detectkit 0.62.0__tar.gz → 0.63.0__tar.gz

{detectkit-0.62.0/detectkit.egg-info → detectkit-0.63.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: detectkit
-Version: 0.62.0
+Version: 0.63.0
 Summary: Metric monitoring with automatic anomaly detection
 Author: detectkit team
 License: MIT
@@ -110,7 +110,7 @@ Dynamic: license-file
 - **Hybrid mode** — read metric SQL from a billed-per-query warehouse while all pipeline state lives in a cheap local database
 - **Idempotent** — resume from interruptions, no duplicate processing
 - **CLI** — `dtk init`, `dtk run --select`, `dtk unlock`, `dtk clean`, tag-based selectors
-- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + three skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
+- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + five skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
 
 ## Installation
 
@@ -122,6 +122,7 @@ With database drivers:
 
 ```bash
 pip install detectkit[clickhouse]   # ClickHouse
+pip install detectkit[duckdb]       # DuckDB - single file, no server
 pip install detectkit[all-db]       # All databases
 ```
 

{detectkit-0.62.0 → detectkit-0.63.0}/README.md

@@ -28,7 +28,7 @@
 - **Hybrid mode** — read metric SQL from a billed-per-query warehouse while all pipeline state lives in a cheap local database
 - **Idempotent** — resume from interruptions, no duplicate processing
 - **CLI** — `dtk init`, `dtk run --select`, `dtk unlock`, `dtk clean`, tag-based selectors
-- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + three skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
+- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + five skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
 
 ## Installation
 
@@ -40,6 +40,7 @@ With database drivers:
 
 ```bash
 pip install detectkit[clickhouse]   # ClickHouse
+pip install detectkit[duckdb]       # DuckDB - single file, no server
 pip install detectkit[all-db]       # All databases
 ```
 

{detectkit-0.62.0 → detectkit-0.63.0}/detectkit/__init__.py

@@ -4,7 +4,7 @@ detectk - Anomaly Detection for Time-Series Metrics
 A Python library for data analysts and engineers to monitor metrics with automatic anomaly detection.
 """
 
-__version__ = "0.62.0"
+__version__ = "0.63.0"
 
 from detectkit.core.interval import Interval
 from detectkit.core.models import ColumnDefinition, TableModel

{detectkit-0.62.0 → detectkit-0.63.0}/detectkit/cli/assets/claude/rules/cli.md

@@ -458,6 +458,13 @@ fail their step when `dtk run` / `dtk autotune` / `dtk clean` exits non-zero
 (see "Exit codes" above), and can additionally parse `dtk run --json`'s stdout
 summary for per-metric detail.
 
+On GitHub Actions specifically, detectkit ships a ready-made composite action
+instead of hand-rolling those steps: `uses: alexeiveselov92/detectkit@vX.Y.Z`
+installs detectkit from PyPI and runs `dtk run` / `dtk autotune` / `dtk clean`
+in a given project directory. It surfaces the 0/1/2 exit-code contract (see
+"Exit codes" above) as the job's own outcome, and exposes the
+`dtk run --json` summary as an action output for downstream steps to consume.
+
 ## Troubleshooting
 
 - **"Metric not found"** — selector doesn't match. Use the bare name

{detectkit-0.62.0 → detectkit-0.63.0}/detectkit/cli/assets/claude/rules/metrics.md

@@ -119,6 +119,14 @@ The query must return a **timestamp** column and a numeric **value** column
 (default names `timestamp` / `value`; remap via `query_columns`). It may also
 return seasonality columns (declare them in `query_columns.seasonality`).
 
+Returned timestamps must already be **bucketed to the interval grid**
+(`toStartOfInterval` / `time_bucket` / `date_trunc`) — gap filling maps rows
+onto the grid by exact match, and the grid's phase comes from
+`loading_start_time`. A batch that returns rows on a different phase (e.g.
+`:28` past the hour vs a grid phased at `:00`) can land 100% off-grid and
+load as silently all-`NULL`; detectkit logs a one-time warning naming both
+phases when that happens.
+
 ## Seasonality features
 
 Two ways to provide seasonality keys that detectors group by:

{detectkit-0.62.0 → detectkit-0.63.0}/detectkit/cli/assets/claude/rules/overview.md

@@ -4,8 +4,9 @@ detectkit is a Python library and CLI (`dtk`) for monitoring time-series
 metrics with automatic anomaly detection and multi-channel alerting. It is
 **dbt-like**: metrics live as YAML + SQL in a project directory, and you run
 them with one command. Core logic is pure numpy (no pandas). **ClickHouse,
-PostgreSQL and MySQL are all fully supported** — only the connection and the SQL
-dialect of your metric queries differ between them.
+PostgreSQL, MySQL/MariaDB and DuckDB are all fully supported** — only the
+connection and the SQL dialect of your metric queries differ between them
+(DuckDB additionally needs no server at all — just a local file).
 
