# Licensed to Elasticsearch B.V. under one or more contributor
# license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright
# ownership. Elasticsearch B.V. licenses this file to you under
# the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
import logging
import time
from operator import methodcaller
from ..compat import Mapping, Queue, map, string_types
from ..exceptions import TransportError
from .errors import BulkIndexError, ScanError
logger = logging.getLogger("elasticsearch.helpers")
def expand_action(data):
"""
From one document or action definition passed in by the user extract the
action/data lines needed for elasticsearch's
:meth:`~elasticsearch.Elasticsearch.bulk` api.
"""
# when given a string, assume user wants to index raw json
if isinstance(data, string_types):
return '{"index":{}}', data
# make sure we don't alter the action
data = data.copy()
op_type = data.pop("_op_type", "index")
action = {op_type: {}}
# If '_source' is a dict use it for source
# otherwise if op_type == 'update' then
# '_source' should be in the metadata.
if (
op_type == "update"
and "_source" in data
and not isinstance(data["_source"], Mapping)
):
action[op_type]["_source"] = data.pop("_source")
for key in (
"_id",
"_index",
"_if_seq_no",
"_if_primary_term",
"_parent",
"_percolate",
"_retry_on_conflict",
"_routing",
"_timestamp",
"_type",
"_version",
"_version_type",
"if_seq_no",
"if_primary_term",
"parent",
"pipeline",
"retry_on_conflict",
"routing",
"version",
"version_type",
):
if key in data:
if key in {
"_if_seq_no",
"_if_primary_term",
"_parent",
"_retry_on_conflict",
"_routing",
"_version",
"_version_type",
}:
action[op_type][key[1:]] = data.pop(key)
else:
action[op_type][key] = data.pop(key)
# no data payload for delete
if op_type == "delete":
return action, None
return action, data.get("_source", data)
class _ActionChunker:
def __init__(self, chunk_size, max_chunk_bytes, serializer):
self.chunk_size = chunk_size
self.max_chunk_bytes = max_chunk_bytes
self.serializer = serializer
self.size = 0
self.action_count = 0
self.bulk_actions = []
self.bulk_data = []
def feed(self, action, data):
ret = None
raw_data, raw_action = data, action
action = self.serializer.dumps(action)
# +1 to account for the trailing new line character
cur_size = len(action.encode("utf-8")) + 1
if data is not None:
data = self.serializer.dumps(data)
cur_size += len(data.encode("utf-8")) + 1
# full chunk, send it and start a new one
if self.bulk_actions and (
self.size + cur_size > self.max_chunk_bytes
or self.action_count == self.chunk_size
):
ret = (self.bulk_data, self.bulk_actions)
self.bulk_actions, self.bulk_data = [], []
self.size, self.action_count = 0, 0
self.bulk_actions.append(action)
if data is not None:
self.bulk_actions.append(data)
self.bulk_data.append((raw_action, raw_data))
else:
self.bulk_data.append((raw_action,))
self.size += cur_size
self.action_count += 1
return ret
def flush(self):
ret = None
if self.bulk_actions:
ret = (self.bulk_data, self.bulk_actions)
self.bulk_actions, self.bulk_data = [], []
return ret
def _chunk_actions(actions, chunk_size, max_chunk_bytes, serializer):
"""
Split actions into chunks by number or size, serialize them into strings in
the process.
"""
chunker = _ActionChunker(
chunk_size=chunk_size, max_chunk_bytes=max_chunk_bytes, serializer=serializer
)
for action, data in actions:
ret = chunker.feed(action, data)
if ret:
yield ret
ret = chunker.flush()
if ret:
yield ret
def _process_bulk_chunk_success(resp, bulk_data, ignore_status, raise_on_error=True):
# if raise on error is set, we need to collect errors per chunk before raising them
errors = []
# go through request-response pairs and detect failures
for data, (op_type, item) in zip(
bulk_data, map(methodcaller("popitem"), resp["items"])
):
status_code = item.get("status", 500)
ok = 200 <= status_code < 300
if not ok and raise_on_error and status_code not in ignore_status:
# include original document source
if len(data) > 1:
item["data"] = data[1]
errors.append({op_type: item})
if ok or not errors:
# if we are not just recording all errors to be able to raise
# them all at once, yield items individually
yield ok, {op_type: item}
if errors:
raise BulkIndexError("%i document(s) failed to index." % len(errors), errors)
def _process_bulk_chunk_error(
error, bulk_data, ignore_status, raise_on_exception=True, raise_on_error=True
):
# default behavior - just propagate exception
if raise_on_exception and error.status_code not in ignore_status:
raise error
# if we are not propagating, mark all actions in current chunk as failed
err_message = str(error)
exc_errors = []
for data in bulk_data:
# collect all the information about failed actions
op_type, action = data[0].copy().popitem()
info = {"error": err_message, "status": error.status_code, "exception": error}
if op_type != "delete":
info["data"] = data[1]
info.update(action)
exc_errors.append({op_type: info})
# emulate standard behavior for failed actions
if raise_on_error and error.status_code not in ignore_status:
raise BulkIndexError(
"%i document(s) failed to index." % len(exc_errors), exc_errors
)
else:
for err in exc_errors:
yield False, err
def _process_bulk_chunk(
client,
bulk_actions,
bulk_data,
raise_on_exception=True,
raise_on_error=True,
ignore_status=(),
*args,
**kwargs
):
"""
Send a bulk request to elasticsearch and process the output.
