import platform
import re
import sqlalchemy
from copy import deepcopy
from elasticsearch.exceptions import NotFoundError
from elasticsearch.helpers import streaming_bulk
from langdetect.lang_detect_exception import LangDetectException
from itertools import groupby
from operator import itemgetter
from queue import Queue, Empty, Full
from onegov.core.utils import is_non_string_iterable
from onegov.search import index_log, log, Searchable, utils
from onegov.search.errors import SearchOfflineError
from typing import Any, Literal, NamedTuple, TYPE_CHECKING
if TYPE_CHECKING:
from collections.abc import Iterable, Iterator, Sequence
from elasticsearch import Elasticsearch
from sqlalchemy.engine import Engine
from sqlalchemy.orm import Session
from typing import TypeAlias
from typing import TypedDict
from uuid import UUID
[docs]
class IndexTask(TypedDict):
[docs]
action: Literal['index']
[docs]
type_name: str # FIXME: not needed for fts
[docs]
properties: dict[str, Any]
class DeleteTask(TypedDict): # FIXME: not needed for fts
action: Literal['delete']
id: UUID | str | int
schema: str
type_name: str
tablename: str
Task: TypeAlias = IndexTask | DeleteTask
[docs]
ES_ANALYZER_MAP = {
'en': 'english',
'de': 'german',
'fr': 'french',
'en_html': 'english_html',
'de_html': 'german_html',
'fr_html': 'french_html',
}
[docs]
ANALYSIS_CONFIG = {
'filter': {
'english_stop': {
'type': 'stop',
'stopwords': '_english_'
},
'english_stemmer': {
'type': 'stemmer',
'language': 'english'
},
'english_possessive_stemmer': {
'type': 'stemmer',
'language': 'possessive_english'
},
'german_stop': {
'type': 'stop',
'stopwords': '_german_'
},
'german_stemmer': {
'type': 'stemmer',
'language': 'light_german'
},
'french_elision': {
'type': 'elision',
'articles_case': True,
'articles': [
'l', 'm', 't', 'qu', 'n', 's',
'j', 'd', 'c', 'jusqu', 'quoiqu',
'lorsqu', 'puisqu'
]
},
'french_stop': {
'type': 'stop',
'stopwords': '_french_'
},
'french_keywords': {
'type': 'keyword_marker',
'keywords': ['Exemple']
},
'french_stemmer': {
'type': 'stemmer',
'language': 'light_french'
}
},
'analyzer': {
'english_html': {
'tokenizer': 'standard',
'char_filter': [
'html_strip'
],
'filter': [
'english_possessive_stemmer',
'lowercase',
'english_stop',
'english_stemmer'
]
},
'german_html': {
'tokenizer': 'standard',
'char_filter': [
'html_strip'
],
'filter': [
'lowercase',
'german_stop',
'german_normalization',
'german_stemmer'
]
},
'french_html': {
'tokenizer': 'standard',
'char_filter': [
'html_strip'
],
'filter': [
'french_elision',
'lowercase',
'french_stop',
'french_keywords',
'french_stemmer'
]
},
'autocomplete': {
'type': 'custom',
'char_filter': ['html_strip'],
'tokenizer': 'standard',
'filter': ['lowercase']
},
'tags': {
'type': 'custom',
'tokenizer': 'keyword',
'filter': ['lowercase']
},
}
}
[docs]
class IndexParts(NamedTuple):
[docs]
def parse_index_name(index_name: str) -> IndexParts:
""" Takes the given index name and returns the hostname, schema,
language and type_name in a dictionary.
* If the index_name doesn't match the pattern, all values are None.
* If the index_name has no version, the version is None.
