Skip to content

examples.rst

Latest commit

 

History

History
662 lines (478 loc) · 20.6 KB

examples.rst

File metadata and controls

662 lines (478 loc) · 20.6 KB

Examples

Connecting to Feldera Sandbox

Ensure that you have an API key to connect to Feldera Sandbox.

To get the key:

  • Login to the Feldera Sandbox.
  • Click on the top right button that says: "Logged in"
  • Click on "Manage API keys"
  • Generate a new API key
  • Give it a name, and copy the API key
from feldera import FelderaClient, PipelineBuilder

client = FelderaClient('https://try.feldera.com', api_key=api_key)

pipeline = PipelineBuilder(client, name, sql).create()

Connecting to Feldera on localhost

from feldera import FelderaClient, PipelineBuilder

client = FelderaClient('http://127.0.0.1:8080', api_key=api_key)

pipeline = PipelineBuilder(client, name, sql).create()

TLS with Self Signed Certificates

To use TLS with self signed certificates, set to the path of the CA bundle or the directory that contains the CA certificates.

from feldera import FelderaClient

client = FelderaClient('https://cluster.feldera.com', verify='path/to/cert')

Setting HTTP Connection Timeouts

To set a timeout for the HTTP connection, pass the timeout parameter to the .class:FelderaClient constructor. If the Feldera backend server takes longer than the specified timeout to respond, a .class:FelderaTimeoutError exception will be raised. This example sets the timeout for each HTTP request to 10 seconds.

from feldera import FelderaClient, PipelineBuilder

client = FelderaClient("http://127.0.0.1:8080", api_key=api_key, timeout=10)

Note

This is for an individual HTTP request, and does not affect things like waiting for a pipeline to start, pause, resume and stop. To set a timeout for these state transitions, set the parameter timeout_s in respective functions.

Creating a Pipeline (OVERWRITING existing pipelines)

sql = """
CREATE TABLE student (
    name STRING,
    id INT
);

CREATE TABLE grades (
    student_id INT,
    science INT,
    maths INT,
    art INT
);

CREATE VIEW average_scores AS SELECT name, ((science + maths + art) / 3) as average FROM {TBL_NAMES[0]} JOIN {TBL_NAMES[1]} on id = student_id ORDER BY average DESC;
"""

# This will stop and overwrite any existing pipeline with the same name.
pipeline = PipelineBuilder(client, name="notebook", sql=sql).create_or_replace()

Creating a Pipeline with Fault Tolerance Enabled

from feldera.runtime_config import RuntimeConfig
from feldera.enums import FaultToleranceModel

client = FelderaClient.localhost()
runtime_config = RuntimeConfig(
    fault_tolerance_model=FaultToleranceModel.AtLeastOnce,
    checkpoint_interval_secs=60
)

pipeline = PipelineBuilder(client, name, sql, runtime_config=runtime_config).create()

Runtime configuration of a Pipeline

from feldera.runtime_config import RuntimeConfig

client = FelderaClient.localhost()
config = {
    "workers": 8,
    "storage": {
        "backend": {
            "name": "default"
        },
        "min_storage_bytes": None,
        "min_step_storage_bytes": None,
        "compression": "default",
        "cache_mib": None
    },
    "fault_tolerance": {
        "model": "at_least_once",
        "checkpoint_interval_secs": 60
    },
    "cpu_profiler": True,
    "tracing": False,
    "tracing_endpoint_jaeger": "",
    "min_batch_size_records": 0,
    "max_buffering_delay_usecs": 0,
    "resources": {
        "cpu_cores_min": None,
        "cpu_cores_max": None,
        "memory_mb_min": None,
        "memory_mb_max": None,
        "storage_mb_max": None,
        "storage_class": None
    },
    "clock_resolution_usecs": 1_000_000,
    "pin_cpus": [],
    "provisioning_timeout_secs": None,
    "max_parallel_connector_init": None,
    "init_containers": None,
    "checkpoint_during_suspend": True,
    "dev_tweaks": {}
}

runtime_config = RuntimeConfig.from_dict(config)

pipeline = PipelineBuilder(client, name, sql, runtime_config=runtime_config).create()

Starting a Pipeline

pipeline.start()

Analyzing Existing Feldera Pipeline for Errors

First let's create a Feldera pipeline that errors from the web console, with the name check_error and invalid SQL as follows:

SELECT invalid

This will fail to compile.

