import copy
import datetime
import functools
import itertools
import json
# setup ES logging
import logging
import re
import time
from importlib import import_module
from typing import List, Mapping, Optional
from elasticsearch import (
Elasticsearch,
ElasticsearchException,
NotFoundError,
RequestError,
TransportError,
helpers,
)
from elasticsearch.serializer import JSONSerializer
from biothings.utils.common import inf, iter_n, merge, nan, splitstr, traverse
# the following imports used by utils.es.Database
from biothings.utils.dataload import dict_walk, update_dict_recur
from biothings.utils.dotfield import parse_dot_fields
from biothings.utils.hub_db import IDatabase
from biothings.utils.serializer import to_json
formatter = logging.Formatter("%(levelname)s:%(name)s:%(message)s")
es_logger = logging.getLogger("elasticsearch")
es_logger.setLevel(logging.WARNING)
ch = logging.StreamHandler()
ch.setFormatter(formatter)
es_logger.addHandler(ch)
es_tracer = logging.getLogger("elasticsearch.trace")
es_tracer.setLevel(logging.WARNING)
ch = logging.StreamHandler()
ch.setFormatter(formatter)
es_tracer.addHandler(ch)
[docs]def verify_ids(
doc_iter,
es_host,
index,
doc_type=None,
step=100000,
):
"""verify how many docs from input interator/list overlapping with existing docs."""
index = index
doc_type = doc_type
es = get_es(es_host)
q = {"query": {"ids": {"values": []}}}
total_cnt = 0
found_cnt = 0
out = []
for doc_batch in iter_n(doc_iter, n=step):
id_li = [doc["_id"] for doc in doc_batch]
# id_li = [doc['_id'].replace('chr', '') for doc in doc_batch]
q["query"]["ids"]["values"] = id_li
xres = es.search(index=index, doc_type=doc_type, body=q, _source=False)
found_cnt += xres["hits"]["total"]
total_cnt += len(id_li)
out.extend([x["_id"] for x in xres["hits"]["hits"]])
return out
[docs]def get_es(es_host, timeout=120, max_retries=3, retry_on_timeout=False):
es = Elasticsearch(
es_host, timeout=timeout, max_retries=max_retries, retry_on_timeout=retry_on_timeout
)
return es
# WARNING: this wrapper changes the _index and _doc_type
# usually this is unwanted but we will leave it this way here
[docs]def wrapper(func):
"""this wrapper allows passing index and doc_type from wrapped method."""
def outter_fn(*args, **kwargs):
self = args[0]
index = kwargs.pop("index", self._index) # pylint: disable=protected-access
doc_type = kwargs.pop("doc_type", self._doc_type) # pylint: disable=protected-access
self._index = index # pylint: disable=protected-access
self._doc_type = doc_type # pylint: disable=protected-access
return func(*args, **kwargs)
outter_fn.__doc__ = func.__doc__
return outter_fn
[docs]class IndexerException(Exception):
pass
[docs]class ESIndex:
"""An Elasticsearch Index Wrapping A Client.
Counterpart for pymongo.collection.Collection"""
# a new approach to biothings.utils.es.ESIndexer
# but not intended to be a replacement in features.
def __init__(self, client, index_name):
self.client = client # synchronous
self.index_name = index_name # MUST exist
@property
@functools.lru_cache()
def doc_type(self):
if int(self.client.info()["version"]["number"].split(".")[0]) < 7:
mappings = self.client.indices.get_mapping(self.index_name)
mappings = mappings[self.index_name]["mappings"]
return next(iter(mappings.keys()))
return None
# SUBCLASS NOTE &&
# BEFORE YOU ADD A METHOD UNDER THIS CLASS:
# An object of this class refers to an existing ES index. All operations
# should target this index. Do not put uncommon methods here. They belong
# to the subclasses. This class is to provide a common framework to support
# Index-specific ES operations, an concept does not exist in the low-level
# ES library, thus only providing low-level common operations, like doc_type
# parsing across ES versions for biothings usecases. (single type per index)
[docs]class ESIndexer:
# RETIRING THIS CLASS
# --
# Since we don't always directly work on
# existing actual indices, the index referred here
# can be an alias or does not exist.
def __init__(
self,
index,
doc_type="_doc",
es_host="localhost:9200",
step=500,
step_size=10, # elasticsearch.helpers.bulk
number_of_shards=1,
number_of_replicas=0,
check_index=True,
**kwargs,
):
self.es_host = es_host
self._es = get_es(es_host, **kwargs)
self._host_major_ver = int(self._es.info()["version"]["number"].split(".")[0])
# the name of the index when ESIndexer is initialized
self.canonical_index_name = index
self._index = index # placeholder, will be updated later
if check_index:
# if index is actually an alias, resolve the alias to
# the real underlying index
self.check_index()
self._doc_type = None
if doc_type:
self._doc_type = doc_type
else:
# assuming index exists, get mapping to discover doc_type
try:
m = self.get_mapping()
assert len(m) == 1, "Expected only one doc type, got: %s" % m.keys()
self._doc_type = list(m).pop()
except Exception as e: # pylint: disable=broad-except
if check_index:
logging.info("Failed to guess doc_type: %s", e)
self.number_of_shards = number_of_shards # set number_of_shards when create_index
self.number_of_replicas = int(
number_of_replicas
) # set number_of_replicas when create_index
self.step = step or 500 # the bulk size when doing bulk operation.
self.step_size = (step_size or 10) * 1048576 # MB -> bytes
self.s = None # number of records to skip, useful to continue indexing after an error.