"""
parts = index_name.split('-')
if len(parts) == 4:
hostname, schema, language, type_name = parts
version = None
elif len(parts) == 5:
hostname, schema, language, type_name, version = parts
else:
hostname = None
schema = None
language = None
type_name = None
version = None
return IndexParts(
hostname=hostname,
schema=schema,
language=language,
type_name=type_name,
version=version
)
[docs]
class IndexerBase:
[docs]
failed_task: 'Task | None' = None
[docs]
def process(
self,
block: bool = False,
timeout: float | None = None
) -> int:
""" Processes the queue until it is empty or until there's an error.
If there's an error, the next call to this function will try to
execute the failed task again. This is mainly meant for elasticsearch
outages.
:block:
If True, the process waits for the queue to be available. Useful
if you run this in a separate thread.
:timeout:
How long the blocking call should block. Has no effect if
``block`` is False.
:return: The number of successfully processed items
"""
try:
processed = 0
while True:
# get the previously failed task or a new one
task = self.failed_task or self.queue.get(block, timeout)
self.failed_task = None
if self.process_task(task):
processed += 1
else:
# if the task failed, keep it for the next run and give up
self.failed_task = task
return processed
except Empty:
pass
return processed
[docs]
def process_task(self, task: 'Task') -> bool:
try:
getattr(self, task['action'])(task)
except SearchOfflineError:
return False
self.queue.task_done()
return True
[docs]
class Indexer(IndexerBase):
""" Takes actions from a queue and executes them on the elasticsearch
cluster. Depends on :class:`IndexManager` for index management and expects
to have the same :class:`TypeRegistry` as :class:`ORMEventTranslator`.
The idea is that this class does the indexing/deindexing, the index manager
sets up the indices and the orm event translator listens for changes in
the ORM.
A queue is used so the indexer can be run in a separate thread.
"""
def __init__(
self,
mappings: 'TypeMappingRegistry',
queue: 'Queue[Task]',
es_client: 'Elasticsearch',
hostname: str | None = None
) -> None:
[docs]
self.es_client = es_client
self.queue = queue
[docs]
self.hostname = hostname or platform.node()
[docs]
self.ixmgr = IndexManager(self.hostname, es_client=self.es_client)
[docs]
self.mappings = mappings
[docs]
def bulk_process(self) -> None:
""" Processes the queue in bulk. This offers better performance but it
is less safe at the moment and should only be used as part of
reindexing.
"""
def actions() -> 'Iterator[dict[str, Any]]':
try:
task = self.queue.get(block=False, timeout=None)
if task['action'] == 'index':
yield {
'_op_type': 'index',
'_index': self.ensure_index(task),
'_id': task['id'],
'doc': task['properties']
}
elif task['action'] == 'delete':
# FIXME: I'm unsure about how this ever worked...
yield {
'_op_type': 'delete',
'_index': self.ensure_index(task), # type:ignore
'_id': task['id'],
'doc': task['properties'] # type:ignore
}
else:
raise NotImplementedError
except Empty:
pass
for success, info in streaming_bulk(self.es_client, actions()):
if success:
self.queue.task_done()
[docs]
def ensure_index(self, task: 'IndexTask') -> str:
return self.ixmgr.ensure_index(
task['schema'],
task['language'],
self.mappings[task['type_name']],
return_index='internal'
)
[docs]
def index(self, task: 'IndexTask') -> None:
index = self.ensure_index(task)
self.es_client.index(
index=index,
id=task['id'], # type:ignore[arg-type]
document=task['properties']
)
[docs]
def delete(self, task: 'DeleteTask') -> None:
# get all the types this model could be stored in (with polymorphic)
# identites, this could be many
mapping = self.mappings[task['type_name']]
if mapping.model:
types = utils.related_types(mapping.model)
else:
types = {mapping.name}
# delete the document from all languages (because we don't know
# which one anymore) - and delete from all related types (polymorphic)
for type in types:
ix = self.ixmgr.get_external_index_name(
schema=task['schema'],
language='*',
type_name=type
)
# for the delete operation we need the internal index names
for internal in self.es_client.indices.get_alias(index=ix).keys():
try:
self.es_client.delete(
index=internal,
id=task['id'] # type:ignore[arg-type]
)
except NotFoundError:
pass
[docs]
class PostgresIndexer(IndexerBase):
[docs]
TEXT_SEARCH_COLUMN_NAME = 'fts_idx'
[docs]
idx_language_mapping = {
'de': 'german',
'fr': 'french',
'it': 'italian',
'en': 'english',
}
[docs]
queue: 'Queue[IndexTask]'
def __init__(
self,
queue: 'Queue[IndexTask]',
engine: 'Engine',
) -> None:
self.queue = queue
[docs]
self.engine = engine
[docs]
def index(
self,
tasks: 'list[IndexTask] | IndexTask',
session: 'Session | None' = None
) -> bool:
""" Update the 'fts_idx' column (full text search index) of the given
object(s)/task(s).
