126 lines
4.7 KiB
Python
126 lines
4.7 KiB
Python
from datetime import datetime
|
|
from pathlib import Path
|
|
from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple
|
|
|
|
import pandas as pd
|
|
from filelock import FileLock
|
|
from tinydb import TinyDB
|
|
|
|
from ..helper.normalise_datetime import normalise_datetime
|
|
from ..views import Filter, SortBy, Trace
|
|
from .tracing_database_driver import TracingDatabaseDriver
|
|
|
|
DEFAULT_TRACING_DB_FILENAME = "tracing_database.json"
|
|
|
|
|
|
operator_mapping = {"=": "eq", "!=": "ne", "<": "lt", "<=": "le", ">": "gt", ">=": "ge"}
|
|
|
|
|
|
class ParallelTinyDbDriver(TracingDatabaseDriver):
|
|
"""TracingDatabaseDriver with TinyDB as a backend.
|
|
|
|
Saves the database as a JSON into a single file. Highly inefficient on inserting,
|
|
not advised for production use.
|
|
|
|
A file lock protects the database across threads and independently started processes.
|
|
"""
|
|
|
|
is_production_ready = False
|
|
path_to_db = Path(DEFAULT_TRACING_DB_FILENAME)
|
|
|
|
def save(self, trace: Trace) -> str:
|
|
return self._safe_execute(lambda db: db.insert(trace.model_dump(mode="json")))
|
|
|
|
def save_batch(self, documents: List[Trace]) -> List[str]:
|
|
traces = [d.model_dump(mode="json") for d in documents]
|
|
return self._safe_execute(lambda db: db.insert_multiple(traces))
|
|
|
|
def get(self, id: str) -> Optional[Trace]:
|
|
value = self._safe_execute(lambda db: db.get(lambda d: d["trace_id"] == id))
|
|
if value:
|
|
value = Trace.model_validate(value)
|
|
return value
|
|
|
|
def query(
|
|
self,
|
|
*,
|
|
skip: int = 0,
|
|
take: Optional[int] = None,
|
|
conjunctive_filters: Sequence[Filter] = [],
|
|
conjunctive_tags: Sequence[str] = [],
|
|
since: Optional[datetime] = None,
|
|
until: Optional[datetime] = None,
|
|
has_feedback: Optional[bool] = None,
|
|
sort_by: Sequence[SortBy] = [],
|
|
) -> Tuple[List[Trace], int]:
|
|
since = normalise_datetime(since) if since is not None else None
|
|
until = normalise_datetime(until) if until is not None else None
|
|
|
|
def does_match(d: Dict[str, Any]) -> bool:
|
|
created = normalise_datetime(datetime.fromisoformat(d["created"]))
|
|
return (
|
|
not set(conjunctive_tags) - set(d["tags"])
|
|
and (since is None or created >= since)
|
|
and (until is None or created <= until)
|
|
and (
|
|
has_feedback is None or has_feedback == (d["feedback"] is not None)
|
|
)
|
|
)
|
|
|
|
documents = self._safe_execute(lambda db: db.search(does_match))
|
|
if not documents:
|
|
return [], 0
|
|
|
|
traces: List[Trace] = [Trace.model_validate(d) for d in documents]
|
|
# The DataFrame keeps its default 0..n-1 index, so filtering/sorting below
|
|
# preserve the row labels that index back into `traces`. This avoids
|
|
# reconstructing Traces from the lossy flattened rows.
|
|
df = pd.DataFrame([t.to_flat_dict() for t in traces])
|
|
|
|
for f in conjunctive_filters:
|
|
operator = f.operator.lower()
|
|
if operator in operator_mapping:
|
|
df = df.loc[
|
|
getattr(df[f.property], operator_mapping[f.operator])(f.value)
|
|
]
|
|
elif operator == "contains":
|
|
pattern = f.value_as_text.casefold()
|
|
df = df.loc[
|
|
df[f.property].map(
|
|
lambda value: (
|
|
pattern in value.casefold()
|
|
if isinstance(value, str)
|
|
else not isinstance(f.value, str) and value == f.value
|
|
)
|
|
)
|
|
]
|
|
|
|
if sort_by:
|
|
df.sort_values(
|
|
[col.column_id for col in sort_by],
|
|
ascending=[col.direction == "asc" for col in sort_by],
|
|
inplace=True,
|
|
)
|
|
|
|
count = len(df)
|
|
result = df.iloc[skip:] if take is None else df.iloc[skip : skip + take]
|
|
return [traces[i] for i in result.index], count
|
|
|
|
def update(self, id: str, new_version: Trace) -> None:
|
|
self._safe_execute(
|
|
lambda db: db.update(
|
|
new_version.model_dump(mode="json"), lambda d: d["trace_id"] == id
|
|
)
|
|
)
|
|
|
|
def delete(self, id: str) -> None:
|
|
self._safe_execute(lambda db: db.remove(lambda d: d["trace_id"] == id))
|
|
|
|
def delete_batch(self, ids: List[str]) -> None:
|
|
selected = set(ids)
|
|
self._safe_execute(lambda db: db.remove(lambda d: d["trace_id"] in selected))
|
|
|
|
def _safe_execute(self, func: Callable[[TinyDB], Any]) -> Any:
|
|
with FileLock(str(self.path_to_db.resolve()) + ".lock"):
|
|
with TinyDB(self.path_to_db) as db:
|
|
return func(db)
|