codebook / potato /data_sources /partial_reader.py
davidjurgens's picture
Deploy: Potato — Codebook Annotation
aceb1b2 verified
Raw
History Blame Contribute Delete
12.7 kB
"""
Partial/incremental loading for data sources.
This module provides functionality for loading data in chunks, enabling:
- Initial loading of first K items
- Batch loading of additional items as annotation progresses
- Auto-loading when annotation reaches a threshold
- State persistence for resumption after restart
"""
import json
import logging
import os
import threading
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Dict, List, Optional, TYPE_CHECKING
if TYPE_CHECKING:
from potato.data_sources.base import DataSource
logger = logging.getLogger(__name__)
@dataclass
class PartialReadState:
"""
Tracks the read state for a data source.
This dataclass maintains information about how much data has been
read from a source, enabling incremental loading.
Attributes:
source_id: Identifier of the data source
items_loaded: Number of items loaded so far
total_estimate: Estimated total items (None if unknown)
file_position: Byte position for file-based sources
line_number: Line number for line-based sources
is_complete: Whether all data has been loaded
last_loaded_at: Unix timestamp of last load
metadata: Additional source-specific state
"""
source_id: str
items_loaded: int = 0
total_estimate: Optional[int] = None
file_position: int = 0
line_number: int = 0
is_complete: bool = False
last_loaded_at: Optional[float] = None
metadata: Dict[str, Any] = field(default_factory=dict)
def to_dict(self) -> Dict[str, Any]:
"""Convert to dictionary for JSON serialization."""
return {
"source_id": self.source_id,
"items_loaded": self.items_loaded,
"total_estimate": self.total_estimate,
"file_position": self.file_position,
"line_number": self.line_number,
"is_complete": self.is_complete,
"last_loaded_at": self.last_loaded_at,
"metadata": self.metadata,
}
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "PartialReadState":
"""Create from dictionary."""
return cls(
source_id=data["source_id"],
items_loaded=data.get("items_loaded", 0),
total_estimate=data.get("total_estimate"),
file_position=data.get("file_position", 0),
line_number=data.get("line_number", 0),
is_complete=data.get("is_complete", False),
last_loaded_at=data.get("last_loaded_at"),
metadata=data.get("metadata", {}),
)
@dataclass
class PartialLoadingConfig:
"""
Configuration for partial/incremental loading.
Attributes:
enabled: Whether partial loading is enabled
initial_count: Number of items to load initially
batch_size: Number of items per incremental load
auto_load_threshold: Auto-load when this fraction is annotated (0.0-1.0)
auto_load_enabled: Whether auto-loading is enabled
"""
enabled: bool = False
initial_count: int = 1000
batch_size: int = 500
auto_load_threshold: float = 0.8
auto_load_enabled: bool = True
@classmethod
def from_dict(cls, config: Dict[str, Any]) -> "PartialLoadingConfig":
"""Create from configuration dictionary."""
partial_config = config.get("partial_loading", {})
return cls(
enabled=partial_config.get("enabled", False),
initial_count=partial_config.get("initial_count", 1000),
batch_size=partial_config.get("batch_size", 500),
auto_load_threshold=partial_config.get("auto_load_threshold", 0.8),
auto_load_enabled=partial_config.get("auto_load_enabled", True),
)
def validate(self) -> List[str]:
"""Validate the configuration."""
errors = []
if self.initial_count < 1:
errors.append("partial_loading.initial_count must be at least 1")
if self.batch_size < 1:
errors.append("partial_loading.batch_size must be at least 1")
if not 0.0 <= self.auto_load_threshold <= 1.0:
errors.append(
"partial_loading.auto_load_threshold must be between 0.0 and 1.0"
)
return errors
class PartialReader:
"""
Manages partial/incremental loading of data sources.
This class coordinates loading data in chunks across multiple sources,
tracking state for each source and providing auto-loading when
annotation progress reaches a threshold.
Thread Safety:
All public methods are thread-safe. Internal state is protected
by a lock.
Attributes:
config: Partial loading configuration
state_file: Path to state persistence file
"""
STATE_FILENAME = ".data_source_state.json"
def __init__(
self,
config: PartialLoadingConfig,
output_dir: str
):
"""
Initialize the partial reader.
Args:
config: Partial loading configuration
output_dir: Directory for state persistence
"""
self.config = config
self.output_dir = Path(output_dir)
self.state_file = self.output_dir / self.STATE_FILENAME
self._states: Dict[str, PartialReadState] = {}
self._lock = threading.RLock()
# Load existing state
self._load_state()
logger.info(
f"PartialReader initialized: initial={config.initial_count}, "
f"batch={config.batch_size}, auto_threshold={config.auto_load_threshold}"
)
def _load_state(self) -> None:
"""Load persisted state from disk."""
if not self.state_file.exists():
return
try:
with open(self.state_file, 'r', encoding='utf-8') as f:
data = json.load(f)
for state_data in data.get("sources", []):
state = PartialReadState.from_dict(state_data)
self._states[state.source_id] = state
logger.debug(f"Loaded partial read state for {len(self._states)} sources")
except (json.JSONDecodeError, KeyError) as e:
logger.warning(f"Failed to load partial read state: {e}")
self._states = {}
def _save_state(self) -> None:
"""Save state to disk."""