In case of a bunch of tasks we are assuming they are all from the
same schema and table in order to optimize the indexing process.
When a session is passed we use that session's transaction context
and use a savepoint instead of our own transaction to perform the
action.
:param tasks: A list of tasks to index
:param session: Supply an active session
:return: True if the indexing was successful, False otherwise
"""
content = []
if not isinstance(tasks, list):
tasks = [tasks]
try:
for task in tasks:
language = (
self.idx_language_mapping.get(task['language'], 'simple'))
data = {
k: str(v)
for k, v in task['properties'].items()
if not k.startswith('es_')}
_id = task['id']
content.append(
{'language': language, 'data': data, '_id': _id})
schema = tasks[0]['schema']
tablename = tasks[0]['tablename']
id_key = tasks[0]['id_key']
table = sqlalchemy.table(
tablename,
(id_col := sqlalchemy
.column(id_key)), # type: ignore[var-annotated]
sqlalchemy.column(self.TEXT_SEARCH_COLUMN_NAME),
schema=schema # type: ignore
)
tsvector_expr = sqlalchemy.text(
'to_tsvector(:language, :data)').bindparams(
sqlalchemy.bindparam('language', type_=sqlalchemy.String),
sqlalchemy.bindparam('data', type_=sqlalchemy.JSON)
)
stmt = (
sqlalchemy.update(table)
.where(id_col == sqlalchemy.bindparam('_id'))
.values({self.TEXT_SEARCH_COLUMN_NAME: tsvector_expr})
)
if session is None:
connection = self.engine.connect()
with connection.begin():
connection.execute(stmt, content)
else:
# use a savepoint instead
with session.begin_nested():
session.execute(stmt, content)
except Exception as ex:
index_log.error(f"Error '{ex}' indexing schema "
f'{tasks[0]["schema"]} table '
f'{tasks[0]["tablename"]}')
return False
return True
# FIXME: bulk_process should probably be the only function we use for
# the Postgres indexer, we don't have to worry about individual
# transactions failing as much
[docs]
def bulk_process(self, session: 'Session | None' = None) -> None:
""" Processes the queue in bulk. This offers better performance but it
is less safe at the moment and should only be used as part of
reindexing.
Gather all index tasks, group them by model and index batch-wise
"""
def task_generator() -> 'Iterator[IndexTask]':
while not self.queue.empty():
task = self.queue.get(block=False, timeout=None)
self.queue.task_done()
yield task
grouped_tasks = groupby(task_generator(), key=itemgetter('action',
'tablename'))
for (action, tablename), tasks in grouped_tasks:
task_list = list(tasks)
if action == 'index':
self.index(task_list, session)
else:
raise NotImplementedError("Action '{action}' not "
'implemented')
[docs]
class TypeMapping:
[docs]
__slots__ = ('name', 'mapping', 'version', 'model')
def __init__(
self,
name: str,
mapping: dict[str, Any],
model: type[Searchable] | None = None
) -> None:
[docs]
self.mapping = self.add_defaults(mapping)
[docs]
self.version = utils.hash_mapping(mapping)
[docs]
def add_defaults(self, mapping: dict[str, Any]) -> dict[str, Any]:
mapping['es_public'] = {
'type': 'boolean'
}
mapping['es_last_change'] = {
'type': 'date'
}
mapping['es_suggestion'] = {
'analyzer': 'autocomplete',
'type': 'completion',
'contexts': [
{
'name': 'es_suggestion_context',
'type': 'category'
}
]
}
mapping['es_tags'] = {
'analyzer': 'tags',
'type': 'text',
}
return mapping
[docs]
def for_language(self, language: str) -> dict[str, Any]:
""" Returns the mapping for the given language. Mappings can
be slightly different for each language. That is, the analyzer
changes.