[docs] def check_index(self):
"""
Check if index is an alias, and update self._index to point to
actual index
TODO: the overall design of ESIndexer is not great. If we are exposing ES
implementation details (such as the abilities to create and delete indices,
create and update aliases, etc.) to the user of this Class, then this method
doesn't seem that out of place.
"""
try:
res = self._es.indices.get_alias(name=self.canonical_index_name)
# this is an alias
if len(res) != 1:
raise RuntimeError(
f"Alias '{self.canonical_index_name}' does not "
"associate with exactly 1 index"
)
self._index = list(res.keys())[0]
except NotFoundError:
# probably intended to be an index name
self._index = self.canonical_index_name
[docs] @wrapper
def get_biothing(self, bid, only_source=False, **kwargs):
rawdoc = self._es.get(index=self._index, id=bid, doc_type=self._doc_type, **kwargs)
if not only_source:
return rawdoc
else:
doc = rawdoc["_source"]
doc["_id"] = rawdoc["_id"]
return doc
[docs] @wrapper
def exists(self, bid):
"""return True/False if a biothing id exists or not."""
try:
doc = self.get_biothing(bid, stored_fields=None)
return doc["found"]
except NotFoundError:
return False
[docs] @wrapper
def mexists(self, bid_list):
q = {"query": {"ids": {"values": bid_list}}}
res = self._es.search(
index=self._index,
doc_type=self._doc_type,
body=q,
stored_fields=None,
size=len(bid_list),
)
# id_set = set([doc['_id'] for doc in res['hits']['hits']]) # TODO: Confirm this line
id_set = {doc["_id"] for doc in res["hits"]["hits"]}
return [(bid, bid in id_set) for bid in bid_list]
[docs] @wrapper
def count(self, q=None, raw=False):
try:
_res = self._es.count(index=self._index, doc_type=self._doc_type, body=q)
return _res if raw else _res["count"]
except NotFoundError:
return None
[docs] @wrapper
def count_src(self, src):
if isinstance(src, str):
src = [src]
cnt_d = {}
for _src in src:
q = {"query": {"constant_score": {"filter": {"exists": {"field": _src}}}}}
cnt_d[_src] = self.count(q)
return cnt_d
[docs] @wrapper
def create_index(self, mapping=None, extra_settings=None):
if not self._es.indices.exists(index=self._index):
body = {
"settings": {
"number_of_shards": self.number_of_shards,
"number_of_replicas": self.number_of_replicas,
}
}
extra_settings = extra_settings or {}
body["settings"].update(extra_settings)
if mapping:
# the mapping is passed in for elasticsearch 6
# if the remote server is of elasticsearch version 7 or later
# drop the doc_type first level key as it is no longer supported
self._populate_es_version()
if self._host_major_ver > 6:
if len(mapping) == 1 and next(iter(mapping)) not in (
"properties",
"dynamic",
"_meta",
):
mapping = next(iter(mapping.values()))
mapping = {"mappings": mapping}
body.update(mapping)
self._es.indices.create(index=self._index, **body)
def _populate_es_version(self):
if not hasattr(self, "_es_version"):
self._es_version = int(self._es.info()["version"]["number"].split(".")[0])
[docs] def exists_index(self, index: Optional[str] = None):
if not index:
index = self._index
return self._es.indices.exists(index)
[docs] def index(self, doc, id=None, action="index"): # pylint: disable=redefined-builtin
"""add a doc to the index. If id is not None, the existing doc will be
updated.
"""
self._es.index(
index=self._index, doc_type=self._doc_type, body=doc, id=id, params={"op_type": action}
)
[docs] def index_bulk(self, docs, step=None, action="index"):
self._populate_es_version()
index_name = self._index
doc_type = self._doc_type
step = step or self.step
def _get_bulk(doc):
# keep original doc
ndoc = copy.copy(doc)
ndoc.update(
{
"_index": index_name,
"_type": doc_type,
"_op_type": action,
}
)
if self._host_major_ver > 6:
ndoc.pop("_type")
return ndoc
actions = (_get_bulk(doc) for doc in docs)
num_ok, errors = helpers.bulk(
self._es, actions, chunk_size=step, max_chunk_bytes=self.step_size
)
if errors:
raise ElasticsearchException(
"%d errors while bulk-indexing: %s" % (len(errors), [str(e) for e in errors])
)
return num_ok, errors
[docs] def delete_doc(self, id): # pylint: disable=redefined-builtin
"""delete a doc from the index based on passed id."""
return self._es.delete(index=self._index, doc_type=self._doc_type, id=id)
[docs] def delete_docs(self, ids, step=None):
"""delete a list of docs in bulk."""
index_name = self._index
doc_type = self._doc_type
step = step or self.step
def _get_bulk(_id):
if self._host_major_ver >= 7:
doc = {"_op_type": "delete", "_index": index_name, "_id": _id}
else:
doc = {"_op_type": "delete", "_index": index_name, "_type": doc_type, "_id": _id}
return doc
actions = (_get_bulk(_id) for _id in ids)
return helpers.bulk(
self._es, actions, chunk_size=step, stats_only=True, raise_on_error=False
)
[docs] def delete_index(self, index=None):
if not index:
index = self._index
self._es.indices.delete(index)
[docs] def update(self, id, extra_doc, upsert=True): # pylint: disable=redefined-builtin
"""update an existing doc with extra_doc.
allow to set upsert=True, to insert new docs.
