import time import asyncio import struct from binascii import unhexlify from collections import Counter, deque from decimal import Decimal from operator import itemgetter from typing import Optional, List, Iterable, Union from elasticsearch import AsyncElasticsearch, NotFoundError, ConnectionError from elasticsearch.helpers import async_streaming_bulk from lbry.error import ResolveCensoredError, TooManyClaimSearchParametersError from lbry.schema.result import Outputs, Censor from lbry.schema.tags import clean_tags from lbry.schema.url import URL, normalize_name from lbry.utils import LRUCache from lbry.wallet.server.db.common import CLAIM_TYPES, STREAM_TYPES from lbry.wallet.server.db.elasticsearch.constants import INDEX_DEFAULT_SETTINGS, REPLACEMENTS, FIELDS, TEXT_FIELDS, \ RANGE_FIELDS, ALL_FIELDS from lbry.wallet.server.util import class_logger from lbry.wallet.server.db.common import ResolveResult class ChannelResolution(str): @classmethod def lookup_error(cls, url): return LookupError(f'Could not find channel in "{url}".') class StreamResolution(str): @classmethod def lookup_error(cls, url): return LookupError(f'Could not find claim at "{url}".') class IndexVersionMismatch(Exception): def __init__(self, got_version, expected_version): self.got_version = got_version self.expected_version = expected_version class SearchIndex: VERSION = 1 def __init__(self, index_prefix: str, search_timeout=3.0, elastic_host='localhost', elastic_port=9200): self.search_timeout = search_timeout self.sync_timeout = 600 # wont hit that 99% of the time, but can hit on a fresh import self.search_client: Optional[AsyncElasticsearch] = None self.sync_client: Optional[AsyncElasticsearch] = None self.index = index_prefix + 'claims' self.logger = class_logger(__name__, self.__class__.__name__) self.claim_cache = LRUCache(2 ** 15) self.search_cache = LRUCache(2 ** 17) self._elastic_host = elastic_host self._elastic_port = elastic_port async def get_index_version(self) -> int: try: template = await self.sync_client.indices.get_template(self.index) return template[self.index]['version'] except NotFoundError: return 0 async def set_index_version(self, version): await self.sync_client.indices.put_template( self.index, body={'version': version, 'index_patterns': ['ignored']}, ignore=400 ) async def start(self) -> bool: if self.sync_client: return False hosts = [{'host': self._elastic_host, 'port': self._elastic_port}] self.sync_client = AsyncElasticsearch(hosts, timeout=self.sync_timeout) self.search_client = AsyncElasticsearch(hosts, timeout=self.search_timeout) while True: try: await self.sync_client.cluster.health(wait_for_status='yellow') break except ConnectionError: self.logger.warning("Failed to connect to Elasticsearch. Waiting for it!") await asyncio.sleep(1) res = await self.sync_client.indices.create(self.index, INDEX_DEFAULT_SETTINGS, ignore=400) acked = res.get('acknowledged', False) if acked: await self.set_index_version(self.VERSION) return acked index_version = await self.get_index_version() if index_version != self.VERSION: self.logger.error("es search index has an incompatible version: %s vs %s", index_version, self.VERSION) raise IndexVersionMismatch(index_version, self.VERSION) await self.sync_client.indices.refresh(self.index) return acked def stop(self): clients = [self.sync_client, self.search_client] self.sync_client, self.search_client = None, None return asyncio.ensure_future(asyncio.gather(*(client.close() for client in clients))) def delete_index(self): return self.sync_client.indices.delete(self.index, ignore_unavailable=True) async def _consume_claim_producer(self, claim_producer): count = 0 async for op, doc in claim_producer: if op == 'delete': yield { '_index': self.index, '_op_type': 'delete', '_id': doc } else: yield { 'doc': {key: value for key, value in doc.items() if key in ALL_FIELDS}, '_id': doc['claim_id'], '_index': self.index, '_op_type': 