Visitar URL original
feat: Add online store read and ODFV transformation duration metrics with per-ODFV opt-in by ntkathole · Pull Request #6125 · feast-dev/feast · GitHub
Skip to content

feat: Add online store read and ODFV transformation duration metrics with per-ODFV opt-in #6125

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
franciscojavierarceo merged 1 commit into from
Mar 24, 2026
Merged

feat: Add online store read and ODFV transformation duration metrics with per-ODFV opt-in #6125

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
feat: Added odfv transformations metrics 
Signed-off-by: ntkathole <nikhilkathole2683@gmail.com>
  • Loading branch information
@ntkathole
ntkathole committed Mar 24, 2026
commit e7eec5250c926908a326b5ebf87d099e39f21e5c
6 changes: 6 additions & 0 deletions docs/getting-started/components/open-telemetry.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,12 @@ Once configured, you can monitor various metrics including:
- `feast_feature_server_memory_usage`: Memory utilization of the feature server
- `feast_feature_server_cpu_usage`: CPU usage statistics
- `feast_feature_server_request_latency_seconds`: Request latency with feature count dimensions
- `feast_feature_server_online_store_read_duration_seconds`: Online store read phase duration
- `feast_feature_server_transformation_duration_seconds`: ODFV read-path transformation duration (per ODFV, requires `track_metrics=True`)
- `feast_feature_server_write_transformation_duration_seconds`: ODFV write-path transformation duration (per ODFV, requires `track_metrics=True`)
- Additional custom metrics based on your configuration

For the full list of metrics, see the [Python Feature Server reference](../../reference/feature-servers/python-feature-server.md#available-metrics).

These metrics can be visualized using Prometheus and other compatible monitoring tools.
56 changes: 44 additions & 12 deletions docs/reference/feature-servers/python-feature-server.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -360,18 +360,50 @@ thread from starting). All categories default to `true`.

### Available metrics

| Metric | Type | Labels | Description |
|--------|------|--------|-------------|
| `feast_feature_server_cpu_usage` | Gauge | — | Process CPU usage % |
| `feast_feature_server_memory_usage` | Gauge | — | Process memory usage % |
| `feast_feature_server_request_total` | Counter | `endpoint`, `status` | Total requests per endpoint |
| `feast_feature_server_request_latency_seconds` | Histogram | `endpoint`, `feature_count`, `feature_view_count` | Request latency with p50/p95/p99 support |
| `feast_online_features_request_total` | Counter | — | Total online feature retrieval requests |
| `feast_online_features_entity_count` | Histogram | — | Entity rows per online feature request |
| `feast_push_request_total` | Counter | `push_source`, `mode` | Push requests by source and mode |
| `feast_materialization_total` | Counter | `feature_view`, `status` | Materialization runs (success/failure) |
| `feast_materialization_duration_seconds` | Histogram | `feature_view` | Materialization duration per feature view |
| `feast_feature_freshness_seconds` | Gauge | `feature_view`, `project` | Seconds since last materialization |
| Metric | Type | Labels | Category | Description |
|--------|------|--------|----------|-------------|
| `feast_feature_server_cpu_usage` | Gauge | — | `resource` | Process CPU usage % |
| `feast_feature_server_memory_usage` | Gauge | — | `resource` | Process memory usage % |
| `feast_feature_server_request_total` | Counter | `endpoint`, `status` | `request` | Total requests per endpoint |
| `feast_feature_server_request_latency_seconds` | Histogram | `endpoint`, `feature_count`, `feature_view_count` | `request` | Request latency with p50/p95/p99 support |
| `feast_online_features_request_total` | Counter | — | `online_features` | Total online feature retrieval requests |
| `feast_online_features_entity_count` | Histogram | — | `online_features` | Entity rows per online feature request |
| `feast_feature_server_online_store_read_duration_seconds` | Histogram | — | `online_features` | Online store read phase duration (sync and async) |
| `feast_feature_server_transformation_duration_seconds` | Histogram | `odfv_name`, `mode` | `online_features` | ODFV read-path transformation duration (requires `track_metrics=True` on the ODFV) |
| `feast_feature_server_write_transformation_duration_seconds` | Histogram | `odfv_name`, `mode` | `online_features` | ODFV write-path transformation duration (requires `track_metrics=True` on the ODFV) |
| `feast_push_request_total` | Counter | `push_source`, `mode` | `push` | Push requests by source and mode |
| `feast_materialization_result_total` | Counter | `feature_view`, `status` | `materialization` | Materialization runs (success/failure) |
| `feast_materialization_duration_seconds` | Histogram | `feature_view` | `materialization` | Materialization duration per feature view |
| `feast_feature_freshness_seconds` | Gauge | `feature_view`, `project` | `freshness` | Seconds since last materialization |

### Per-ODFV transformation metrics

The `transformation_duration_seconds` and `write_transformation_duration_seconds`
metrics are gated behind **two** conditions — both must be true for any
instrumentation to run:

1. **Server-level**: the `online_features` category must be enabled in the
metrics configuration.
2. **ODFV-level**: the `OnDemandFeatureView` must have `track_metrics=True`.