"""
kwargs = _add_helper_meta_to_kwargs(kwargs, "bp")
if not isinstance(ignore_status, (list, tuple)):
ignore_status = (ignore_status,)
try:
# send the actual request
resp = client.bulk("\n".join(bulk_actions) + "\n", *args, **kwargs)
except TransportError as e:
gen = _process_bulk_chunk_error(
error=e,
bulk_data=bulk_data,
ignore_status=ignore_status,
raise_on_exception=raise_on_exception,
raise_on_error=raise_on_error,
)
else:
gen = _process_bulk_chunk_success(
resp=resp,
bulk_data=bulk_data,
ignore_status=ignore_status,
raise_on_error=raise_on_error,
)
for item in gen:
yield item
def _add_helper_meta_to_kwargs(kwargs, helper_meta):
params = (kwargs or {}).pop("params", {})
params["__elastic_client_meta"] = (("h", helper_meta),)
kwargs["params"] = params
return kwargs
def streaming_bulk(
client,
actions,
chunk_size=500,
max_chunk_bytes=100 * 1024 * 1024,
raise_on_error=True,
expand_action_callback=expand_action,
raise_on_exception=True,
max_retries=0,
initial_backoff=2,
max_backoff=600,
yield_ok=True,
ignore_status=(),
*args,
**kwargs
):
"""
Streaming bulk consumes actions from the iterable passed in and yields
results per action. For non-streaming usecases use
:func:`~elasticsearch.helpers.bulk` which is a wrapper around streaming
bulk that returns summary information about the bulk operation once the
entire input is consumed and sent.
If you specify ``max_retries`` it will also retry any documents that were
rejected with a ``429`` status code. To do this it will wait (**by calling
time.sleep which will block**) for ``initial_backoff`` seconds and then,
every subsequent rejection for the same chunk, for double the time every
time up to ``max_backoff`` seconds.
:arg client: instance of :class:`~elasticsearch.Elasticsearch` to use
:arg actions: iterable containing the actions to be executed
:arg chunk_size: number of docs in one chunk sent to es (default: 500)
:arg max_chunk_bytes: the maximum size of the request in bytes (default: 100MB)
:arg raise_on_error: raise ``BulkIndexError`` containing errors (as `.errors`)
from the execution of the last chunk when some occur. By default we raise.
:arg raise_on_exception: if ``False`` then don't propagate exceptions from
call to ``bulk`` and just report the items that failed as failed.
:arg expand_action_callback: callback executed on each action passed in,
should return a tuple containing the action line and the data line
(`None` if data line should be omitted).
:arg max_retries: maximum number of times a document will be retried when
``429`` is received, set to 0 (default) for no retries on ``429``
:arg initial_backoff: number of seconds we should wait before the first
retry. Any subsequent retries will be powers of ``initial_backoff *
2**retry_number``
:arg max_backoff: maximum number of seconds a retry will wait
:arg yield_ok: if set to False will skip successful documents in the output
:arg ignore_status: list of HTTP status code that you want to ignore
"""
actions = map(expand_action_callback, actions)
for bulk_data, bulk_actions in _chunk_actions(
actions, chunk_size, max_chunk_bytes, client.transport.serializer
):
for attempt in range(max_retries + 1):
to_retry, to_retry_data = [], []
if attempt:
time.sleep(min(max_backoff, initial_backoff * 2 ** (attempt - 1)))
try:
for data, (ok, info) in zip(
bulk_data,
_process_bulk_chunk(
client,
bulk_actions,
bulk_data,
raise_on_exception,
raise_on_error,
ignore_status,
*args,
**kwargs
),
):
if not ok:
action, info = info.popitem()
# retry if retries enabled, we get 429, and we are not
# in the last attempt
if (
max_retries
and info["status"] == 429
and (attempt + 1) <= max_retries
):
# _process_bulk_chunk expects strings so we need to
# re-serialize the data
to_retry.extend(
map(client.transport.serializer.dumps, data)
)
to_retry_data.append(data)
else:
yield ok, {action: info}
elif yield_ok:
yield ok, info
except TransportError as e:
# suppress 429 errors since we will retry them
if attempt == max_retries or e.status_code != 429:
raise
else:
if not to_retry:
break
# retry only subset of documents that didn't succeed
bulk_actions, bulk_data = to_retry, to_retry_data
def bulk(client, actions, stats_only=False, ignore_status=(), *args, **kwargs):
"""
Helper for the :meth:`~elasticsearch.Elasticsearch.bulk` api that provides
a more human friendly interface - it consumes an iterator of actions and
sends them to elasticsearch in chunks. It returns a tuple with summary
information - number of successfully executed actions and either list of
errors or number of errors if ``stats_only`` is set to ``True``. Note that
by default we raise a ``BulkIndexError`` when we encounter an error so
options like ``stats_only`` only apply when ``raise_on_error`` is set to
``False``.