'update', 'doc_as_upsert': True } count += 1 if count % 100 == 0: self.logger.info("Indexing in progress, %d claims.", count) if count: self.logger.info("Indexing done for %d claims.", count) else: self.logger.debug("Indexing done for %d claims.", count) async def claim_consumer(self, claim_producer): touched = set() async for ok, item in async_streaming_bulk(self.sync_client, self._consume_claim_producer(claim_producer), raise_on_error=False): if not ok: self.logger.warning("indexing failed for an item: %s", item) else: item = item.popitem()[1] touched.add(item['_id']) await self.sync_client.indices.refresh(self.index) self.logger.debug("Indexing done.") def update_filter_query(self, censor_type, blockdict, channels=False): blockdict = {blocked.hex(): blocker.hex() for blocked, blocker in blockdict.items()} if channels: update = expand_query(channel_id__in=list(blockdict.keys()), censor_type=f"<{censor_type}") else: update = expand_query(claim_id__in=list(blockdict.keys()), censor_type=f"<{censor_type}") key = 'channel_id' if channels else 'claim_id' update['script'] = { "source": f"ctx._source.censor_type={censor_type}; " f"ctx._source.censoring_channel_id=params[ctx._source.{key}];", "lang": "painless", "params": blockdict } return update async def update_trending_score(self, params): update_trending_score_script = """ double softenLBC(double lbc) { return (Math.pow(lbc, 1.0 / 3.0)); } double logsumexp(double x, double y) { double top; if(x > y) top = x; else top = y; double result = top + Math.log(Math.exp(x-top) + Math.exp(y-top)); return(result); } double logdiffexp(double big, double small) { return big + Math.log(1.0 - Math.exp(small - big)); } double squash(double x) { if(x < 0.0) return -Math.log(1.0 - x); else return Math.log(x + 1.0); } double unsquash(double x) { if(x < 0.0) return 1.0 - Math.exp(-x); else return Math.exp(x) - 1.0; } double log_to_squash(double x) { return logsumexp(x, 0.0); } double squash_to_log(double x) { //assert x > 0.0; return logdiffexp(x, 0.0); } double squashed_add(double x, double y) { // squash(unsquash(x) + unsquash(y)) but avoiding overflow. // Cases where the signs are the same if (x < 0.0 && y < 0.0) return -logsumexp(-x, logdiffexp(-y, 0.0)); if (x >= 0.0 && y >= 0.0) return logsumexp(x, logdiffexp(y, 0.0)); // Where the signs differ if (x >= 0.0 && y < 0.0) if (Math.abs(x) >= Math.abs(y)) return logsumexp(0.0, logdiffexp(x, -y)); else return -logsumexp(0.0, logdiffexp(-y, x)); if (x < 0.0 && y >= 0.0) { // Addition is commutative, hooray for new math return squashed_add(y, x); } return 0.0; } double squashed_multiply(double x, double y) { // squash(unsquash(x)*unsquash(y)) but avoiding overflow. int sign; if(x*y >= 0.0) sign = 1; else sign = -1; return sign*logsumexp(squash_to_log(Math.abs(x)) + squash_to_log(Math.abs(y)), 0.0); } // Squashed inflated units double inflateUnits(int height) { double timescale = 576.0; // Half life of 400 = e-folding time of a day // by coincidence, so may as well go with it return log_to_squash(height / timescale); } double spikePower(double newAmount) { if (newAmount < 50.0) { return(0.5); } else if (newAmount < 85.0) { return(newAmount / 100.0); } else { return(0.85); } } double spikeMass(double oldAmount, double newAmount) { double softenedChange = softenLBC(Math.abs(newAmount - oldAmount)); double changeInSoftened = Math.abs(softenLBC(newAmount) - softenLBC(oldAmount)); double power = spikePower(newAmount); if (oldAmount > newAmount) { -1.0 * Math.pow(changeInSoftened, power) * Math.pow(softenedChange, 1.0 - power) } else { Math.pow(changeInSoftened, power) * Math.pow(softenedChange, 1.0 - power) } } for (i in params.src.changes) { double units = inflateUnits(i.height); if (ctx._source.trending_score == null) { ctx._source.trending_score = 0.0; } double bigSpike = squashed_multiply(units, squash(spikeMass(i.prev_amount, i.new_amount))); ctx._source.trending_score = squashed_add(ctx._source.trending_score, bigSpike); } """ start = time.perf_counter() def producer(): for claim_id, claim_updates in params.items(): yield { '_id': claim_id, '_index': self.index, '_op_type': 'update', 'script': { 'lang': 'painless', 'source': update_trending_score_script, 'params': {'src': { 'changes': [ { 'height': p.height, 'prev_amount': p.prev_amount / 1E8, 'new_amount': p.new_amount / 1E8, } for p in claim_updates ] }} }, } if not params: return async for ok, item in async_streaming_bulk(self.sync_client, producer(), raise_on_error=False): if not ok: self.logger.warning("updating trending failed for an item: %s", item) await self.sync_client.indices.refresh(self.index) self.logger.info("updated trending scores in %ims", int((time.perf_counter() - start) * 1000)) async def apply_filters(self, blocked_streams, blocked_channels, filtered_streams, filtered_channels): if filtered_streams: await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.SEARCH, filtered_streams), slices=4) await self.sync_client.indices.refresh(self.index) if filtered_channels: await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.SEARCH, filtered_channels), slices=4) await self.sync_client.indices.refresh(self.index) await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.SEARCH, filtered_channels, True), slices=4) await self.sync_client.indices.refresh(self.index) if blocked_streams: await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.RESOLVE, blocked_streams), slices=4) await self.sync_client.indices.refresh(self.index) if blocked_channels: await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.RESOLVE, blocked_channels), slices=4) await self.sync_client.indices.refresh(self.index) await self.sync_client.update_by_query( self.index, body=self.update_filter_query(Censor.RESOLVE, blocked_channels, True), slices=4) await self.sync_client.indices.refresh(self.index) self.clear_caches() def clear_caches(self): self.search_cache.clear() self.claim_cache.clear() async def cached_search(self, kwargs): total_referenced = [] cache_item = ResultCacheItem.from_cache(str(kwargs), self.search_cache) if cache_item.result is not None: return cache_item.result async with cache_item.lock: if cache_item.result: return cache_item.result censor = Censor(Censor.SEARCH) if kwargs.get('no_totals'): response, offset, total = await self.search(**kwargs, censor_type=Censor.NOT_CENSORED) else: response, offset, total = await self.search(**kwargs) censor.apply(response) total_referenced.extend(response) if censor.censored: response, _, _ = await self.search(**kwargs, censor_type=Censor.NOT_CENSORED) total_referenced.extend(response) response = [ ResolveResult( name=r['claim_name'], normalized_name=r['normalized_name'], claim_hash=r['claim_hash'], tx_num=r['tx_num'], position=r['tx_nout'], tx_hash=r['tx_hash'], height=r['height'], amount=r['amount'], short_url=r['short_url'], is_controlling=r['is_controlling'], canonical_url=r['canonical_url'], creation_height=r['creation_height'], activation_height=r['activation_height'], expiration_height=r['expiration_height'], effective_amount=r['effective_amount'], support_amount=r['support_amount'], last_takeover_height=r['last_take_over_height'], claims_in_channel=r['claims_in_channel'], channel_hash=r['channel_hash'], reposted_claim_hash=r['reposted_claim_hash'], reposted=r['reposted'], signature_valid=r['signature_valid'] ) for r in response ] extra = [ ResolveResult( name=r['claim_name'], normalized_name=r['normalized_name'], claim_hash=r['claim_hash'], tx_num=r['tx_num'], position=r['tx_nout'], tx_hash=r['tx_hash'], height=r['height'], amount=r['amount'], short_url=r['short_url'], is_controlling=r['is_controlling'], canonical_url=r['canonical_url'], creation_height=r['creation_height'], activation_height=r['activation_height'], expiration_height=r['expiration_height'], effective_amount=r['effective_amount'], support_amount=r['support_amount'], last_takeover_height=r['last_take_over_height'], claims_in_channel=r['claims_in_channel'], channel_hash=r['channel_hash'], reposted_claim_hash=r['reposted_claim_hash'], reposted=r['reposted'], signature_valid=r['signature_valid'] ) for