This defaults to `False`, so no ODFV incurs timing overhead unless explicitly
opted in:

```python
from feast.on_demand_feature_view import on_demand_feature_view
@on_demand_feature_view(
sources=[my_feature_view, my_request_source],
schema=[Field(name="output", dtype=Float64)],
track_metrics=True, # opt in to transformation timing
)
def my_transform(inputs: pd.DataFrame) -> pd.DataFrame:
...
```

The `odfv_name` label lets you filter or group by individual ODFV,
and the `mode` label (`python`, `pandas`, `substrait`) lets you compare
transformation engines.

### Scraping with Prometheus

Expand Down
2 changes: 1 addition & 1 deletion docs/reference/feature-store-yaml.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ feature_server:
enabled: true # Enable Prometheus metrics server on port 8000
resource: true # CPU / memory gauges
request: true # endpoint latency histograms & request counters
online_features: true # online feature retrieval counters
online_features: true # online feature retrieval counters + store read & ODFV transform timing
push: true # push request counters
materialization: true # materialization counters & duration histograms
freshness: true # per-feature-view freshness gauges
Expand Down
148 changes: 88 additions & 60 deletions sdk/python/feast/feature_store.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2097,76 +2097,104 @@ def _transform_on_demand_feature_view_df(
Raises:
Exception: For unsupported OnDemandFeatureView modes
"""
if feature_view.mode == "python" and isinstance(
feature_view.feature_transformation, PythonTransformation
):
input_dict = (
df.to_dict(orient="records")[0]
if feature_view.singleton
else df.to_dict(orient="list")
_should_track = False
try:
from feast.metrics import _config as _metrics_config

_should_track = _metrics_config.online_features and getattr(
feature_view, "track_metrics", False
)
except Exception:
pass

if feature_view.singleton:
transformed_rows = []
if _should_track:
import time as _time

for i, row in df.iterrows():
output = feature_view.feature_transformation.udf(row.to_dict())
if i == 0:
transformed_rows = output
else:
for k in output:
if isinstance(output[k], list):
transformed_rows[k].extend(output[k])
else:
transformed_rows[k].append(output[k])

transformed_data = pd.DataFrame(transformed_rows)
else:
transformed_data = feature_view.feature_transformation.udf(input_dict)
_t0 = _time.monotonic()

if feature_view.write_to_online_store:
entities = [
self.get_entity(entity) for entity in (feature_view.entities or [])
]
join_keys = [entity.join_key for entity in entities if entity]
join_keys = [k for k in join_keys if k in input_dict.keys()]
transformed_df = (
pd.DataFrame(transformed_data)
if not isinstance(transformed_data, pd.DataFrame)
else transformed_data
)
input_df = pd.DataFrame(
[input_dict] if feature_view.singleton else input_dict
try:
if feature_view.mode == "python" and isinstance(
feature_view.feature_transformation, PythonTransformation
):
input_dict = (
df.to_dict(orient="records")[0]
if feature_view.singleton
else df.to_dict(orient="list")
)
if input_df.shape[0] == transformed_df.shape[0]:

if feature_view.singleton:
transformed_rows = []

for i, row in df.iterrows():
output = feature_view.feature_transformation.udf(row.to_dict())
if i == 0:
transformed_rows = output
else:
for k in output:
if isinstance(output[k], list):
transformed_rows[k].extend(output[k])
else:
transformed_rows[k].append(output[k])

transformed_data = pd.DataFrame(transformed_rows)
else:
transformed_data = feature_view.feature_transformation.udf(
input_dict
)