We can use this Python SDK to connect to this Feldera pipeline to check if it has any errors as follows:

pipeline = Pipeline.get("check_error", client)
err = pipeline.errors()

if len(err) != 0:
    print("got err: ", err)

Here, err is a list of all errors in this pipeline. The above code will emit the following output:

got err:  [{'sql_compilation': {'exit_code': 1, 'messages': [{'start_line_number': 1, 'start_column': 1, 'end_line_number': 1, 'end_column': 14, 'warning': False, 'error_type': 'Not supported', 'message': 'Raw \'SELECT\' statements are not supported; did you forget to CREATE VIEW?: SELECT "invalid"', 'snippet': '    1|SELECT invalid\n      ^^^^^^^^^^^^^^\n'}]}}]

Using Pandas DataFrames

# populate pandas dataframes
df_students = pd.read_csv('students.csv')
df_grades = pd.read_csv('grades.csv')

pipeline.start()

# subscribe to listen to outputs from a view
out = pipeline.listen("average_scores")

# feed pandas dataframes as input
pipeline.input_pandas("students", df_students)
pipeline.input_pandas("grades", df_grades)

# wait for the pipeline to complete and stop
pipeline.wait_for_completion(True)

# get the output of the view as a pandas dataframe
df = out.to_pandas()

# clear the storage and delete the pipeline
pipeline.delete(True)

Using Completion Tokens

Completion tokens can be used to check whether all inputs ingested before the token was issued have been fully processed, and the resulting outputs have been written to all data sinks.

The following methods automatically use completion tokens:

To generate a completion token for a connector attached to the pipeline use :meth:`.Pipeline.generate_completion_token`. To check the status of this completion token use :meth:`.Pipeline.completion_token_status`.

# generate completion token for table "t0" and connector "myconnector"
token = pipeline.generate_completion_token("t0", "myconnector")

# check the status of this completion token
status = pipeline.completion_token_status(token)
print(status)

# wait until the pipeline processes this completion token
pipeline.wait_for_token(token)

Executing ad-hoc SQL Queries

Ad-hoc SQL queries can be executed on running or paused pipelines. Ad-hoc queries provide a way to query the state of materialized views or tables.

For more information, refer to the docs at: https://docs.feldera.com/sql/ad-hoc

We provide the following methods to execute ad-hoc queries:

  1. :meth:`.Pipeline.execute` - Execute an ad-hoc query and discard the result. Useful for INSERT queries.
  2. :meth:`.Pipeline.query` (Lazy) - Executes an ad-hoc query and returns a generator to iterate over the result.
  3. :meth:`.Pipeline.query_tabular` (Lazy) - Executes an ad-hoc query and returns a generator that yields a string representing the query result in human-readable tabular format.
  4. :meth:`.Pipeline.query_parquet` - Executes an ad-hoc query and saves the result to the specified path as a parquet file.
# execute an `INSERT` ad-hoc SQL query
pipeline.execute("INSERT into students VALUES ('John', 1)")

# executing a `SELECT` ad-hoc SQL query
students = list(pipeline.query("SELECT * FROM students"))

Iterating over Output Chunks

Use :meth:`.foreach_chunk` to process each chunk of data from a view or table. It takes a callback, and calls the callback on each chunk of received data.

# define your callback to run on every chunk of data received
# ensure that it takes two parameters, the chunk (DataFrame) and the sequence number
def callback(df: pd.DataFrame, seq_no: int):
    print(f"\nSeq No: {seq_no}, DF size: {df.shape[0]}\n")

pipeline = PipelineBuilder(client, name="notebook", sql=sql).create_or_replace()

# run the pipeline
pipeline.start()

# register the callback for data received from the selected view
pipeline.foreach_chunk("view_name", callback)

pipeline.input_pandas("table_name", df)

# wait for the pipeline to finish and stop
pipeline.wait_for_completion(True)

# clear the storage and delete the pipeline
pipeline.delete(True)

Waiting for Completion

To wait (block) till the pipeline has been completed, use :meth:`.Pipeline.wait_for_completion`.

pipeline.wait_for_completion()

Optionally, to forcibly stop (without checkpointing) the pipeline after completion:

pipeline.wait_for_completion(force_stop=True)

Warning

If the data source is streaming, this will block forever.

