?;
// Environment with explicit overrides (explicit values win)
let config = LokiConfig::builder()
.request_timeout(Duration::from_secs(5))
.build_from_env()?;
// Everything in code
let explicit = LokiConfig::builder()
.url("https://grafana-loki.example.com")
.tenant_id("team-a")
.bearer_token(std::env::var("MY_LOKI_TOKEN").unwrap_or_default())
.connect_timeout(Duration::from_secs(5))
.request_timeout(Duration::from_secs(15))
.total_timeout(Duration::from_secs(60))
.retry(RetryPolicy::default().with_max_retries(5))
.ca_cert_path("/etc/ssl/internal-ca.pem")
.proxy("http://proxy.internal:3128")
.header("x-team", "payments")
.user_agent_suffix("my-app/1.4")
.build()?;
let custom = LokiClient::new(explicit)?;
println!("{:?}", custom.config()); // bearer_token shown as [REDACTED]
// Environment only
client, err := loki.NewClientFromEnv()
// Everything in code: zero values mean "use the default"
cfg := loki.Config{
CommonConfig: loki.CommonConfig{
URL: "https://grafana-loki.example.com",
ConnectTimeout: 5 * time.Second,
RequestTimeout: 15 * time.Second,
TotalTimeout: time.Minute,
Retry: &loki.RetryPolicy{MaxRetries: 5, InitialBackoff: 200 * time.Millisecond},
TLS: loki.TLSConfig{CACertPath: "/etc/ssl/internal-ca.pem"},
Proxy: "http://proxy.internal:3128",
Headers: map[string]string{"x-team": "payments"},
UserAgentSuffix: "my-app/1.4",
},
TenantID: "team-a",
BearerToken: loki.NewSecret(os.Getenv("MY_LOKI_TOKEN")),
}
if err := cfg.Validate(); err != nil {
log.Fatal(err) // errors.Is(err, loki.ErrConfig)
}
custom, err := loki.NewClient(cfg)
// Environment with explicit overrides (explicit values win; nil env = process environment)
merged, err := loki.ConfigFromEnvWith(loki.Config{
CommonConfig: loki.CommonConfig{RequestTimeout: 5 * time.Second},
}, nil)
fmt.Println(merged) // credentials shown as [REDACTED]
import os
from grow_grafana_loki import RetryPolicy, TlsConfig, LokiClient, LokiConfig
# Environment only
client = LokiClient.from_env()
# Environment with explicit overrides (explicit values win)
config = LokiConfig.from_env(request_timeout=5.0)
# Everything in code: durations are float seconds (datetime.timedelta also works)
explicit = LokiConfig(
url="https://grafana-loki.example.com",
tenant_id="team-a",
bearer_token=os.environ.get("MY_LOKI_TOKEN"),
connect_timeout=5.0,
request_timeout=15.0,
total_timeout=60.0,
retry=RetryPolicy(max_retries=5, initial_backoff=0.2),
tls=TlsConfig(ca_cert_path="/etc/ssl/internal-ca.pem"),
proxy="http://proxy.internal:3128",
headers={"x-team": "payments"},
user_agent_suffix="my-app/1.4",
)
custom = LokiClient(explicit)
print(repr(explicit)) # bearer_token shown as [REDACTED]
import java.time.Duration;
import codes.grow.packages.loki.LokiClient;
import codes.grow.packages.loki.LokiConfig;
import codes.grow.packages.loki.core.RetryPolicy;
// Environment only
LokiClient client = LokiClient.fromEnv();
// Everything in code
LokiConfig config = LokiConfig.builder()
.url("https://grafana-loki.example.com")
.tenantId("team-a")
.bearerToken(System.getenv("MY_LOKI_TOKEN"))
.connectTimeout(Duration.ofSeconds(5))
.requestTimeout(Duration.ofSeconds(15))
.totalTimeout(Duration.ofMinutes(1))
.retry(RetryPolicy.builder().maxRetries(5).initialBackoff(Duration.ofMillis(200)).build())
.proxy("http://proxy.internal:3128")
.header("x-team", "payments")
.userAgentSuffix("my-app/1.4")
.build();
LokiClient custom = LokiClient.create(config);
System.out.println(config); // bearerToken shown as [REDACTED]
Retry policy¶
| Field | Default | Meaning |
|---|---|---|
maxRetries |
3 | retries after the first attempt (so up to 4 attempts) |
initialBackoff |
100 ms | backoff cap of the first retry |
maxBackoff |
10 s | cap of every backoff |
multiplier |
2 | growth of the cap per retry |
maxRetryAfter |
60 s | longest server-requested Retry-After the client waits for |
Backoff for retry n (0-based) is full jitter: a random wait between 0 and
min(maxBackoff, initialBackoff × multiplier^n).
Requests are retried on HTTP 429, 408, 502, 503, 504 and on network errors. Idempotent
requests (GET HEAD OPTIONS PUT DELETE, or marked idempotent) are also retried on HTTP 500.
Other requests (for example POST) are retried only on 429 and on connection errors that happened
before the request was sent, never after an uncertain send.
On 429 and 503 the client honours Retry-After (seconds or HTTP date) and waits
max(retryAfter, backoff). When that wait exceeds maxRetryAfter or the remaining
totalTimeout, it stops at once with RATE_LIMITED/UNAVAILABLE and sets retryAfter on the
error. Set maxRetries to 0 to turn retries off.
Timeouts and cancellation¶
connectTimeoutbounds connection setup,requestTimeouteach attempt,totalTimeoutthe whole call including backoff sleeps. An exhausted timeout raisesTIMEOUT.- Cancellation is honoured during requests and backoff sleeps and raises
CANCELLED: TypeScript{ signal }(AbortSignal), Rust dropping the future, Goctx, Pythoncancel=(CancelToken) or task cancellation, Java…Asyncfutures (cancel(true)).
TLS and proxies¶
- Add a private CA with
tls.caCertPath(or the PEM text intls.caCertPem). - Mutual TLS needs
tls.clientCertPathandtls.clientKeyPathtogether. tls.insecureSkipVerifydisables certificate checks. Use it for local tests only; the client logs a warning once.HTTP_PROXY/HTTPS_PROXYare ignored so that credentials never leave through an unexpected proxy. Setproxy(orGROW_GRAFANA_LOKI_PROXY) to use one.