Because the :class:`IndexManager` puts each language into its own
index we do not have to worry about creating different versions
of the same mapping here.
"""
return self.supplement_analyzer(deepcopy(self.mapping), language)
[docs]
def supplement_analyzer(
self,
dictionary: dict[str, Any],
language: str
) -> dict[str, Any]:
""" Iterate through the dictionary found in the type mapping and
replace the 'localized' type with a 'text' type that includes a
language specific analyzer.
"""
supplement = None
for key, value in dictionary.items():
if hasattr(value, 'items'):
dictionary[key] = self.supplement_analyzer(value, language)
elif key == 'type' and value.startswith('localized'):
supplement = value.replace('localized', language)
break
if supplement:
assert 'analyzer' not in dictionary
dictionary[key] = 'text'
dictionary['analyzer'] = ES_ANALYZER_MAP[supplement]
return dictionary
[docs]
class TypeMappingRegistry:
[docs]
mappings: dict[str, TypeMapping]
def __init__(self) -> None:
self.mappings = {}
[docs]
def __getitem__(self, key: str) -> TypeMapping:
return self.mappings[key]
[docs]
def __iter__(self) -> 'Iterator[TypeMapping]':
yield from self.mappings.values()
[docs]
def register_orm_base(self, base: type[object]) -> None:
""" Takes the given SQLAlchemy base and registers all
:class:`~onegov.search.mixins.Searchable` objects.
"""
for model in utils.searchable_sqlalchemy_models(base):
self.register_type(model.es_type_name, model.es_properties, model)
[docs]
def register_type(
self,
type_name: str,
mapping: dict[str, Any],
model: type[Searchable] | None = None
) -> None:
""" Registers the given type with the given mapping. The mapping is
as dictionary representing the part below the ``mappings/type_name``.
See:
`<https://www.elastic.co/guide/en/elasticsearch/reference/current/\
indices-create-index.html#mappings>`_
When the mapping changes, a new index is created internally and the
alias to this index (the external name of the index) is pointed to
this new index.
As a consequence, a change in the mapping requires a reindex.
"""
assert type_name not in self.mappings
self.mappings[type_name] = TypeMapping(type_name, mapping, model)
@property
[docs]
def registered_fields(self) -> set[str]:
""" Goes through all the registered types and returns the a set with
all fields used by the mappings.
"""
return {key for mapping in self for key in mapping.mapping.keys()}
[docs]
class IndexManager:
""" Manages the creation/destruction of indices. The indices it creates
have an internal name and an external alias. To facilitate that, versions
are used.
"""
[docs]
created_indices: set[str]
def __init__(self, hostname: str, es_client: 'Elasticsearch') -> None:
assert hostname and es_client
[docs]
self.hostname = hostname
[docs]
self.es_client = es_client
self.created_indices = set()
@property
[docs]
def normalized_hostname(self) -> str:
return utils.normalize_index_segment(
self.hostname, allow_wildcards=False)
[docs]
def query_indices(self) -> set[str]:
""" Queryies the elasticsearch cluster for indices belonging to this
hostname. """
return set(
self.es_client.cat.indices( # type:ignore[union-attr]
index=f'{self.normalized_hostname}-*', h='index'
).splitlines()
)
[docs]
def query_aliases(self) -> set[str]:
""" Queryies the elasticsearch cluster for aliases belonging to this
hostname. """
result = set()
infos = self.es_client.indices.get_alias(
index='{}-*'.format(self.normalized_hostname)
)
for info in infos.values():
for alias in info['aliases']:
result.add(alias)
return result
[docs]
def ensure_index(
self,
schema: str,
language: str,
mapping: TypeMapping,
return_index: Literal['external', 'internal'] = 'external'
) -> str:
""" Takes the given database schema, language and type name and
creates an internal index with a version number and an external
alias without the version number.