End-to-End Example with Kafka Sink

This example shows creating and running a pipeline with Feldera's internal data generator and writing to a Kafka sink.

from feldera import FelderaClient, PipelineBuilder

client = FelderaClient('http://localhost:8080')

sql = """
        CREATE TABLE Stocks (
        symbol VARCHAR NOT NULL,
        price_time BIGINT NOT NULL,  -- UNIX timestamp
        price DECIMAL(38, 2) NOT NULL
        ) with (
          'connectors' = '[{
            "transport": {
              "name": "datagen",
              "config": {
                "plan": [{
                    "limit": 5,
                    "rate": 1,
                    "fields": {
                        "symbol": { "values": ["AAPL", "GOOGL", "SPY", "NVDA"] },
                        "price": { "strategy": "uniform", "range": [100, 10000] }
                    }
                }]
              }
            }
          }]'
        );

        CREATE VIEW googl_stocks
        WITH (
            'connectors' = '[
                {
                    "name": "kafka-3",
                    "transport": {
                        "name": "kafka_output",
                        "config": {
                            "bootstrap.servers": "localhost:9092",
                            "topic": "googl_stocks",
                            "auto.offset.reset": "earliest"
                        }
                    },
                    "format": {
                        "name": "json",
                        "config": {
                            "update_format": "insert_delete",
                            "array": false
                        }
                    }
                }
            ]'
        )
        AS SELECT * FROM Stocks WHERE symbol = 'GOOGL';
        """

pipeline = PipelineBuilder(client, name="kafka_example", sql=sql).create_or_replace()

# Start the pipeline in paused state, attach listener, then unpause the pipeline.
# This ensures that the listener gets all the output from the view.
pipeline.start_paused()
out = pipeline.listen("googl_stocks")
pipeline.resume()

# important: `wait_for_completion` will block forever here
pipeline.wait_for_idle()
pipeline.stop(force=True)
df = out.to_pandas()
assert df.shape[0] != 0

# clear the storage and delete the pipeline
pipeline.delete(True)

Retrieve a support-bundle for a pipeline

This example shows how to download a support bundle for a pipeline using the Python SDK.

# Create a client (assuming Feldera is running on localhost:8080)
client = FelderaClient.localhost(port=8080)

# Define a simple SQL program
sql_program = """
CREATE TABLE users(id INT, name STRING);
CREATE MATERIALIZED VIEW user_count AS SELECT COUNT(*) as count FROM users;
"""

# Create a pipeline
pipeline_name = "support-bundle-example"
pipeline = PipelineBuilder(
    client,
    pipeline_name,
    sql_program
).create_or_replace()

print(f"Created pipeline: {pipeline.name}")

# Start the pipeline
pipeline.start()
print("Pipeline started")

# Generate support bundle as bytes
print("Generating support bundle...")
support_bundle_bytes = pipeline.support_bundle()
print(f"Support bundle size: {len(support_bundle_bytes)} bytes")

# Verify it's a valid ZIP file
try:
    with zipfile.ZipFile(io.BytesIO(support_bundle_bytes), 'r') as zip_file:
        file_list = zip_file.namelist()
        print(f"Support bundle contains {len(file_list)} files:")
        for file_name in file_list[:5]:  # Show first 5 files
            print(f"  - {file_name}")
        if len(file_list) > 5:
            print(f"  ... and {len(file_list) - 5} more files")
except zipfile.BadZipFile:
    print("Warning: Support bundle is not a valid ZIP file")

# Save support bundle to a file
output_path = f"{pipeline_name}-support-bundle.zip"
pipeline.support_bundle(output_path=output_path)
print(f"Support bundle saved to: {output_path}")

# Verify the saved file
if os.path.exists(output_path):
    file_size = os.path.getsize(output_path)
    print(f"Saved file size: {file_size} bytes")

    # Clean up
    os.unlink(output_path)
    print("Cleaned up saved file")

# Stop the pipeline
pipeline.stop(force=True)
pipeline.clear_storage()
pipeline.delete()
print("Pipeline stopped and deleted")

Specifying Data Sources / Sinks

To connect Feldera to various data sources or sinks, you can define them in the SQL code. Refer to the connector documentation at: https://docs.feldera.com/connectors/

Benchmarking Pipelines

The :mod:`feldera.benchmarking` module provides utilities to collect and upload benchmark metrics for Feldera pipelines. It polls :meth:`.Pipeline.stats` in a loop, aggregates the snapshots into :class:`.BenchmarkMetrics`, and can optionally upload a Bencher Metric Format (BMF) report to a Bencher-compatible server.

Note

These utilities only observe a running pipeline — they do not start, stop, or otherwise manage pipeline lifetime. The caller is responsible for starting the pipeline before calling :func:`.bench` or :func:`.collect_metrics`, and for stopping it afterwards.