r in await self._get_referenced_rows(total_referenced) ] result = Outputs.to_base64( response, extra, offset, total, censor ) cache_item.result = result return result async def get_many(self, *claim_ids): await self.populate_claim_cache(*claim_ids) return filter(None, map(self.claim_cache.get, claim_ids)) async def populate_claim_cache(self, *claim_ids): missing = [claim_id for claim_id in claim_ids if self.claim_cache.get(claim_id) is None] if missing: results = await self.search_client.mget( index=self.index, body={"ids": missing} ) for result in expand_result(filter(lambda doc: doc['found'], results["docs"])): self.claim_cache.set(result['claim_id'], result) async def search(self, **kwargs): try: return await self.search_ahead(**kwargs) except NotFoundError: return [], 0, 0 # return expand_result(result['hits']), 0, result.get('total', {}).get('value', 0) async def search_ahead(self, **kwargs): # 'limit_claims_per_channel' case. Fetch 1000 results, reorder, slice, inflate and return per_channel_per_page = kwargs.pop('limit_claims_per_channel', 0) or 0 remove_duplicates = kwargs.pop('remove_duplicates', False) page_size = kwargs.pop('limit', 10) offset = kwargs.pop('offset', 0) kwargs['limit'] = 1000 cache_item = ResultCacheItem.from_cache(f"ahead{per_channel_per_page}{kwargs}", self.search_cache) if cache_item.result is not None: reordered_hits = cache_item.result else: async with cache_item.lock: if cache_item.result: reordered_hits = cache_item.result else: query = expand_query(**kwargs) search_hits = deque((await self.search_client.search( query, index=self.index, track_total_hits=False, _source_includes=['_id', 'channel_id', 'reposted_claim_id', 'creation_height'] ))['hits']['hits']) if remove_duplicates: search_hits = self.__remove_duplicates(search_hits) if per_channel_per_page > 0: reordered_hits = self.__search_ahead(search_hits, page_size, per_channel_per_page) else: reordered_hits = [(hit['_id'], hit['_source']['channel_id']) for hit in search_hits] cache_item.result = reordered_hits result = list(await self.get_many(*(claim_id for claim_id, _ in reordered_hits[offset:(offset + page_size)]))) return result, 0, len(reordered_hits) def __remove_duplicates(self, search_hits: deque) -> deque: known_ids = {} # claim_id -> (creation_height, hit_id), where hit_id is either reposted claim id or original dropped = set() for hit in search_hits: hit_height, hit_id = hit['_source']['creation_height'], hit['_source']['reposted_claim_id'] or hit['_id'] if hit_id not in known_ids: known_ids[hit_id] = (hit_height, hit['_id']) else: previous_height, previous_id = known_ids[hit_id] if hit_height < previous_height: known_ids[hit_id] = (hit_height, hit['_id']) dropped.add(previous_id) else: dropped.add(hit['_id']) return deque(hit for hit in search_hits if hit['_id'] not in dropped) def __search_ahead(self, search_hits: list, page_size: int, per_channel_per_page: int): reordered_hits = [] channel_counters = Counter() next_page_hits_maybe_check_later = deque() while search_hits or next_page_hits_maybe_check_later: if reordered_hits and len(reordered_hits) % page_size == 0: channel_counters.clear() elif not reordered_hits: pass else: break # means last page was incomplete and we are left with bad replacements for _ in range(len(next_page_hits_maybe_check_later)): claim_id, channel_id = next_page_hits_maybe_check_later.popleft() if per_channel_per_page > 0 and channel_counters[channel_id] < per_channel_per_page: reordered_hits.append((claim_id, channel_id)) channel_counters[channel_id] += 1 else: next_page_hits_maybe_check_later.append((claim_id, channel_id)) while search_hits: hit = search_hits.popleft() hit_id, hit_channel_id = hit['_id'], hit['_source']['channel_id'] if hit_channel_id is None or per_channel_per_page <= 0: reordered_hits.append((hit_id, hit_channel_id)) elif channel_counters[hit_channel_id] < per_channel_per_page: reordered_hits.append((hit_id, hit_channel_id)) channel_counters[hit_channel_id] += 1 if len(reordered_hits) % page_size == 0: break else: next_page_hits_maybe_check_later.append((hit_id, hit_channel_id)) return reordered_hits async def _get_referenced_rows(self, txo_rows: List[dict]): txo_rows = [row for row in txo_rows if isinstance(row, dict)] referenced_ids = set(filter(None, map(itemgetter('reposted_claim_id'), txo_rows))) referenced_ids |= set(filter(None, (row['channel_id'] for row in txo_rows))) referenced_ids |= set(filter(None, (row['censoring_channel_id'] for row in txo_rows))) referenced_txos = [] if referenced_ids: referenced_txos.extend(await self.get_many(*referenced_ids)) referenced_ids = set(filter(None, (row['channel_id'] for row in referenced_txos))) if referenced_ids: referenced_txos.extend(await self.get_many(*referenced_ids)) return referenced_txos def expand_query(**kwargs): if "amount_order" in kwargs: kwargs["limit"] = 1 kwargs["order_by"] = "effective_amount" kwargs["offset"] = int(kwargs["amount_order"]) - 1 if 'name' in kwargs: kwargs['name'] = normalize_name(kwargs.pop('name')) if kwargs.get('is_controlling') is False: kwargs.pop('is_controlling') query = {'must': [], 'must_not': []} collapse = None if 'fee_currency' in kwargs and kwargs['fee_currency'] is not None: kwargs['fee_currency'] = kwargs['fee_currency'].upper() for key, value in kwargs.items(): key = key.replace('claim.', '') many = key.endswith('__in') or isinstance(value, list) if many and len(value) > 2048: raise TooManyClaimSearchParametersError(key, 2048) if many: key = key.replace('__in', '') value = list(filter(None, value)) if value is None or isinstance(value, list) and len(value) == 0: continue key = REPLACEMENTS.get(key, key) if key in FIELDS: partial_id = False if key == 'claim_type': if isinstance(value, str): value = CLAIM_TYPES[value] else: value = [CLAIM_TYPES[claim_type] for claim_type in value] elif key == 'stream_type': value = [STREAM_TYPES[value]] if isinstance(value, str) else list(map(STREAM_TYPES.get, value)) if key == '_id': if isinstance(value, Iterable): value = [item[::-1].hex() for item in value] else: value = value[::-1].hex() if not many and key in ('_id', 'claim_id') and len(value) < 20: partial_id = True if key in ('signature_valid', 'has_source'): continue # handled later if key in TEXT_FIELDS: key += '.keyword' ops = {'<=': 'lte', '>=': 'gte', '<': 'lt', '>': 'gt'} if partial_id: query['must'].append({"prefix": {"claim_id": value}}) elif key in RANGE_FIELDS and isinstance(value, str) and value[0] in ops: operator_length = 2 if value[:2] in ops else 1 operator, value = value[:operator_length], value[operator_length:] if key == 'fee_amount': value = str(Decimal(value)*1000) query['must'].append({"range": {key: {ops[operator]: value}}}) elif key in RANGE_FIELDS and isinstance(value, list) and all(v[0] in ops for v in value): range_constraints = [] for v in value: operator_length = 2 if v[:2] in ops else 1 operator, stripped_op_v = v[:operator_length], v[operator_length:] if key == 'fee_amount': stripped_op_v = str(Decimal(stripped_op_v)*1000) range_constraints.append((operator, stripped_op_v)) query['must'].append({"range": {key: {ops[operator]: v for operator, v in range_constraints}}}) elif many: query['must'].append({"terms": {key: value}}) else: if key == 'fee_amount': value = str(Decimal(value)*1000) query['must'].append({"term": {key: {"value": value}}}) elif key == 'not_channel_ids': for channel_id in value: query['must_not'].append({"term": {'channel_id.keyword': channel_id}}) query['must_not'].append({"term": {'_id': channel_id}}) elif key == 'channel_ids': query['must'].append({"terms": {'channel_id.keyword': value}}) elif