codebook / potato /data_sources /manager.py
davidjurgens's picture
Deploy: Potato — Codebook Annotation
aceb1b2 verified
Raw
History Blame Contribute Delete
17.6 kB
"""
Data Source Manager
This module provides the central manager for all data sources, implementing
the singleton pattern for thread-safe access across the application.
"""
import logging
import threading
from typing import Any, Dict, Iterator, List, Optional, Type, TYPE_CHECKING
from potato.data_sources.base import DataSource, SourceConfig, SourceType
from potato.data_sources.credentials import CredentialManager
from potato.data_sources.cache_manager import CacheManager
from potato.data_sources.partial_reader import PartialReader, PartialLoadingConfig
if TYPE_CHECKING:
from potato.item_state_management import ItemStateManager
logger = logging.getLogger(__name__)
# Singleton instance with thread-safe initialization
DATA_SOURCE_MANAGER: Optional["DataSourceManager"] = None
_MANAGER_LOCK = threading.Lock()
# Registry of source type implementations
_SOURCE_REGISTRY: Dict[SourceType, Type[DataSource]] = {}
def register_source_type(source_type: SourceType, source_class: Type[DataSource]) -> None:
"""
Register a data source implementation.
Args:
source_type: The SourceType enum value
source_class: The DataSource subclass
"""
_SOURCE_REGISTRY[source_type] = source_class
logger.debug(f"Registered source type: {source_type.value} -> {source_class.__name__}")
def get_source_class(source_type: SourceType) -> Optional[Type[DataSource]]:
"""
Get the source class for a given type.
Args:
source_type: The SourceType to look up
Returns:
The DataSource subclass, or None if not registered
"""
return _SOURCE_REGISTRY.get(source_type)
def get_registered_types() -> List[str]:
"""Get list of registered source type names."""
return [t.value for t in _SOURCE_REGISTRY.keys()]
class DataSourceManager:
"""
Central manager for all data sources.
This class provides:
- Registration and lifecycle management of data sources
- Credential management with environment variable substitution
- Caching for remote sources
- Partial/incremental loading coordination
- Thread-safe access to all sources
Attributes:
config: The application configuration
credential_manager: Handles credential resolution
cache_manager: Manages cached remote files
partial_reader: Coordinates incremental loading
"""
def __init__(
self,
config: Dict[str, Any],
item_state_manager: "ItemStateManager"
):
"""
Initialize the data source manager.
Args:
config: Application configuration dictionary
item_state_manager: The ItemStateManager for adding items
"""
self._config = config
self._item_state_manager = item_state_manager
self._sources: Dict[str, DataSource] = {}
self._lock = threading.RLock()
# Initialize sub-managers
self.credential_manager = CredentialManager.from_config(config)
# Set up cache manager if caching is enabled
cache_config = config.get("data_cache", {})
if cache_config.get("enabled", True):
cache_dir = cache_config.get(
"cache_dir",
".potato_cache/data_sources"
)
# Resolve relative to task_dir
task_dir = config.get("task_dir", ".")
if not cache_dir.startswith("/"):
import os
cache_dir = os.path.join(task_dir, cache_dir)
self.cache_manager = CacheManager(
cache_dir=cache_dir,
ttl_seconds=cache_config.get("ttl_seconds", 3600),
max_size_mb=cache_config.get("max_size_mb", 500)
)
else:
self.cache_manager = None
# Set up partial reader if incremental loading is configured
partial_config = PartialLoadingConfig.from_dict(config)
if partial_config.enabled:
output_dir = config.get("output_annotation_dir", ".")
self.partial_reader = PartialReader(partial_config, output_dir)
else:
self.partial_reader = None
# Get item property keys
item_props = config.get("item_properties", {})
self._id_key = item_props.get("id_key", "id")
self._text_key = item_props.get("text_key", "text")
# Initialize sources from configuration
self._init_sources()
logger.info(f"DataSourceManager initialized with {len(self._sources)} sources")
def _init_sources(self) -> None:
"""Initialize data sources from configuration."""