if feature_view.write_to_online_store:
entities = [
self.get_entity(entity)
for entity in (feature_view.entities or [])
]
join_keys = [entity.join_key for entity in entities if entity]
join_keys = [k for k in join_keys if k in input_dict.keys()]
transformed_df = (
pd.DataFrame(transformed_data)
if not isinstance(transformed_data, pd.DataFrame)
else transformed_data
)
input_df = pd.DataFrame(
[input_dict] if feature_view.singleton else input_dict
)
if input_df.shape[0] == transformed_df.shape[0]:
for k in input_dict:
if k not in transformed_data:
transformed_data[k] = input_dict[k]
transformed_df = pd.DataFrame(transformed_data)
else:
transformed_df = pd.merge(
transformed_df,
input_df,
how="left",
on=join_keys,
)
else:
# overwrite any transformed features and update the dictionary
for k in input_dict:
if k not in transformed_data:
transformed_data[k] = input_dict[k]
transformed_df = pd.DataFrame(transformed_data)
else:
transformed_df = pd.merge(
transformed_df,
input_df,
how="left",
on=join_keys,
)
else:
# overwrite any transformed features and update the dictionary
for k in input_dict:
if k not in transformed_data:
transformed_data[k] = input_dict[k]

return pd.DataFrame(transformed_data)
return pd.DataFrame(transformed_data)

elif feature_view.mode == "pandas" and isinstance(
feature_view.feature_transformation, PandasTransformation
):
transformed_df = feature_view.feature_transformation.udf(df)
for col in df.columns:
transformed_df[col] = df[col]
return transformed_df
else:
raise Exception("Unsupported OnDemandFeatureView mode")
elif feature_view.mode == "pandas" and isinstance(
feature_view.feature_transformation, PandasTransformation
):
transformed_df = feature_view.feature_transformation.udf(df)
for col in df.columns:
transformed_df[col] = df[col]
return transformed_df
else:
raise Exception("Unsupported OnDemandFeatureView mode")
finally:
if _should_track:
from feast.metrics import track_write_transformation

track_write_transformation(
feature_view.name,
feature_view.mode,
_time.monotonic() - _t0,
)

def _validate_vector_features(self, feature_view, df: pd.DataFrame) -> None:
"""
Expand Down
9 changes: 7 additions & 2 deletions sdk/python/feast/infra/feature_servers/base_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,15 +62,20 @@ class MetricsConfig(FeastConfigBaseModel):
online_features: StrictBool = True
"""Emit online feature retrieval metrics
(feast_online_features_request_total,
feast_online_features_entity_count)."""
feast_online_features_entity_count,
feast_feature_server_online_store_read_duration_seconds,
feast_feature_server_transformation_duration_seconds,
feast_feature_server_write_transformation_duration_seconds).
ODFV transformation metrics additionally require track_metrics=True
on the OnDemandFeatureView definition."""

push: StrictBool = True
"""Emit push/write request counters
(feast_push_request_total)."""

materialization: StrictBool = True
"""Emit materialization success/failure counters and duration histograms
(feast_materialization_total,
(feast_materialization_result_total,
feast_materialization_duration_seconds)."""

freshness: StrictBool = True
Expand Down
36 changes: 36 additions & 0 deletions sdk/python/feast/infra/online_stores/online_store.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,6 +185,19 @@ def get_online_features(
native_entity_values=True,
)

_track_read = False
try:
from feast.metrics import _config as _metrics_config

_track_read = _metrics_config.online_features
except Exception:
pass
Comment thread
ntkathole marked this conversation as resolved.

if _track_read:
import time as _time

_read_start = _time.monotonic()

for table, requested_features in grouped_refs:
# Get the correct set of entity values with the correct join keys.
table_entity_values, idxs, output_len = utils._get_unique_entities(
Expand Down Expand Up @@ -218,6 +231,11 @@ def get_online_features(
output_len,
)

if _track_read:
from feast.metrics import track_online_store_read

track_online_store_read(_time.monotonic() - _read_start)
Comment thread
devin-ai-integration[bot] marked this conversation as resolved.

if requested_on_demand_feature_views:
utils._augment_response_with_on_demand_transforms(
online_features_response,
Expand Down Expand Up @@ -293,6 +311,19 @@ async def query_table(table, requested_features):

return idxs, read_rows, output_len

_track_read = False
try:
from feast.metrics import _config as _metrics_config

_track_read = _metrics_config.online_features
except Exception:
pass

if _track_read:
import time as _time

_read_start = _time.monotonic()

all_responses = await asyncio.gather(
*[
query_table(table, requested_features)
Expand All @@ -318,6 +349,11 @@ async def query_table(table, requested_features):
output_len,
)

if _track_read:
from feast.metrics import track_online_store_read

track_online_store_read(_time.monotonic() - _read_start)

if requested_on_demand_feature_views:
utils._augment_response_with_on_demand_transforms(
online_features_response,
Expand Down
Loading
Loading