key == 'claim_ids': query['must'].append({"terms": {'claim_id.keyword': value}}) elif key == 'media_types': query['must'].append({"terms": {'media_type.keyword': value}}) elif key == 'any_languages': query['must'].append({"terms": {'languages': clean_tags(value)}}) elif key == 'any_languages': query['must'].append({"terms": {'languages': value}}) elif key == 'all_languages': query['must'].extend([{"term": {'languages': tag}} for tag in value]) elif key == 'any_tags': query['must'].append({"terms": {'tags.keyword': clean_tags(value)}}) elif key == 'all_tags': query['must'].extend([{"term": {'tags.keyword': tag}} for tag in clean_tags(value)]) elif key == 'not_tags': query['must_not'].extend([{"term": {'tags.keyword': tag}} for tag in clean_tags(value)]) elif key == 'not_claim_id': query['must_not'].extend([{"term": {'claim_id.keyword': cid}} for cid in value]) elif key == 'limit_claims_per_channel': collapse = ('channel_id.keyword', value) if kwargs.get('has_channel_signature'): query['must'].append({"exists": {"field": "signature"}}) if 'signature_valid' in kwargs: query['must'].append({"term": {"is_signature_valid": bool(kwargs["signature_valid"])}}) elif 'signature_valid' in kwargs: query.setdefault('should', []) query["minimum_should_match"] = 1 query['should'].append({"bool": {"must_not": {"exists": {"field": "signature"}}}}) query['should'].append({"term": {"is_signature_valid": bool(kwargs["signature_valid"])}}) if 'has_source' in kwargs: query.setdefault('should', []) query["minimum_should_match"] = 1 is_stream_or_repost = {"terms": {"claim_type": [CLAIM_TYPES['stream'], CLAIM_TYPES['repost']]}} query['should'].append( {"bool": {"must": [{"match": {"has_source": kwargs['has_source']}}, is_stream_or_repost]}}) query['should'].append({"bool": {"must_not": [is_stream_or_repost]}}) query['should'].append({"bool": {"must": [{"term": {"reposted_claim_type": CLAIM_TYPES['channel']}}]}}) if kwargs.get('text'): query['must'].append( {"simple_query_string": {"query": kwargs["text"], "fields": [ "claim_name^4", "channel_name^8", "title^1", "description^.5", "author^1", "tags^.5" ]}}) query = { "_source": {"excludes": ["description", "title"]}, 'query': {'bool': query}, "sort": [], } if "limit" in kwargs: query["size"] = kwargs["limit"] if 'offset' in kwargs: query["from"] = kwargs["offset"] if 'order_by' in kwargs: if isinstance(kwargs["order_by"], str): kwargs["order_by"] = [kwargs["order_by"]] for value in kwargs['order_by']: if 'trending_group' in value: # fixme: trending_mixed is 0 for all records on variable decay, making sort slow. continue is_asc = value.startswith('^') value = value[1:] if is_asc else value value = REPLACEMENTS.get(value, value) if value in TEXT_FIELDS: value += '.keyword' query['sort'].append({value: "asc" if is_asc else "desc"}) if collapse: query["collapse"] = { "field": collapse[0], "inner_hits": { "name": collapse[0], "size": collapse[1], "sort": query["sort"] } } return query def expand_result(results): inner_hits = [] expanded = [] for result in results: if result.get("inner_hits"): for _, inner_hit in result["inner_hits"].items(): inner_hits.extend(inner_hit["hits"]["hits"]) continue result = result['_source'] result['claim_hash'] = unhexlify(result['claim_id'])[::-1] if result['reposted_claim_id']: result['reposted_claim_hash'] = unhexlify(result['reposted_claim_id'])[::-1] else: result['reposted_claim_hash'] = None result['channel_hash'] = unhexlify(result['channel_id'])[::-1] if result['channel_id'] else None result['txo_hash'] = unhexlify(result['tx_id'])[::-1] + struct.pack(' str: return self._result @result.setter def result(self, result: str): self._result = result if result is not None: self.has_result.set() @classmethod def from_cache(cls, cache_key, cache): cache_item = cache.get(cache_key) if cache_item is None: cache_item = cache[cache_key] = ResultCacheItem() return cache_item