"""
body = {"doc": extra_doc}
if upsert:
body["doc_as_upsert"] = True
return self._es.update(index=self._index, doc_type=self._doc_type, id=id, body=body)
[docs] def update_docs(self, partial_docs, upsert=True, step=None, **kwargs):
"""update a list of partial_docs in bulk.
allow to set upsert=True, to insert new docs.
"""
index_name = self._index
doc_type = self._doc_type
step = step or self.step
def _get_bulk(doc):
if self._host_major_ver >= 7:
doc = {"_op_type": "update", "_index": index_name, "_id": doc["_id"], "doc": doc}
else:
doc = {
"_op_type": "update",
"_index": index_name,
"_type": doc_type,
"_id": doc["_id"],
"doc": doc,
}
if upsert:
doc["doc_as_upsert"] = True
return doc
actions = (_get_bulk(doc) for doc in partial_docs)
return helpers.bulk(self._es, actions, chunk_size=step, **kwargs)
[docs] def get_mapping(self):
"""return the current index mapping"""
if self._host_major_ver <= 6:
m = self._es.indices.get_mapping(
index=self._index,
doc_type=self._doc_type,
)
return m[self._index]["mappings"]
elif self._host_major_ver <= 8:
m = self._es.indices.get_mapping(index=self._index)
# fake the mapping doc_type
m = {self._doc_type: m[self._index]["mappings"]}
return m
else:
raise RuntimeError(
f"Server Elasticsearch version is {self._host_major_ver} "
"which is unsupported when using old ESIndexer class"
)
[docs] def update_mapping(self, m):
if self._host_major_ver <= 6:
assert list(m) == [
self._doc_type
], "Bad mapping format, should have one doc_type, got: %s" % list(m)
assert "properties" in m[self._doc_type], "Bad mapping format, no 'properties' key"
return self._es.indices.put_mapping(index=self._index, doc_type=self._doc_type, body=m)
elif self._host_major_ver <= 8:
# this is basically guessing based on heuristics
if len(m) == 1:
if "properties" not in m: # basically {'_doc': mapping}
m = next(iter(m.values())) # take out the mapping
else: # basically just the mapping or type: properties
pass # ignoring the possibility of doc_type='properties'
else: # we don't expect {} as input, so len(m) > 1
if "properties" not in m:
raise RuntimeError(f"Possibly invalid mapping {m}")
return self._es.indices.put_mapping(index=self._index, body=m)
else:
raise RuntimeError(
f"Server Elasticsearch version is {self._host_major_ver} "
"which is unsupported when using old ESIndexer class"
)
[docs] @wrapper
def build_index(
self, collection, verbose=True, query=None, bulk=True, update=False, allow_upsert=True
):
index_name = self._index
# update some settings for bulk indexing
body = {
"index": {
"refresh_interval": "-1", # disable refresh temporarily
"auto_expand_replicas": "0-all",
}
}
self._es.indices.put_settings(body=body, index=index_name)
try:
self._build_index_sequential(
collection, verbose, query=query, bulk=bulk, update=update, allow_upsert=True
)
finally:
# restore some settings after bulk indexing is done.
body = {"index": {"refresh_interval": "1s"}} # default settings
self._es.indices.put_settings(body=body, index=index_name)
try:
self._es.indices.flush()
self._es.indices.refresh()
except: # pylint: disable=bare-except # noqa
pass
time.sleep(1)
src_cnt = collection.count(query)
es_cnt = self.count()
if src_cnt != es_cnt:
raise IndexerException(
"Total count of documents does not match [{}, should be {}]".format(
es_cnt, src_cnt
)
)
return es_cnt
def _build_index_sequential(
self, collection, verbose=False, query=None, bulk=True, update=False, allow_upsert=True
):
def rate_control(cnt, t):
delay = 0
if t > 90:
delay = 30
elif t > 60:
delay = 10
if delay:
time.sleep(delay)
from biothings.utils.mongo import doc_feeder
src_docs = doc_feeder(
collection, step=self.step, s=self.s, batch_callback=rate_control, query=query
)
if bulk:
if update:
# input doc will update existing one
# if allow_upsert, create new one if not exist
res = self.update_docs(src_docs, upsert=allow_upsert)
else:
# input doc will overwrite existing one
res = self.index_bulk(src_docs)
# if len(res[1]) > 0:
if res[1]:
raise IndexerException("Error: {} docs failed indexing.".format(len(res[1])))
return res[0]
else:
cnt = 0
for doc in src_docs:
self.index(doc)
cnt += 1
return cnt
[docs] @wrapper
def optimize(self, max_num_segments=1):
"""optimize the default index."""
params = {
"wait_for_merge": False,
"max_num_segments": max_num_segments,
}
return self._es.indices.forcemerge(index=self._index, params=params)
[docs] def clean_field(self, field, dryrun=True, step=5000):
"""remove a top-level field from ES index, if the field is the only field of the doc,
remove the doc as well.
step is the size of bulk update on ES
try first with dryrun turned on, and then perform the actual updates with dryrun off.