# Ensure output directory exists
self.output_dir.mkdir(parents=True, exist_ok=True)
data = {
"version": 1,
"sources": [state.to_dict() for state in self._states.values()]
}
try:
with open(self.state_file, 'w', encoding='utf-8') as f:
json.dump(data, f, indent=2)
except IOError as e:
logger.error(f"Failed to save partial read state: {e}")
def get_state(self, source_id: str) -> Optional[PartialReadState]:
"""
Get the current state for a source.
Args:
source_id: The source identifier
Returns:
PartialReadState or None if not tracked
"""
with self._lock:
return self._states.get(source_id)
def get_or_create_state(self, source_id: str) -> PartialReadState:
"""
Get or create state for a source.
Args:
source_id: The source identifier
Returns:
PartialReadState for the source
"""
with self._lock:
if source_id not in self._states:
self._states[source_id] = PartialReadState(source_id=source_id)
self._save_state()
return self._states[source_id]
def update_state(
self,
source_id: str,
items_added: int,
file_position: Optional[int] = None,
line_number: Optional[int] = None,
is_complete: bool = False,
total_estimate: Optional[int] = None
) -> PartialReadState:
"""
Update the state after loading items.
Args:
source_id: The source identifier
items_added: Number of items added in this batch
file_position: New file position (for file sources)
line_number: New line number (for line-based sources)
is_complete: Whether all data has been loaded
total_estimate: Updated total estimate
Returns:
Updated PartialReadState
"""
import time
with self._lock:
state = self.get_or_create_state(source_id)
state.items_loaded += items_added
state.last_loaded_at = time.time()
if file_position is not None:
state.file_position = file_position
if line_number is not None:
state.line_number = line_number
if total_estimate is not None:
state.total_estimate = total_estimate
if is_complete:
state.is_complete = True
self._save_state()
return state
def mark_complete(self, source_id: str) -> None:
"""
Mark a source as completely loaded.
Args:
source_id: The source identifier
"""
with self._lock:
state = self.get_or_create_state(source_id)
state.is_complete = True
self._save_state()
def should_load_more(
self,
source_id: str,
annotated_count: int,
total_loaded: int
) -> bool:
"""
Check if more data should be loaded based on annotation progress.
This method implements the auto-load threshold logic.
Args:
source_id: The source identifier
annotated_count: Number of items annotated
total_loaded: Total number of items loaded
Returns:
True if more data should be loaded
"""
if not self.config.auto_load_enabled:
return False
with self._lock:
state = self._states.get(source_id)
if state and state.is_complete:
return False
if total_loaded == 0:
return False
progress = annotated_count / total_loaded
return progress >= self.config.auto_load_threshold
def get_load_count(
self,
source_id: str,
is_initial: bool = False
) -> int:
"""
Get the number of items to load.
Args:
source_id: The source identifier
is_initial: Whether this is the initial load
Returns:
Number of items to load
"""
if is_initial:
return self.config.initial_count
return self.config.batch_size
def get_start_position(self, source_id: str) -> int:
"""
Get the starting position for the next load.
Args:
source_id: The source identifier
Returns:
Number of items already loaded (start position for next batch)
"""
with self._lock:
state = self._states.get(source_id)
if state:
return state.items_loaded
return 0
def reset_state(self, source_id: str) -> None:
"""
Reset the state for a source.
Args:
source_id: The source identifier
"""
with self._lock:
if source_id in self._states:
del self._states[source_id]
self._save_state()
def clear_all_state(self) -> None:
"""Clear state for all sources."""
with self._lock:
self._states.clear()
self._save_state()
def get_all_states(self) -> Dict[str, PartialReadState]:
"""
Get all source states.
Returns:
Dictionary mapping source_id to PartialReadState
"""
with self._lock:
return dict(self._states)
def get_stats(self) -> Dict[str, Any]:
"""
Get statistics about partial loading.
Returns:
Dictionary with loading statistics
"""
with self._lock:
total_loaded = sum(s.items_loaded for s in self._states.values())
complete_count = sum(1 for s in self._states.values() if s.is_complete)
return {
"enabled": self.config.enabled,
"initial_count": self.config.initial_count,
"batch_size": self.config.batch_size,
"auto_load_threshold": self.config.auto_load_threshold,
"sources_tracked": len(self._states),
"sources_complete": complete_count,
"total_items_loaded": total_loaded,
}