Collect and Display Metrics

Stop any existing run, start fresh, wait for all bounded input to be processed, then print a human-readable results table.

from feldera import FelderaClient, PipelineBuilder, bench

client = FelderaClient("http://localhost:8080")

sql = """
CREATE TABLE events (id INT, value DOUBLE) WITH (
    'connectors' = '[{
        "transport": {
            "name": "datagen",
            "config": {"plan": [{"limit": 1000000, "rate": 100000}]}
        }
    }]'
);
CREATE MATERIALIZED VIEW totals AS
    SELECT COUNT(*) AS n, SUM(value) AS total FROM events;
"""

pipeline = PipelineBuilder(client, name="my_bench", sql=sql).create_or_replace()

# Stop any running instance for a reproducible baseline, then start fresh.
pipeline.stop()
pipeline.start()

# Poll stats until pipeline_complete is True (all bounded input consumed).
result = bench(pipeline)

# Stop the pipeline now that collection is done.
pipeline.stop()

# Print the results table.
print(result.format_table())

# Or access the BMF dict directly.
print(result.to_json())

Collect Metrics for a Fixed Duration

For streaming pipelines whose input never ends naturally, pass duration_secs to stop collection after a fixed wall-clock window.

from feldera import FelderaClient, bench

client = FelderaClient("http://localhost:8080")
pipeline = client.get_pipeline("my_streaming_pipeline")

pipeline.stop()
pipeline.start()

# Collect for 60 seconds regardless of pipeline_complete.
result = bench(pipeline, duration_secs=60)

pipeline.stop()

print(result.format_table())

Aggregate Metrics Across Multiple Runs

Run the benchmark several times and combine the results with :meth:`.BenchmarkResult.aggregate`. The aggregated result averages throughput, uptime, and state-amplification across runs, takes the min-of-mins and max-of-maxes for memory and storage, and can be passed directly to :func:`.upload_to_bencher` just like a single-run result.

from feldera import FelderaClient, bench
from feldera.benchmarking import BenchmarkResult

client = FelderaClient("http://localhost:8080")
pipeline = client.get_pipeline("my_bench")

runs = []
for _ in range(3):
    pipeline.stop()
    pipeline.start()
    runs.append(bench(pipeline))

pipeline.stop()

result = BenchmarkResult.aggregate(runs)
print(result.format_table())   # shows avg with stddev %

Upload Results to Bencher

After collecting metrics, call :func:`.upload_to_bencher` to POST the BMF report to a Bencher-compatible server.

Passing feldera_client enriches the run context with the Feldera instance edition and revision.

API token and project can be supplied as parameters or via the BENCHER_API_TOKEN and BENCHER_PROJECT environment variables.

from feldera import FelderaClient, PipelineBuilder, bench, upload_to_bencher

client = FelderaClient("http://localhost:8080")

sql = """
CREATE TABLE events (id INT, value DOUBLE) WITH (
    'connectors' = '[{
        "transport": {
            "name": "datagen",
            "config": {"plan": [{"limit": 1000000, "rate": 100000}]}
        }
    }]'
);
CREATE MATERIALIZED VIEW totals AS
    SELECT COUNT(*) AS n, SUM(value) AS total FROM events;
"""

pipeline = PipelineBuilder(client, name="my_bench", sql=sql).create_or_replace()

pipeline.stop()
pipeline.start()
result = bench(pipeline)
pipeline.stop()

print(result.format_table())

# Upload to https://benchmarks.feldera.io (the default host).
upload_to_bencher(
    result,
    project="my-project",          # or set BENCHER_PROJECT env var
    token="YOUR_BENCHER_TOKEN",     # or set BENCHER_API_TOKEN env var
    branch="main",
    feldera_client=client,          # adds edition/revision to run context
)

Note

The host parameter (or BENCHER_HOST environment variable) can point to any Bencher-compatible server. It defaults to https://benchmarks.feldera.io.

Track Results Against a Baseline Branch

Use start_point to initialise a new branch from an existing one and optionally inherit its alert thresholds.

from feldera import FelderaClient, bench, upload_to_bencher

client = FelderaClient("http://localhost:8080")
pipeline = client.get_pipeline("my_bench")

pipeline.stop()
pipeline.start()
result = bench(pipeline)
pipeline.stop()

upload_to_bencher(
    result,
    project="my-project",
    token="YOUR_BENCHER_TOKEN",
    branch="feature/my-optimisation",  # the branch being tested
    start_point="main",                 # branch to branch off from
    start_point_clone_thresholds=True,  # inherit alert thresholds
    start_point_max_versions=10,        # how many historical runs to consider
    feldera_client=client,
)