When errors are being collected original document data is included in the
error dictionary which can lead to an extra high memory usage. If you need
to process a lot of data and want to ignore/collect errors please consider
using the :func:`~elasticsearch.helpers.streaming_bulk` helper which will
just return the errors and not store them in memory.
:arg client: instance of :class:`~elasticsearch.Elasticsearch` to use
:arg actions: iterator containing the actions
:arg stats_only: if `True` only report number of successful/failed
operations instead of just number of successful and a list of error responses
:arg ignore_status: list of HTTP status code that you want to ignore
Any additional keyword arguments will be passed to
:func:`~elasticsearch.helpers.streaming_bulk` which is used to execute
the operation, see :func:`~elasticsearch.helpers.streaming_bulk` for more
accepted parameters.
"""
success, failed = 0, 0
# list of errors to be collected is not stats_only
errors = []
# make streaming_bulk yield successful results so we can count them
kwargs["yield_ok"] = True
for ok, item in streaming_bulk(
client, actions, ignore_status=ignore_status, *args, **kwargs
):
# go through request-response pairs and detect failures
if not ok:
if not stats_only:
errors.append(item)
failed += 1
else:
success += 1
return success, failed if stats_only else errors
def parallel_bulk(
client,
actions,
thread_count=4,
chunk_size=500,
max_chunk_bytes=100 * 1024 * 1024,
queue_size=4,
expand_action_callback=expand_action,
ignore_status=(),
*args,
**kwargs
):
"""
Parallel version of the bulk helper run in multiple threads at once.
:arg client: instance of :class:`~elasticsearch.Elasticsearch` to use
:arg actions: iterator containing the actions
:arg thread_count: size of the threadpool to use for the bulk requests
:arg chunk_size: number of docs in one chunk sent to es (default: 500)
:arg max_chunk_bytes: the maximum size of the request in bytes (default: 100MB)
:arg raise_on_error: raise ``BulkIndexError`` containing errors (as `.errors`)
from the execution of the last chunk when some occur. By default we raise.
:arg raise_on_exception: if ``False`` then don't propagate exceptions from
call to ``bulk`` and just report the items that failed as failed.
:arg expand_action_callback: callback executed on each action passed in,
should return a tuple containing the action line and the data line
(`None` if data line should be omitted).
:arg queue_size: size of the task queue between the main thread (producing
chunks to send) and the processing threads.
:arg ignore_status: list of HTTP status code that you want to ignore
"""
# Avoid importing multiprocessing unless parallel_bulk is used
# to avoid exceptions on restricted environments like App Engine
from multiprocessing.pool import ThreadPool
actions = map(expand_action_callback, actions)
class BlockingPool(ThreadPool):
def _setup_queues(self):
super(BlockingPool, self)._setup_queues() # type: ignore
# The queue must be at least the size of the number of threads to
# prevent hanging when inserting sentinel values during teardown.
self._inqueue = Queue(max(queue_size, thread_count))
self._quick_put = self._inqueue.put
pool = BlockingPool(thread_count)
try:
for result in pool.imap(
lambda bulk_chunk: list(
_process_bulk_chunk(
client,
bulk_chunk[1],
bulk_chunk[0],
ignore_status=ignore_status,
*args,
**kwargs
)
),
_chunk_actions(
actions, chunk_size, max_chunk_bytes, client.transport.serializer
),
):
for item in result:
yield item
finally:
pool.close()
pool.join()
def scan(
client,
query=None,
scroll="5m",
raise_on_error=True,
preserve_order=False,
size=1000,
request_timeout=None,
clear_scroll=True,
scroll_kwargs=None,
**kwargs
):
"""
Simple abstraction on top of the
:meth:`~elasticsearch.Elasticsearch.scroll` api - a simple iterator that
yields all hits as returned by underlining scroll requests.
