nicusX commented on code in PR #1: URL: https://github.com/apache/flink-connector-prometheus/pull/1#discussion_r1755176342
########## flink-connector-prometheus/README.md: ########## @@ -0,0 +1,260 @@ +## Flink Prometheus connector (sink) + +Implementation of the Prometheus sink connector for DataStream API. + +The sink writes to Prometheus using the Remote-Write interface, based +on [Remote-Write specifications version 1.0](https://prometheus.io/docs/concepts/remote_write_spec/) + +### Guarantees and input restrictions + +Due to the strict [ordering](https://prometheus.io/docs/concepts/remote_write_spec/#ordering) +and [format](https://prometheus.io/docs/concepts/remote_write_spec/#labels) requirements +of Prometheus Remote-Write, the sink guarantees that input data are written to Prometheus only if input data are in +order and well-formed. + +For efficiency, the connector does not do any validation. +If input is out of order or malformed, **the write request is rejected by Prometheus**. +If the write request is rejected, depending on the configured [error handling behaviour](#error-handling-behavior) for +"Prometheus non-retriable errors", the sink will either throw an exception (`FAIL`, default behavior) or discard the +offending +request and continue (`DISCARD_AND_CONTINUE`). See [error handling behaviour](#error-handling-behavior), below, for +further details. + +The sink receives as input time-series, each containing one or more samples. +To optimise the write throughput, input time-series are batched, in the order they are received, and written with a +single write-request. + +If a write-request contains any out-of-order or malformed data, **the entire request is rejected** and all time series +are discarded. +The reason is Remote-Write +specifications [explicitly forbids retrying](https://prometheus.io/docs/concepts/remote_write_spec/#retries-backoff) of +rejected write requests (4xx responses). +and the Prometheus response does not contain enough information to efficiently partially retry the write, discarding the +offending data. + +### Responsibilities of the application + +It is responsibility of the application sending the data to the sink in the correct order and format. + +1. Input time-series must be well-formed, e.g. only valid and non-duplicated labels, + samples in timestamp order (see [Labels and Ordering](https://prometheus.io/docs/concepts/remote_write_spec/#labels) + in Prometheus Remote-Write specs). +2. Input time-series with identical labels are sent to the sink in timestamp order. +3. If sink parallelism > 1 is used, the input stream must be partitioned so that all time-series with identical labels + go to the same sink subtask. A `KeySelector` is provided to partition input correctly ( + see [Partitioning](#partitioning), below). + +#### Sink input objects + +To help sending well-formed data to the sink, the connector +expect [`PrometheusTimeSeries`](./src/main/java/org/apache/flink/connector/prometheus/sink/PrometheusTimeSeries.java) +POJOs as input. + +Each `PrometheusTimeSeries` instance maps 1-to-1 to +a [remote-write `TimeSeries`](https://prometheus.io/docs/concepts/remote_write_spec/#protocol). Each object contains: + +* exactly one `metericName`, mapped to the special `__name__` label +* optionally, any number of additional labels { k: String, v:String } - MUST BE IN ORDER BY KEY +* one or more `Samples` { value: double, timestamp: long } - MUST BE IN TIMESTAMP ORDER + +`PrometheusTimeSeries` provides a builder interface. + +```java + +// List<Tuple2<Double, Long>> samples = ... + +PrometheusTimeSeries.Builder tsBuilder = PrometheusTimeSeries.builder() + .withMetricName("CPU") // mapped to `__name__` label + .addLabel("InstanceID", instanceId) + .addLabel("AccountID", accountId); + +for( +Tuple2<Double, Long> sample :samples){ + tsBuilder. + +addSample(sample.f0, sample.f1); +} + +PrometheusTimeSeries ts = tsBuilder.build(); +``` + +#### Input data validation + +Prometheus imposes strict constraints to the content sent to remote-write, including label format and ordering, sample +time ordering ecc. + +For efficiency, the sink **does not do any validation or reordering** of the input. It's responsibility +of the application ensuring that input is well-formed. + +Any malformed data will be rejected on write to Prometheus. Depending on +the [error handling behaviours](#error-handling-behavior) +configured, the sink will throw an exception stopping the job (default), or drop the entire write-request, log the fact, +and continue. + +For complete details about these constraints, refer to +the [remote-write specifications](https://prometheus.io/docs/concepts/remote_write_spec/). + +### Batching, blocking writes and retry + +The sink batches multiple time-series into a single write-request, retaining the order.. + +Batching is based on the number of samples. Each write-request contains up to 500 samples, with a max buffering time of +5 seconds +(both configurable). The number of time-series doesn't matter. + +As by [Prometheus Remote-Write specifications](https://prometheus.io/docs/concepts/remote_write_spec/#retries-backoff), +the sink retries 5xx and 429 responses. Retrying is blocking, to retain sample ordering, and uses and exponential +backoff. + +The exponential backoff starts with an initial delay (default 30 ms) and increases it exponentially up to a max retry +delay (default 5 sec). It continues retrying until the max number of retries is reached (default reties forever). + +On non-retriable error response (4xx, except 429, non retryable exceptions), or on reaching the retry limit, depending +on +the configured [error handling behaviour](#error-handling-behavior) for "Max retries exceeded", the sink will either +throw +an exception (`FAIL`, default behaviour), or **discard the entire write-request**, log a warning and continue. See +[error handling behaviour](#error-handling-behavior), below, for further details. + +### Initializing the sink + +Example of sink initialisation (for documentation purposes, we are setting all parameters to their default values): +Sink + +```java +PrometheusSink sink = PrometheusSink.builder() + .setMaxBatchSizeInSamples(500) // Batch size (write-request size), in samples (default: 500) + .setMaxRecordSizeInSamples(500) // Max sink input record size, in samples (default: 500), must be <= maxBatchSizeInSamples Review Comment: Good point. I will explain this in the actual docs, in the other PR -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