:schema:
The database schema this index is based on.
:language:
The language in ISO 639-1 format.
:mapping:
The :class:`TypeMapping` mapping used in this index.
:return_index:
The index name to return. Either 'external' or 'internal'.
:return:
The (external/aliased) name of the created index.
"""
assert schema and language and mapping
assert len(language) == 2
assert return_index == 'external' or return_index == 'internal'
external = self.get_external_index_name(schema, language, mapping.name)
internal = self.get_internal_index_name(
schema, language, mapping.name, mapping.version)
return_value = return_index == 'external' and external or internal
if internal in self.created_indices:
return return_value
if self.es_client.indices.exists(index=internal):
self.created_indices.add(internal)
return return_value
# create the index
self.es_client.indices.create(
index=internal,
mappings={
'properties': mapping.for_language(language)
},
settings={
'analysis': ANALYSIS_CONFIG,
'index': {
'number_of_shards': 1,
'number_of_replicas': 0,
'refresh_interval': '5s'
}
}
)
# point the alias to the new index
self.es_client.indices.put_alias(name=external, index=internal)
# cache the result
self.created_indices.add(internal)
return return_value
[docs]
def remove_expired_indices(
self,
current_mappings: 'Iterable[TypeMapping]'
) -> int:
""" Removes all expired indices. An index is expired if it's version
number is no longer known in the current mappings.
:return: The number of indices that were deleted.
"""
active_versions = {m.version for m in current_mappings}
count = 0
for index in self.query_indices():
info = parse_index_name(index)
if info.version and info.version not in active_versions:
self.es_client.indices.delete(index=index)
self.created_indices.remove(index)
count += 1
return count
[docs]
def get_managed_indices_wildcard(self, schema: str) -> str:
""" Returns a wildcard index name for all indices managed. """
return '-'.join((
utils.normalize_index_segment(
self.hostname, allow_wildcards=False),
utils.normalize_index_segment(
schema, allow_wildcards=False),
'*'
))
[docs]
def get_external_index_names(
self,
schema: str,
languages: 'Iterable[str]' = '*',
types: 'Iterable[str]' = '*'
) -> str:
""" Returns a comma separated string of external index names that
match the given arguments. Useful to pass on to elasticsearch when
targeting multiple indices.
"""
return ','.join(
self.get_external_index_name(schema, language, type_name)
for language in languages
for type_name in types
)
[docs]
def get_external_index_name(
self,
schema: str,
language: str,
type_name: str
) -> str:
""" Generates the external index name from the given parameters. """
return '-'.join(
utils.normalize_index_segment(segment, allow_wildcards=True)
for segment in (self.hostname, schema, language, type_name)
)
[docs]
def get_internal_index_name(
self,
schema: str,
language: str,
type_name: str,
version: str
) -> str:
""" Generates the internal index name from the given parameters. """
return '-'.join((
self.get_external_index_name(schema, language, type_name),
utils.normalize_index_segment(version, allow_wildcards=False)
))
[docs]
class ORMLanguageDetector(utils.LanguageDetector):
[docs]
html_strip_expression = re.compile(r'<[^<]+?>')
[docs]
def localized_properties(self, obj: Searchable) -> 'Iterator[str]':
for key, definition in obj.es_properties.items():
if definition.get('type', '').startswith('localized'):
yield key
[docs]
def localized_texts(
self,
obj: Searchable,
max_chars: int | None = None
) -> 'Iterator[str]':
chars = 0
for p in self.localized_properties(obj):
text = getattr(obj, p, '')
if not isinstance(text, str):
continue
yield text.strip()
chars += len(text)
if max_chars is not None and max_chars <= chars:
break
[docs]
def detect_object_language(self, obj: Searchable) -> str:
properties = self.localized_properties(obj)
if not properties:
# here, the mapping will be the same for all languages
return self.supported_languages[0]
text = ' '.join(self.localized_texts(obj, max_chars=1024))
text = self.html_strip_expression.sub('', text).strip()
if not text:
return self.supported_languages[0]
try:
return self.detect(text)
except LangDetectException:
return self.supported_languages[0]
[docs]
class ORMEventTranslator:
""" Handles the onegov.core orm events, translates them into indexing
actions and puts the result into a queue for the indexer to consume.