 ## The pipeline: load → detect → alert
 

{detectkit-0.62.0 → detectkit-0.63.0}/detectkit/loaders/metric_loader.py

@@ -9,6 +9,7 @@ Loads time-series data from databases with:
 - Integration with InternalTablesManager
 """
 
+import logging
 from datetime import datetime, timedelta
 
 import numpy as np
@@ -21,6 +22,30 @@ from detectkit.loaders.query_template import QueryTemplate
 from detectkit.utils.datetime_utils import now_utc_naive, to_naive_utc
 from detectkit.utils.json_utils import json_dumps_sorted
 
+logger = logging.getLogger(__name__)
+
+
+def _format_offset_seconds(seconds: int) -> str:
+    """Render an offset-within-interval (seconds) as a human clock offset.
+
+    Used to describe a grid/source "phase" without assuming a particular
+    interval unit: ``0`` -> ``":00"``, ``1680`` -> ``":28"`` (28 minutes),
+    ``90`` -> ``":01:30"``, ``3661`` -> ``"1:01:01"``.
+    """
+    hours, remainder = divmod(int(seconds), 3600)
+    minutes, secs = divmod(remainder, 60)
+    if hours:
+        return f"{hours}:{minutes:02d}:{secs:02d}"
+    if secs:
+        return f":{minutes:02d}:{secs:02d}"
+    return f":{minutes:02d}"
+
+
+def _format_datetime64(ts: np.datetime64) -> str:
+    """Render a datetime64 as ``YYYY-MM-DD HH:MM:SS`` (naive UTC convention)."""
+    dt: datetime = ts.astype("datetime64[s]").astype(datetime)
+    return dt.strftime("%Y-%m-%d %H:%M:%S")
+
 
 class MetricLoader:
     """
@@ -76,6 +101,9 @@ class MetricLoader:
         self.internal_manager = internal_manager
         self.source_profile_name = source_profile_name
         self.query_template = QueryTemplate()
+        # Emitted at most once per loader instance (one instance per metric
+        # per run) - see _check_grid_alignment.
+        self._warned_grid_misalignment = False
 
     def load(
         self,
@@ -246,6 +274,7 @@ class MetricLoader:
             timestamp_array, value_array = self._fill_gaps(
                 timestamp_array, value_array, from_date, to_date, interval_seconds
             )
+            self._check_grid_alignment(original_timestamps, timestamp_array, interval_seconds)
 
             # Realign query-provided seasonality to the gap-filled grid.
             # Gap rows can appear ANYWHERE in the range and _fill_gaps also
@@ -423,6 +452,60 @@ class MetricLoader:
 
         return full_timestamps, filled_values
 
+    def _check_grid_alignment(
+        self,
+        original_timestamps: np.ndarray,
+        grid_timestamps: np.ndarray,
+        interval_seconds: int,
+    ) -> None:
+        """Warn once if a non-empty query result landed entirely off the grid.
+
+        ``_fill_gaps`` maps source rows onto the gap-filled grid by exact
+        timestamp match. If the query returns rows bucketed to a different
+        grid "phase" than ``loading_start_time`` establishes (e.g. rows at
+        :28 past the hour against a 10min grid phased at :00), NONE of them
+        match and the batch loads as silently 100% NULL - no error, no
+        partial data, just an empty-looking series. Partial alignment is not
+        the trap this guards against (some rows landing is normal, e.g. a
+        source with occasional off-grid duplicates), so this only fires when
+        the overlap is exactly zero. Observability only - does not change
+        what gets loaded.
+        """
+        if self._warned_grid_misalignment:
+            return
+        if len(original_timestamps) == 0 or len(grid_timestamps) == 0:
+            return
+        if np.any(np.isin(original_timestamps, grid_timestamps)):
+            return
+
+        self._warned_grid_misalignment = True
+
+        grid_phase_seconds = int(
+            grid_timestamps[0].astype("datetime64[s]").astype(np.int64) % interval_seconds
+        )
+        source_offsets = (
+            original_timestamps.astype("datetime64[s]").astype(np.int64) % interval_seconds
+        )
+        offsets, counts = np.unique(source_offsets, return_counts=True)
+        source_phase_seconds = int(offsets[int(np.argmax(counts))])
+
+        sample_ts = original_timestamps[0]
+        nearest_ts = grid_timestamps[int(np.argmin(np.abs(grid_timestamps - sample_ts)))]
+
+        logger.warning(
+            "%s: query returned %d row(s) but none align with the metric's %s grid "
+            "(grid phase %s, source phase %s). e.g. source row %s vs nearest grid "
+            "slot %s. Bucket the query's timestamps (toStartOfInterval / time_bucket "
+            "/ date_trunc) or set loading_start_time to match the source phase.",
+            self.config.name,
+            len(original_timestamps),
+            self.config.interval,
+            _format_offset_seconds(grid_phase_seconds),
+            _format_offset_seconds(source_phase_seconds),
+            _format_datetime64(sample_ts),
+            _format_datetime64(nearest_ts),
+        )
+
     def _extract_seasonality(
         self,
         timestamps: np.ndarray,

{detectkit-0.62.0 → detectkit-0.63.0/detectkit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: detectkit
-Version: 0.62.0
+Version: 0.63.0
 Summary: Metric monitoring with automatic anomaly detection
 Author: detectkit team
 License: MIT
@@ -110,7 +110,7 @@ Dynamic: license-file
 - **Hybrid mode** — read metric SQL from a billed-per-query warehouse while all pipeline state lives in a cheap local database
 - **Idempotent** — resume from interruptions, no duplicate processing
 - **CLI** — `dtk init`, `dtk run --select`, `dtk unlock`, `dtk clean`, tag-based selectors
-- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + three skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
+- **AI-native onboarding** — `dtk init-claude` sets up Claude Code context (CLAUDE.md + rules + five skills) so an assistant can scaffold metrics, configure databases, and file feedback upstream
 
 ## Installation
 
@@ -122,6 +122,7 @@ With database drivers:
 
 ```bash
 pip install detectkit[clickhouse]   # ClickHouse
+pip install detectkit[duckdb]       # DuckDB - single file, no server
 pip install detectkit[all-db]       # All databases
 ```