data_sources = self._config.get("data_sources", [])
for index, source_dict in enumerate(data_sources):
try:
# Process credentials in the source config
processed_config = self.credential_manager.process_config(source_dict)
# Parse source configuration
source_config = SourceConfig.from_dict(processed_config, index)
if not source_config.enabled:
logger.debug(f"Skipping disabled source: {source_config.source_id}")
continue
# Get the source class for this type
source_class = get_source_class(source_config.source_type)
if not source_class:
logger.warning(
f"No implementation for source type: {source_config.source_type.value}. "
f"Available types: {get_registered_types()}"
)
continue
# Create the source instance
source = source_class(source_config)
# Validate configuration
errors = source.validate_config()
if errors:
logger.error(
f"Invalid configuration for source {source_config.source_id}: "
f"{'; '.join(errors)}"
)
continue
# Check availability
if not source.is_available():
logger.warning(
f"Source {source_config.source_id} is not available. "
f"Check dependencies and credentials."
)
# Still register the source, but log the warning
# It may become available later
self._sources[source_config.source_id] = source
logger.info(
f"Initialized source: {source_config.source_id} "
f"(type={source_config.source_type.value})"
)
except Exception as e:
logger.error(f"Failed to initialize source at index {index}: {e}")
def get_source(self, source_id: str) -> Optional[DataSource]:
"""
Get a data source by ID.
Args:
source_id: The source identifier
Returns:
The DataSource instance, or None if not found
"""
with self._lock:
return self._sources.get(source_id)
def get_all_sources(self) -> Dict[str, DataSource]:
"""
Get all registered sources.
Returns:
Dictionary mapping source_id to DataSource
"""
with self._lock:
return dict(self._sources)
def list_sources(self) -> List[Dict[str, Any]]:
"""
List all sources with their status.
Returns:
List of source status dictionaries
"""
with self._lock:
statuses = []
for source in self._sources.values():
status = source.get_status()
# Add partial loading state if available
if self.partial_reader:
state = self.partial_reader.get_state(source.source_id)
if state:
status["items_loaded"] = state.items_loaded
status["is_complete"] = state.is_complete
status["last_loaded_at"] = state.last_loaded_at
statuses.append(status)
return statuses
def load_initial_data(self) -> int:
"""
Load initial data from all sources.
If partial loading is enabled, loads only the initial_count items
from each source. Otherwise, loads all data.
Returns:
Total number of items loaded
"""
total_loaded = 0
with self._lock:
for source_id, source in self._sources.items():
try:
count = self._load_from_source(source, is_initial=True)
total_loaded += count
logger.info(f"Loaded {count} items from {source_id}")
except Exception as e:
logger.error(f"Failed to load from {source_id}: {e}")
return total_loaded
def load_more(
self,
source_id: str,
count: Optional[int] = None
) -> int:
"""
Load more items from a specific source.
Args:
source_id: The source to load from
count: Number of items to load (uses batch_size if not specified)
Returns:
Number of items loaded
Raises:
ValueError: If source_id is not found
"""
with self._lock:
source = self._sources.get(source_id)
if not source:
raise ValueError(f"Unknown source: {source_id}")
return self._load_from_source(source, is_initial=False, count=count)
def _load_from_source(
self,
source: DataSource,
is_initial: bool = True,
count: Optional[int] = None
) -> int:
"""
Load items from a source into the ItemStateManager.
Args:
source: The data source
is_initial: Whether this is the initial load
count: Number of items to load (overrides config)
Returns:
Number of items loaded
"""
source_id = source.source_id
# Check if source is complete
if self.partial_reader:
state = self.partial_reader.get_state(source_id)
if state and state.is_complete:
logger.debug(f"Source {source_id} is already complete")
return 0
# Determine how many items to load and from what position
if self.partial_reader and self.partial_reader.config.enabled:
start = self.partial_reader.get_start_position(source_id)
if count is None:
count = self.partial_reader.get_load_count(source_id, is_initial)
else:
start = 0
count = None # Load all
# Check if source supports partial reading
if start > 0 and not source.supports_partial_reading():
logger.warning(
f"Source {source_id} does not support partial reading, "
f"cannot continue from position {start}"
)
return 0
# Load items
items_loaded = 0
is_complete = False
try:
for item in source.read_items(start=start, count=count):
# Validate ID key exists
if self._id_key not in item:
logger.warning(
f"Missing id_key '{self._id_key}' in item from {source_id}"
)
continue
instance_id = str(item[self._id_key])
# Check for duplicates
if self._item_state_manager.has_item(instance_id):
logger.debug(f"Skipping duplicate ID: {instance_id}")
continue
# Add item to state manager
try:
self._item_state_manager.add_item(instance_id, item)
items_loaded += 1
except ValueError as e:
logger.warning(f"Failed to add item {instance_id}: {e}")
# Check if we loaded fewer items than requested (source exhausted)
if count is not None and items_loaded < count:
is_complete = True
except StopIteration:
is_complete = True
# Update partial reader state
if self.partial_reader:
total_estimate = source.get_total_count()
self.partial_reader.update_state(
source_id=source_id,
items_added=items_loaded,
is_complete=is_complete,
total_estimate=total_estimate
)
return items_loaded
def refresh_source(self, source_id: str) -> bool:
"""
Refresh a data source (re-fetch from remote).