The queue may be limited. Once the limit is reached, new events are no
longer processed and an error is logged.
"""
[docs]
converters = {
'date': lambda dt: dt and dt.isoformat(),
}
[docs]
es_queue: 'Queue[Task]'
[docs]
psql_queue: 'Queue[IndexTask]'
def __init__(
self,
mappings: TypeMappingRegistry,
max_queue_size: int = 0,
languages: 'Sequence[str]' = ('de', 'fr', 'en')
) -> None:
[docs]
self.mappings = mappings
self.es_queue = Queue(maxsize=max_queue_size)
self.psql_queue = Queue(maxsize=max_queue_size)
[docs]
self.detector = ORMLanguageDetector(languages)
[docs]
def on_insert(self, schema: str, obj: object) -> None:
if not self.stopped:
if isinstance(obj, Searchable):
self.index(schema, obj)
[docs]
def on_update(self, schema: str, obj: object) -> None:
if not self.stopped:
if isinstance(obj, Searchable):
self.delete(schema, obj)
self.index(schema, obj)
[docs]
def on_delete(self, schema: str, obj: object) -> None:
if not self.stopped:
if isinstance(obj, Searchable):
self.delete(schema, obj)
[docs]
def put(self, translation: 'Task') -> None:
try:
self.es_queue.put_nowait(translation)
if translation['action'] == 'index':
# we only need to provide index tasks for fts
self.psql_queue.put_nowait(translation)
except Full:
log.error('The orm event translator queue is full!')
[docs]
def index(self, schema: str, obj: Searchable) -> None:
if obj.es_skip:
return
if obj.es_language == 'auto':
language = self.detector.detect_object_language(obj)
else:
language = obj.es_language
translation: IndexTask = {
'action': 'index',
'id': getattr(obj, obj.es_id),
'id_key': obj.es_id,
'schema': schema,
'type_name': obj.es_type_name, # FIXME: not needed for fts
'tablename': obj.__tablename__, # type:ignore[attr-defined]
'language': language,
'properties': {}
}
mapping_ = self.mappings[obj.es_type_name].for_language(language)
for prop, mapping in mapping_.items():
if prop == 'es_suggestion':
continue
convert = self.converters.get(mapping['type'], lambda v: v)
raw = getattr(obj, prop)
if is_non_string_iterable(raw):
translation['properties'][prop] = [convert(v) for v in raw]
else:
translation['properties'][prop] = convert(raw)
if obj.es_public:
contexts = {'es_suggestion_context': ['public']}
else:
contexts = {'es_suggestion_context': ['private']}
suggestion = obj.es_suggestion
if is_non_string_iterable(suggestion):
translation['properties']['es_suggestion'] = {
'input': suggestion,
'contexts': contexts
}
else:
translation['properties']['es_suggestion'] = {
'input': [suggestion],
'contexts': contexts
}
self.put(translation)
[docs]
def delete(self, schema: str, obj: Searchable) -> None:
translation: DeleteTask = {
'action': 'delete',
'schema': schema,
'type_name': obj.es_type_name,
'tablename': obj.__tablename__, # type:ignore[attr-defined]
'id': getattr(obj, obj.es_id)
}
self.put(translation)