"""
if self._host_major_ver >= 7:
raise RuntimeError("clean_field is no longer supported")
q = {"query": {"constant_score": {"filter": {"exists": {"field": field}}}}}
cnt_orphan_doc = 0
cnt = 0
_li = []
for doc in self.doc_feeder(query=q):
if set(doc) == {"_id", field}:
cnt_orphan_doc += 1
# delete orphan doc
_li.append(
{"delete": {"_index": self._index, "_type": self._doc_type, "_id": doc["_id"]}}
)
else:
# otherwise, just remove the field from the doc
_li.append(
{"update": {"_index": self._index, "_type": self._doc_type, "_id": doc["_id"]}}
)
# this script update requires "script.disable_dynamic: false" setting
# in elasticsearch.yml
_li.append({"script": 'ctx._source.remove("{}")'.format(field)})
cnt += 1
if len(_li) == step:
if not dryrun:
self._es.bulk(body=_li)
_li = []
if _li:
if not dryrun:
self._es.bulk(body=_li)
return {"total": cnt, "updated": cnt - cnt_orphan_doc, "deleted": cnt_orphan_doc}
[docs] @wrapper
def doc_feeder_using_helper(self, step=None, verbose=True, query=None, scroll="10m", **kwargs):
# verbose unimplemented
step = step or self.step
q = query if query else {"query": {"match_all": {}}}
for rawdoc in helpers.scan(
client=self._es,
query=q,
scroll=scroll,
index=self._index,
doc_type=self._doc_type,
**kwargs,
):
if rawdoc.get("_source", False):
doc = rawdoc["_source"]
doc["_id"] = rawdoc["_id"]
yield doc
else:
yield rawdoc
[docs] @wrapper
def doc_feeder(
self, step=None, verbose=True, query=None, scroll="10m", only_source=True, **kwargs
):
step = step or self.step
q = query if query else {"query": {"match_all": {}}}
_q_cnt = self.count(q=q, raw=True)
n = _q_cnt["count"]
n_shards = _q_cnt["_shards"]["total"]
assert n_shards == _q_cnt["_shards"]["successful"]
# Not sure if scroll size is per shard anymore in the new ES...should check this
_size = int(step / n_shards)
assert _size * n_shards == step
cnt = 0
# t0 = time.time()
# if verbose:
# t1 = time.time()
res = self._es.search(
index=self._index,
doc_type=self._doc_type,
body=q,
size=_size,
search_type="scan",
scroll=scroll,
**kwargs,
)
# double check initial scroll request returns no hits
# assert len(res['hits']['hits']) == 0
assert not res["hits"]["hits"]
while True:
# if verbose:
# t1 = time.time()
res = self._es.scroll(scroll_id=res["_scroll_id"], scroll=scroll)
# if len(res['hits']['hits']) == 0:
if not res["hits"]["hits"]:
break
else:
for rawdoc in res["hits"]["hits"]:
if rawdoc.get("_source", False) and only_source:
doc = rawdoc["_source"]
doc["_id"] = rawdoc["_id"]
yield doc
else:
yield rawdoc
cnt += 1
assert (
cnt == n
), "Error: scroll query terminated early [{}, {}], please retry.\nLast response:\n{}".format(
cnt, n, res
)
[docs] @wrapper
def get_id_list(self, step=None, verbose=True):
step = step or self.step
cur = self.doc_feeder(step=step, _source=False, verbose=verbose)
for doc in cur:
yield doc["_id"]
[docs] @wrapper
def get_docs(self, ids, step=None, only_source=True, **mget_args):
"""Return matching docs for given ids iterable, if not found return None.
A generator is returned to the matched docs. If only_source is False,
the entire document is returned, otherwise only the source is returned."""
# chunkify
step = step or self.step
for chunk in iter_n(ids, step):
if self._host_major_ver > 6:
chunk_res = self._es.mget(body={"ids": chunk}, index=self._index, **mget_args)
else:
chunk_res = self._es.mget(
body={"ids": chunk}, index=self._index, doc_type=self._doc_type, **mget_args
)
for rawdoc in chunk_res["docs"]:
if ("found" not in rawdoc) or (("found" in rawdoc) and not rawdoc["found"]):
continue
elif not only_source:
yield rawdoc
else:
doc = rawdoc["_source"]
doc["_id"] = rawdoc["_id"]
yield doc
[docs] def find_biggest_doc(
self, fields_li, min=5, return_doc=False
): # pylint: disable=redefined-builtin
"""return the doc with the max number of fields from fields_li."""
for n in range(len(fields_li), min - 1, -1):
for field_set in itertools.combinations(fields_li, n):
q = " AND ".join(["_exists_:" + field for field in field_set])
q = {"query": {"query_string": {"query": q}}}
cnt = self.count(q)
if cnt > 0:
if return_doc:
res = self._es.search(
index=self._index, doc_type=self._doc_type, body=q, size=cnt
)
return res
else:
return (cnt, q)
[docs] def snapshot(self, repo, snapshot, mode=None, **params):
body = {
"indices": self._index,
"include_global_state": False
# there is no reason to include global state in our application
# we want to separate the staging env from the production env
# (global state includes index templates and ingest pipeline)
# but this doesn't mean this setting has to be here
# maybe move this line to where it belongs later
}
if mode == "purge":
# Note: this works, just for small one when deletion is done instantly
try:
self._es.snapshot.get(repo, snapshot)
# if we can get it, we have to delete it
self._es.snapshot.delete(repo, snapshot)
except NotFoundError:
# ok, nothing to delete/purge
pass
try:
return self._es.snapshot.create(repo, snapshot, body=body, params=params)
except RequestError as e:
try:
err_msg = e.info["error"]["reason"]
except KeyError:
err_msg = e.error
raise IndexerException("Can't snapshot '%s': %s" % (self._index, err_msg))
[docs] def restore(self, repo_name, snapshot_name, index_name=None, purge=False):
indices = self.get_indices_from_snapshots(repo_name, snapshot_name)
if not index_name:
index_name = indices[0]
indices = ",".join(indices)
if purge:
try:
self._es.indices.get(index=index_name)
# if we get there, it exists, delete it
self._es.indices.delete(index=index_name)
except NotFoundError:
# no need to delete it,
pass
try:
# this is just about renaming index within snapshot to index_name
body = {
"indices": indices,
"rename_replacement": index_name,
"ignore_unavailable": True,
"rename_pattern": "(.+)",
# set to False, snapshots were created without global state anyway.
# In ES8, an error will be raised if set to True
"include_global_state": False,
}
return self._es.snapshot.restore(repo_name, snapshot_name, body=body)
except TransportError as e:
raise IndexerException(
"Can't restore snapshot '%s' (does index '%s' already exist ?): %s"
% (snapshot_name, index_name, e)
)
[docs] def get_alias(self, index: str = None, alias_name: str = None) -> List[str]:
"""
Get indices with alias associated with given index name or alias name
Args:
index: name of index
alias_name: name of alias
Returns:
Mapping of index names with their aliases
"""
return self._es.indices.get_alias(index=index, name=alias_name)
[docs] def get_settings(self, index: str = None) -> Mapping[str, Mapping]:
"""
Get indices with settings associated with given index name
Args:
index: name of index
Returns:
Mapping of index names with their settings
"""
return self._es.indices.get_settings(index=index)
[docs] def get_indice_names_by_settings(
self, index: str = None, sort_by_creation_date=False, reverse=False
) -> List[str]:
"""
Get list of indices names associated with given index name, using indices' settings
Args:
index: name of index
sort_by_creation_date: sort the result by indice's creation_date
reverse: control the direction of the sorting
Returns:
list of index names (str)
"""
indices_settings = self.get_settings(index)
names_with_creation_date = [
(indice_name, setting["settings"]["index"]["creation_date"])
for indice_name, setting in indices_settings.items()
]
if sort_by_creation_date:
names_with_creation_date = sorted(
names_with_creation_date,
key=lambda name_with_creation_date: name_with_creation_date[1],
reverse=reverse,
)
indice_names = [name for name, _ in names_with_creation_date]
return indice_names
[docs] def update_alias(self, alias_name: str, index: Optional[str] = None):
"""
Create or update an ES alias pointing to an index
Creates or updates an alias in Elasticsearch, associated with the
given index name or the underlying index of the ESIndexer instance.
When the alias name does not exist, it will be created. If an existing
alias already exists, it will be updated to only associate with the
index.
When the alias name already exists, an exception will be raised,
UNLESS the alias name is the same as index name that the ESIndexer is
initialized with. In this case, the existing index with the name
collision will be deleted, and the alias will be created
in its place. This feature is intended for seamless migration from
an index to an alias associated with an index for zero-downtime
installs.
Args:
alias_name: name of the alias
index: name of the index to associate with alias. If None, the
index of the ESIndexer instance is used.
Raises:
IndexerException
"""
if index is None:
index = self._index
if not self._es.indices.exists(alias_name):
# case 1 it does not already exist
# create the alias pointing to _index
self._es.indices.put_alias(index=index, name=alias_name)
else: # case 2 it already exists
if self._es.indices.exists_alias(name=alias_name):
# if it is an alias, blindly update
# This removes any other indices associated with the alias
actions = {"actions": [{"add": {"index": index, "alias": alias_name}}]}
removes = [
{"remove": {"index": index_name, "alias": alias_name}}
for index_name in self.get_alias(alias_name)
]
actions["actions"].extend(removes)
self._es.indices.update_aliases(body=actions)
else: # it is an index
# if not _index and is the canonical index name
# then delete the index and create alias
if alias_name == self.canonical_index_name:
self._es.indices.delete(alias_name)
self._es.indices.put_alias(index=index, name=alias_name)
else:
raise IndexerException(
f"Cannot create alias {alias_name} "
"an index with the same name "
"already exists"
)
[docs] def get_repository(self, repo_name):
try:
return self._es.snapshot.get_repository(repo_name)
except NotFoundError:
raise IndexerException("Repository '%s' doesn't exist" % repo_name)
[docs] def create_repository(self, repo_name, settings):
try:
self._es.snapshot.create_repository(repo_name, settings)
except TransportError as e:
raise IndexerException("Can't create snapshot repository '%s': %s" % (repo_name, e))
[docs] def get_snapshots(self, repo_name, snapshot_name):
try:
snapshots = self._es.snapshot.get(repository=repo_name, snapshot=snapshot_name)
return [snapshot for snapshot in snapshots["snapshots"]]
except NotFoundError as ex:
message = str(ex)
if ex.error == "repository_missing_exception":
message = "Repository '%s' doesn't exist" % repo_name
if ex.error == "snapshot_missing_exception":
message = "Snapshot '%s' doesn't exist" % snapshot_name
raise IndexerException(message)
[docs] def get_indices_from_snapshots(self, repo_name, snapshot_name):
snapshots = self.get_snapshots(repo_name, snapshot_name)
indices = []
for snapshot in snapshots:
indices.extend([index for index in (snapshot.get("indices") or [])])
return indices
[docs] def get_snapshot_status(self, repo, snapshot):
return self._es.snapshot.status(repo, snapshot)
[docs] def get_restore_status(self, index_name=None):
index_name = index_name or self._index
recov = self._es.indices.recovery(index=index_name)
if index_name not in recov:
return {"status": "INIT", "progress": "0%"}
shards = recov[index_name]["shards"]
# get all shards status
shards_status = [s["stage"] for s in shards]
done = len([s for s in shards_status if s == "DONE"])
if set(shards_status) == {"DONE"}:
return {"status": "DONE", "progress": "100%"}
else:
return {
"status": "IN_PROGRESS",
"progress": "%.2f%%" % (done / len(shards_status) * 100),
}
[docs] def get_internal_number_of_replicas(self):
try:
index_settings = self._es.indices.get_settings(self._index)
return index_settings[self._index]["settings"]["index"]["number_of_replicas"]
except Exception:
return
[docs] def set_internal_number_of_replicas(self, number_of_replicas=None):
if not number_of_replicas:
number_of_replicas = self.number_of_replicas
settings = json.dumps({"index": {"number_of_replicas": number_of_replicas}})
self._es.indices.put_settings(settings, index=self._index)
[docs]class MappingError(Exception):
pass
[docs]def generate_es_mapping(inspect_doc, init=True, level=0):
"""Generate an ES mapping according to "inspect_doc", which is
produced by biothings.utils.inspect module"""
map_tpl = {
int: {"type": "integer"},
bool: {"type": "boolean"},
float: {"type": "float"},
str: {
"type": "keyword",
"normalizer": "keyword_lowercase_normalizer",
}, # not splittable (like an ID for instance)
splitstr: {"type": "text"},
}
# inspect_doc, if it's been jsonified, contains keys with type as string,
# such as "<class 'str'>". This is not a real type and we need to convert them
# back to actual types. This is transparent if inspect_doc isalready in proper format
pat = re.compile(r"<class '(\w+)'>")
def str2type(k):
if isinstance(k, str):
mat = pat.findall(k)
if mat:
return eval(mat[0]) # actual type
else:
return k
else:
return k
inspect_doc = dict_walk(inspect_doc, str2type)
mapping = {}
errors = []
none_type = type(None)
if init and "_id" not in inspect_doc:
errors.append("No _id key found, document won't be indexed. (doc: %s)" % inspect_doc)
for rootk in inspect_doc:
if rootk == "_id":
keys = list(inspect_doc[rootk].keys())
if str in keys and splitstr in keys:
keys.remove(str)
if not len(keys) == 1 or (keys[0] != str and keys[0] != splitstr):
errors.append("_id fields should all be a string type (got: %s)" % keys)
# it was just a check, it's not part of the mapping
continue
if rootk == "_stats":
continue
if isinstance(rootk, type(None)): # if rootk == type(None):
# value can be null, just skip it
continue
# some inspect report have True as value, others have dict (will all have dict eventually)
if inspect_doc[rootk] is True:
inspect_doc[rootk] = {}
keys = list(inspect_doc[rootk].keys())
# if dict, it can be a dict containing the type (no explore needed) or a dict
# containing more keys (explore needed)
if list in keys:
# we explore directly the list w/ inspect_doc[rootk][list] as param.
# (similar to skipping list type, as there's no such list type in ES mapping)
# carefull: there could be list of list, if which case we move further into the structure
# to skip them
toexplore = inspect_doc[rootk][list]
while list in toexplore:
toexplore = toexplore[list]
if len(toexplore) > 1:
# we want to make sure that, whatever the structure, the types involved were the same
# Exception: None is allowed with other types (translates to 'null' in ES)
# other_types = set([k for k in toexplore.keys() if k != list and isinstance(k, type) and k is not type(None)]) # TODO: Confirm this line
other_types = {
k
for k in toexplore.keys()
if k != list and isinstance(k, type) and not isinstance(k, none_type)
}
# some mixes are allowed by ES
if {int, float}.issubset(other_types):
other_types.discard(int) # float > int
toexplore.pop(int)
if len(other_types) > 1:
raise Exception("Mixing types for key '%s': %s" % (rootk, other_types))
res = generate_es_mapping(toexplore, init=False, level=level + 1)
# is it the only key or do we have more ? (ie. some docs have data as "x", some
# others have list("x")
# list was either a list of values (end of tree) or a list of dict. Depending
# on that, we add "properties" (when list of dict) or not (when list of values)
if type in set(map(type, inspect_doc[rootk][list])):
mapping[rootk] = res
else:
mapping[rootk] = {"properties": {}}
mapping[rootk]["properties"] = res
elif set(map(type, keys)) == {type}:
# it's a type declaration, no explore
# typs = list(map(type, [k for k in keys if k is not type(None)])) # TODO: Confirm this line
typs = list(map(type, [k for k in keys if not isinstance(k, none_type)]))
if len(typs) > 1:
errors.append("More than one type (key:%s,types:%s)" % (repr(rootk), repr(keys)))
try:
typ = list(inspect_doc[rootk].keys())
# ther can still be more than one type, if we have a None combined with
# the "correct" one. We allow None as a combined type, but we want to ignore
# it when we want to find the mapping
if len(typ) == 1:
typ = typ[0]
else:
# typ = [t for t in typ if t is not type(None)][0] # TODO: Confirm this line
typ = [t for t in typ if not isinstance(t, none_type)][0]
if typ is nan or typ is inf:
raise TypeError(typ)
mapping[rootk] = map_tpl[typ]
except KeyError:
errors.append("Can't find map type %s for key %s" % (inspect_doc[rootk], rootk))
except TypeError:
errors.append("Type %s for key %s isn't allowed in ES mapping" % (typ, rootk))
elif inspect_doc[rootk] == {}:
typ = rootk
return map_tpl[typ]
else:
mapping[rootk] = {"properties": {}}
mapping[rootk]["properties"] = generate_es_mapping(
inspect_doc[rootk], init=False, level=level + 1
)
if errors:
raise MappingError("Error while generating mapping", errors)
return mapping
# from biothings.utils.common import json_serial
# from biothings.utils.dataload import update_dict_recur
# from biothings.utils.dotfield import parse_dot_fields
########################
# ES as HUB DB backend #
########################
# from biothings.utils.hub_db import IDatabase
[docs]def get_hub_db_conn():
return Database()
[docs]def get_src_conn():
return get_hub_db_conn()
[docs]def get_src_dump():
db = Database()
return db[db.CONFIG.DATA_SRC_DUMP_COLLECTION]
[docs]def get_src_master():
db = Database()
return db[db.CONFIG.DATA_SRC_MASTER_COLLECTION]
[docs]def get_src_build():
db = Database()
return db[db.CONFIG.DATA_SRC_BUILD_COLLECTION]
[docs]def get_src_build_config():
db = Database()
return db[db.CONFIG.DATA_SRC_BUILD_CONFIG_COLLECTION]
[docs]def get_data_plugin():
db = Database()
return db[db.CONFIG.DATA_PLUGIN_COLLECTION]
[docs]def get_api():
db = Database()
return db[db.CONFIG.API_COLLECTION]
[docs]def get_cmd():
db = Database()
return db[db.CONFIG.CMD_COLLECTION]
[docs]def get_event():
db = Database()
return db[db.CONFIG.EVENT_COLLECTION]
[docs]def get_hub_config():
db = Database()
return db[getattr(db.CONFIG, "HUB_CONFIG_COLLECTION", "hub_config")]
[docs]def get_source_fullname(col_name):
return col_name
[docs]def get_last_command():
cmds = list(sorted(get_cmd()._read().values(), key=lambda cmd: cmd["_id"]))
return cmds[-1] if cmds else None
# ES7+ FOR HUB_DB
# IS *NOT* DESIGNED FOR
# MANAGING HEAVY WORKLOADS
[docs]class Database(IDatabase):
CONFIG = None # will be set by bt.utils.hub_db.setup()
def __init__(self):
super(Database, self).__init__()
self.name = self.CONFIG.DATA_HUB_DB_DATABASE
self.host = self.CONFIG.HUB_DB_BACKEND["host"]
self.client = Elasticsearch(self.host, serializer=_HubDBEncoder())
if not self.client.indices.exists(index=self.name):
self.client.indices.create(
index=self.name,
settings={
"number_of_shards": 1,
"number_of_replicas": 0,
},
mappings={"enabled": False},
)
@property
def address(self):
return self.host
# ES API OPS
# ON "COLLECTION"
def _exists(self, _id):
return self.client.exists(index=self.name, id=_id)
def _read(self, _id):
doc = self.client.get(index=self.name, id=_id)
return doc["_source"]
def _write(self, _id, doc):
assert doc.pop("_id", None) in (_id, None)
self.client.index(index=self.name, document=doc, id=_id)
self.client.indices.refresh(index=self.name)
def _modify(self, _id, func):
doc = self._read(_id)
doc = func(doc) or doc
self._write(_id, doc)
# HIGH LEVEL
# HUB_DB ABSTRACTION
[docs] def create_collection(self, colname):
return self[colname]
def __getitem__(self, colname):
return Collection(colname, self)
[docs]class Collection:
def __init__(self, colname, db):
self.name = colname
self.db = db
if not self.db._exists(colname):
self._write({})
# COLLECTION OPS
def _read(self):
return self.db._read(self.name)
def _write(self, col):
self.db._write(self.name, col)
# COLLECTION DOC OPS
def _exists_one(self, _id):
return str(_id) in self._read()
def _write_one(self, doc):
def func(collection):
collection[str(doc["_id"])] = doc
self.db._modify(self.name, func)
# HUB_DB ABSTRACTION
def __getitem__(self, _id):
return self.find_one({"_id": _id})
[docs] def find_one(self, *args, **kwargs):
results = self.find(*args, **kwargs)
return results[0] if results else None
[docs] def find(self, filter=None, projection=None, *args, **kwargs):
if args or kwargs:
raise NotImplementedError()
results = []
logger = logging.getLogger(__name__)
for doc in self._read().values():
_doc = dict(traverse(doc)) # dotdict
_doc.update(dict(traverse(doc, True)))
for k, v in (filter or {}).items():
if isinstance(v, dict) and "$exists" in v:
logger.error("Ignored filter: {'%s': %s}", k, v)
continue
if _doc.get(k) != v:
break
else: # no break
results.append(_pyobj(doc))
if projection: # used by BuildManager.build_info
logger.error("Ignored projection: %s", projection)
return results
[docs] def insert_one(self, document, *args, **kwargs):
if args or kwargs:
raise NotImplementedError()
if self._exists_one(document["_id"]):
raise ValueError("Document already exists.")
self._write_one(document)
[docs] def replace_one(self, filter, replacement, upsert=False, *args, **kwargs):
if args or kwargs:
raise NotImplementedError()
doc = self.find_one(filter) or {}
if not (doc or upsert):
raise ValueError("No Match.")