Args:
source_id: The source to refresh
Returns:
True if refresh was successful
Raises:
ValueError: If source_id is not found
"""
with self._lock:
source = self._sources.get(source_id)
if not source:
raise ValueError(f"Unknown source: {source_id}")
# Invalidate cache
if self.cache_manager:
self.cache_manager.invalidate(source_id)
# Reset partial reader state
if self.partial_reader:
self.partial_reader.reset_state(source_id)
return source.refresh()
def check_auto_load(
self,
annotated_count: int,
total_loaded: int
) -> Dict[str, int]:
"""
Check if any sources should auto-load more data.
Args:
annotated_count: Total number of annotated items
total_loaded: Total number of loaded items
Returns:
Dictionary mapping source_id to items loaded (for sources that triggered)
"""
if not self.partial_reader or not self.partial_reader.config.auto_load_enabled:
return {}
results = {}
with self._lock:
for source_id, source in self._sources.items():
if self.partial_reader.should_load_more(
source_id, annotated_count, total_loaded
):
try:
loaded = self._load_from_source(source, is_initial=False)
if loaded > 0:
results[source_id] = loaded
logger.info(
f"Auto-loaded {loaded} items from {source_id}"
)
except Exception as e:
logger.error(f"Auto-load failed for {source_id}: {e}")
return results
def clear_cache(self) -> int:
"""
Clear the download cache for all sources.
Returns:
Number of cache entries cleared
"""
if self.cache_manager:
return self.cache_manager.clear()
return 0
def get_stats(self) -> Dict[str, Any]:
"""
Get comprehensive statistics.
Returns:
Dictionary with source and manager statistics
"""
stats = {
"source_count": len(self._sources),
"sources": self.list_sources(),
}
if self.cache_manager:
stats["cache"] = self.cache_manager.get_stats()
if self.partial_reader:
stats["partial_loading"] = self.partial_reader.get_stats()
return stats
def close(self) -> None:
"""Close all sources and release resources."""
with self._lock:
for source in self._sources.values():
try:
source.close()
except Exception as e:
logger.warning(f"Error closing source {source.source_id}: {e}")
self._sources.clear()
def init_data_source_manager(config: Dict[str, Any]) -> Optional[DataSourceManager]:
"""
Initialize the global DataSourceManager singleton.
This function creates the manager if data_sources is configured in
the configuration. Thread-safe initialization using double-checked
locking pattern.
Args:
config: Application configuration dictionary
Returns:
The DataSourceManager instance, or None if not configured
"""
global DATA_SOURCE_MANAGER
# Check if data_sources is configured
if "data_sources" not in config or not config["data_sources"]:
return None
# Double-checked locking
if DATA_SOURCE_MANAGER is None:
with _MANAGER_LOCK:
if DATA_SOURCE_MANAGER is None:
from potato.item_state_management import get_item_state_manager
ism = get_item_state_manager()
DATA_SOURCE_MANAGER = DataSourceManager(config, ism)
return DATA_SOURCE_MANAGER
def get_data_source_manager() -> Optional[DataSourceManager]:
"""
Get the global DataSourceManager singleton.
Returns:
The DataSourceManager instance, or None if not initialized
"""
return DATA_SOURCE_MANAGER
def clear_data_source_manager() -> None:
"""
Clear the global DataSourceManager singleton.
This function closes all sources and clears the singleton instance.
Thread-safe. Used primarily for testing.
"""
global DATA_SOURCE_MANAGER
with _MANAGER_LOCK:
if DATA_SOURCE_MANAGER is not None:
DATA_SOURCE_MANAGER.close()
DATA_SOURCE_MANAGER = None