By default scan does not return results in any pre-determined order. To
have a standard order in the returned documents (either by score or
explicit sort definition) when scrolling, use ``preserve_order=True``. This
may be an expensive operation and will negate the performance benefits of
using ``scan``.
:arg client: instance of :class:`~elasticsearch.Elasticsearch` to use
:arg query: body for the :meth:`~elasticsearch.Elasticsearch.search` api
:arg scroll: Specify how long a consistent view of the index should be
maintained for scrolled search
:arg raise_on_error: raises an exception (``ScanError``) if an error is
encountered (some shards fail to execute). By default we raise.
:arg preserve_order: don't set the ``search_type`` to ``scan`` - this will
cause the scroll to paginate with preserving the order. Note that this
can be an extremely expensive operation and can easily lead to
unpredictable results, use with caution.
:arg size: size (per shard) of the batch send at each iteration.
:arg request_timeout: explicit timeout for each call to ``scan``
:arg clear_scroll: explicitly calls delete on the scroll id via the clear
scroll API at the end of the method on completion or error, defaults
to true.
:arg scroll_kwargs: additional kwargs to be passed to
:meth:`~elasticsearch.Elasticsearch.scroll`
Any additional keyword arguments will be passed to the initial
:meth:`~elasticsearch.Elasticsearch.search` call::
scan(es,
query={"query": {"match": {"title": "python"}}},
index="orders-*",
doc_type="books"
)
"""
scroll_kwargs = scroll_kwargs or {}
_add_helper_meta_to_kwargs(scroll_kwargs, "s")
if not preserve_order:
query = query.copy() if query else {}
query["sort"] = "_doc"
# initial search
resp = client.search(
body=query, scroll=scroll, size=size, request_timeout=request_timeout, **kwargs
)
scroll_id = resp.get("_scroll_id")
try:
while scroll_id and resp["hits"]["hits"]:
for hit in resp["hits"]["hits"]:
yield hit
# Default to 0 if the value isn't included in the response
shards_successful = resp["_shards"].get("successful", 0)
shards_skipped = resp["_shards"].get("skipped", 0)
shards_total = resp["_shards"].get("total", 0)
# check if we have any errors
if (shards_successful + shards_skipped) < shards_total:
shards_message = "Scroll request has only succeeded on %d (+%d skipped) shards out of %d."
logger.warning(
shards_message,
shards_successful,
shards_skipped,
shards_total,
)
if raise_on_error:
raise ScanError(
scroll_id,
shards_message
% (
shards_successful,
shards_skipped,
shards_total,
),
)
resp = client.scroll(
body={"scroll_id": scroll_id, "scroll": scroll}, **scroll_kwargs
)
scroll_id = resp.get("_scroll_id")
finally:
if scroll_id and clear_scroll:
client.clear_scroll(
body={"scroll_id": [scroll_id]},
ignore=(404,),
params={"__elastic_client_meta": (("h", "s"),)},
)
def reindex(
client,
source_index,
target_index,
query=None,
target_client=None,
chunk_size=500,
scroll="5m",
scan_kwargs={},
bulk_kwargs={},
):
"""
Reindex all documents from one index that satisfy a given query
to another, potentially (if `target_client` is specified) on a different cluster.
If you don't specify the query you will reindex all the documents.
Since ``2.3`` a :meth:`~elasticsearch.Elasticsearch.reindex` api is
available as part of elasticsearch itself. It is recommended to use the api
instead of this helper wherever possible. The helper is here mostly for
backwards compatibility and for situations where more flexibility is
needed.
.. note::
This helper doesn't transfer mappings, just the data.
:arg client: instance of :class:`~elasticsearch.Elasticsearch` to use (for
read if `target_client` is specified as well)
:arg source_index: index (or list of indices) to read documents from
:arg target_index: name of the index in the target cluster to populate
:arg query: body for the :meth:`~elasticsearch.Elasticsearch.search` api
:arg target_client: optional, is specified will be used for writing (thus
enabling reindex between clusters)
:arg chunk_size: number of docs in one chunk sent to es (default: 500)
:arg scroll: Specify how long a consistent view of the index should be
maintained for scrolled search
:arg scan_kwargs: additional kwargs to be passed to
:func:`~elasticsearch.helpers.scan`
:arg bulk_kwargs: additional kwargs to be passed to
:func:`~elasticsearch.helpers.bulk`
"""
target_client = client if target_client is None else target_client
docs = scan(client, query=query, index=source_index, scroll=scroll, **scan_kwargs)
def _change_doc_index(hits, index):
for h in hits:
h["_index"] = index
if "fields" in h:
h.update(h.pop("fields"))
yield h
kwargs = {"stats_only": True}
kwargs.update(bulk_kwargs)
return bulk(
target_client,
_change_doc_index(docs, target_index),
chunk_size=chunk_size,
**kwargs
)