_id = doc.get("_id") or filter["_id"]
replacement["_id"] = _id
self._write_one(replacement)
# update operations support
# a subset of mongo operators
# for partial document editing
def _update_one(self, doc, update, *args, **kwargs):
if args or kwargs:
raise NotImplementedError()
if not len(update) == 1:
raise ValueError("Invalid operator.")
if next(iter(update)) not in ("$set", "$unset", "$push", "$addToSet", "$pull"):
raise NotImplementedError(next(iter(update)))
# https://docs.mongodb.com/manual/reference/operator/update/set/
# https://docs.mongodb.com/manual/reference/operator/update/unset/
# https://docs.mongodb.com/manual/reference/operator/update/push/
# https://docs.mongodb.com/manual/reference/operator/update/addToSet/
# https://docs.mongodb.com/manual/reference/operator/update/pull/
if "$set" in update:
_update = json.loads(to_json(update["$set"]))
_update = parse_dot_fields(_update)
doc = update_dict_recur(doc, _update)
elif "$unset" in update:
for dotk, v in traverse(doc):
if dotk in update["$unset"]:
v["__REMOVE__"] = True
doc = merge({}, doc)
elif "$push" in update:
for key, val in update["$push"].items():
if "." in key: # not all mongo operators are fully implemented
raise NotImplementedError("nested key in $push: %s" % key)
doc.setdefault(key, []).append(val)
elif "$addToSet" in update:
for key, val in update["$addToSet"].items():
if "." in key: # not all mongo operators are fully implemented
raise NotImplementedError("nested key in $addToSet: %s" % key)
field = doc.setdefault(key, [])
if val not in field:
field.append(val)
else: # "$pull" in update:
for key, val in update["$pull"].items():
if "." in key: # not all mongo operators are fully implemented
raise NotImplementedError("nested key in $pull: %s" % key)
if not isinstance(val, (str, int)):
raise NotImplementedError("value or condition in $pull: %s" % val)
if isinstance(doc.get(key), list):
doc[key][:] = [x for x in doc[key] if x != val]
self._write_one(doc)
[docs] def update_one(self, filter, update, upsert=False, *args, **kwargs):
doc = self.find_one(filter) or {}
if not (doc or upsert):
raise ValueError("No Match.")
self._update_one(doc, update, *args, **kwargs)
[docs] def update_many(self, filter, update, upsert=False, *args, **kwargs):
docs = self.find(filter)
if not docs and upsert:
if any("." in k for k in filter):
raise ValueError("dotfield in upsert.")
docs = [filter]
for doc in docs:
self._update_one(doc, update, *args, **kwargs)
# DEPRECATED
# -----------------
[docs] def update(self, *args, **kwargs):
# In the future,
# Use replace_one(), update_one(), or update_many() instead.
self.update_many(*args, **kwargs)
[docs] def save(self, doc, *args, **kwargs):
# In the future,
# Use insert_one() or replace_one() instead.
self._write_one(doc)
[docs] def remove(self, query):
# In the future,
# Use delete_one() or delete_many() instead.
docs = self.find(query)
collection = self._read()
for doc in docs:
del collection[doc["_id"]]
self._write(collection)
[docs] def count(self):
# In the future,
# Use count_documents() or estimated_document_count() instead.
return len(self._read())
# JSON <-> BSON
# -----------------
# The original author of the biothings.hub decided to have the interface
# of hubdb to model that of MongoDB, the prevailing choice of hub db.
# However, MongoDB stores BSON, extended from JSON to add some optional
# non-JSON-native data types, like dates and binary data, while ES only
# supports JSON documents. Upper layers of biothings.hub has subsequently
# been developed to take full advantage of BSON and specifically have
# stored document values in datetime type in hubdb. Thus, the ES as hubdb
# feature needs to provide the support of these additional data types.
# To achieve generic support of additional data types, non-JSON-serializable
# data by default are serialized to their string representations using "repr".
# For example, datetime objects are stored like datetime.datetime(...).
# Is using "eval" safe? For our internal-use purpose, and combining the fact
# that this module is designed for small demonstrations, better genericity
# currently outweighs the potential security risk, that we may easily support
# serializing additional data types without manually handling each type.
# when the circumstances change, it is advised to reconsider this implementation.
def _pyobj(doc): # ES doc -> Python object
for _, _doc in traverse(doc):
if isinstance(_doc, dict):
for k, v in list(_doc.items()):
_doc[k] = _eval(v)
elif isinstance(_doc, list):
_doc[:] = map(_eval, _doc)
return doc
_PY_REPR = re.compile(r"([\w.]+)\(.*\)")
def _eval(v):
try:
match = _PY_REPR.match(v)
if match:
clsstr = match.group(1)
modstr = clsstr.rsplit(".", 1)[0]
return eval(v, {modstr: import_module(modstr)})
except Exception:
...
return v
class _HubDBEncoder(JSONSerializer):
TYPES = (
datetime.datetime,
# ...
)
def default(self, obj):
if isinstance(obj, self.TYPES):
return repr(obj)
return super().default(obj)