|
| 1 | +--- |
| 2 | +title: Span Sampling & Filtering |
| 3 | +--- |
| 4 | +<Alert level="info"> |
| 5 | + This document uses key words such as "MUST", "SHOULD", and "MAY" as defined in [RFC 2119](https://www.ietf.org/rfc/rfc2119.txt) to indicate requirement levels. |
| 6 | +</Alert> |
| 7 | + |
| 8 | +Any APIs exposed to the user to sample or filter spans must adhere to the following design principles: |
| 9 | + |
| 10 | +- The APIs are optimized for trace completeness |
| 11 | +- The APIs are optimized for conclusive sampling decisions |
| 12 | + |
| 13 | +## Sample root spans with `tracesSampleRate` & `tracesSampler` |
| 14 | + |
| 15 | +The SDK is automatically initialized with a `tracesSampleRate` of `0`. |
| 16 | +When starting a root span, the configured rate is compared against a random number between 0 and 1 to decide if this root span will be sampled or not. |
| 17 | + |
| 18 | +If the SDK is configured with a `tracesSampler`, the `tracesSampleRate` no longer applies. |
| 19 | +The `tracesSampler` callback must receive sufficient arguments from users to define their own sampling rules. |
| 20 | +This can include but is not limited to certain attributes from the root span, such as HTTP headers. |
| 21 | +The return value of the `tracesSampler` is a float between `0.0` and `1.0`. |
| 22 | + |
| 23 | +```js |
| 24 | +Sentry.init({ |
| 25 | + tracesSampler: ({ name, attributes, parentSampleRate }) => { |
| 26 | + // Inherit the trace parent's sample rate if there is one. Sampling is deterministic |
| 27 | + // for one trace, e.g. if the parent was sampled, all children will be sampled at the same rate. |
| 28 | + if (typeof parentSampleRate === "number") { |
| 29 | + return parentSampleRate; |
| 30 | + } |
| 31 | + |
| 32 | + // Else, use a default sample rate (replacing tracesSampleRate). |
| 33 | + return 0.5; |
| 34 | + } |
| 35 | +}) |
| 36 | +``` |
| 37 | + |
| 38 | +The `parentSampleRate` is a propagated value inside the baggage, using key `sentry-sample_rand`. |
| 39 | +The value stems from a truly random number between 0 and 1, generated when a new trace is started. If the SDK does not receive such a number in an incoming trace, a new, truly random number between 0 and 1 is generated. |
| 40 | + |
| 41 | +In the following cases, the SDK must compare sample rates against this `parentSampleRate` instead of `math.random()`: |
| 42 | + |
| 43 | + - When a `tracesSampler` is configured, i.e. `trace["sentry-sample_rand"] < tracesSampler()` |
| 44 | + |
| 45 | + - When the SDK is the head of trace, this applies to sample decisions based on `tracesSampleRate`, e.g. `trace['sentry-sample_rand'] < config.tracesSampleRate` |
| 46 | + |
| 47 | +If the `sentry-sample_rate` (`parentSampleRate`) is not available for any reason for an inbound trace, but the trace has the sampled flag set to true, the SDK injects `parentSampleRate: 1.0` into the `tracesSampler`. |
| 48 | + |
| 49 | +If no `tracesSampler` is configured, a propagated sampling decision via the traceparent takes precedence over the `tracesSampleRate`. This behavior can be disabled by defining a `tracesSampler`. |
| 50 | + |
| 51 | +## Parent Sampling Origins |
| 52 | + |
| 53 | +If the SDK can parse an org ID from the configured DSN, this value must be propagated as a baggage entry with the key `sentry-org`. Given a DSN of `https://[email protected]/1`, the org ID is 1, based on `o1`. |
| 54 | + |
| 55 | +On incoming traces, the SDK must compare the `sentry-org` baggage value against its own parsed value from the DSN. Only if both match, the parent sampling decisions applies. |
| 56 | + |
| 57 | +This behavior can be disabled by setting `strictTracePropagation: false` in the SDK init call. |
| 58 | + |
| 59 | +The SDK must be configurable with an optional `org: <org-id>` setting that takes precedence over the parsed value from the DSN. |
| 60 | + |
| 61 | +## Filter spans with `ignoreSpans` & integration config |
| 62 | + |
| 63 | +The SDK must implement a mechanism for users to filter out spans. The result must be binary (true/false). |
| 64 | +The `ignoreSpans` option accepts a glob pattern or string. |
| 65 | +The `integrations` option can perform in similar fashion or make explicit opt-out possible via a bool flag. |
| 66 | + |
| 67 | +If both options are not feasible to be implemented in certain SDKs, other approaches must be explored that have the same outcome. |
| 68 | + |
| 69 | +```js |
| 70 | +Sentry.init({ |
| 71 | + ignoreSpans: [ |
| 72 | + 'GET /about', |
| 73 | + 'events.signal *', |
| 74 | + ], |
| 75 | + integrations: [ |
| 76 | + fsIntegration: { |
| 77 | + ignoreSpans: [ |
| 78 | + 'fs.read', |
| 79 | + ], |
| 80 | + readSpans: true, |
| 81 | + writeSpans: false, |
| 82 | + } |
| 83 | + ] |
| 84 | +}) |
| 85 | +``` |
| 86 | + |
| 87 | +## Sanitize span attributes with `beforeSendSpans` |
| 88 | + |
| 89 | +This callback must not allow the removal of any spans from the span tree. |
| 90 | +It receives a deep copy of all spans in the span tree and their attributes. |
| 91 | + |
| 92 | +``` |
| 93 | +[ |
| 94 | + { |
| 95 | + 'name': 'GET /', |
| 96 | + 'attributes': [ |
| 97 | + 'http.request.method': 'GET', |
| 98 | + 'http.response.status_code': 200, |
| 99 | + ] |
| 100 | + }, |
| 101 | +] |
| 102 | +``` |
| 103 | + |
| 104 | +Users can mutate any exposed properties to perform sanitation on sensitive data or Pii. |
| 105 | +The return value `beforeSendSpans` should be merged with the original span tree prior to emission. |
0 commit comments