def get_inc_num(conn: sql_base.Connection, table_name: str, field: str): query_sql = ''' select count(id) as inc_count from {} where {} >= date_sub(CURDATE(), interval 1 day); ''' query_sql = query_sql.format(table_name, field) inc_count = conn.get(query_sql).get("inc_count") return inc_count
class UpdateInnerCode(object): def __init__(self): self._spider_conn = Connection( host=SPIDER_MYSQL_HOST, port=SPIDER_MYSQL_PORT, user=SPIDER_MYSQL_USER, password=SPIDER_MYSQL_PASSWORD, database=SPIDER_MYSQL_DB, ) self._juyuan_conn = Connection( host=JUY_HOST, port=JUY_PORT, user=JUY_USER, password=JUY_PASSWD, database=JUY_DB, ) self.batch_num = 10000 def max_id(self): sql = '''select max(id) as max_id from announcement_base2; ''' return self._spider_conn.get(sql).get("max_id") def load_innercode_map(self): sql = '''select secucode, innercode from secumain; ''' _map = dict() ret = self._juyuan_conn.query(sql) for r in ret: _map[r.get("secucode")] = r.get("innercode") return _map def get_old_inner_code(self, secucode: str): sql = '''select InnerCode from LC_CodeChange where secucode = '{}';'''.format( secucode) r = self._juyuan_conn.get(sql) if r: inner_code = r.get("InnerCode") return inner_code def start(self, begin: int, end: int): inner_map = self.load_innercode_map() # secucode 无前缀 # max_id = self.max_id() # print(max_id) base_sql = '''select id, secucode from announcement_base2 where id between {} and {} and InnerCode = 0; ''' for i in range(begin, end): sql = base_sql.format(i * self.batch_num, i * self.batch_num + self.batch_num) print(sql) datas = self._spider_conn.query(sql) print(len(datas)) for data in datas: inner_code = inner_map.get(data.get('secucode')[2:]) if inner_code is None: inner_code = self.get_old_inner_code( data.get('secucode')[2:]) if inner_code is not None: inner_map.update( {data.get('secucode')[2:]: inner_code}) self._spider_conn.table_update( 'announcement_base2', {'InnerCode': inner_code}, "id", data.get("id")) else: self._spider_conn.table_update('announcement_base2', {'InnerCode': inner_code}, "id", data.get("id"))
class JuchaoLiveNewsSpider(object): """巨潮快讯爬虫""" def __init__(self): self.web_url = 'http://www.cninfo.com.cn/new/commonUrl/quickNews?url=/disclosure/quickNews&queryDate=2020-08-13' self.api_url = 'http://www.cninfo.com.cn/new/quickNews/queryQuickNews?queryDate={}&type=' self.fields = ['code', 'name', 'link', 'title', 'type', 'pub_date'] self.table_name = 'juchao_kuaixun' self.name = '巨潮快讯' self._spider_conn = Connection( host=SPIDER_MYSQL_HOST, port=SPIDER_MYSQL_PORT, user=SPIDER_MYSQL_USER, password=SPIDER_MYSQL_PASSWORD, database=SPIDER_MYSQL_DB, ) self._juyuan_conn = Connection( host=JUY_HOST, port=JUY_PORT, user=JUY_USER, password=JUY_PASSWD, database=JUY_DB, ) self.headers = { 'User-Agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_2) AppleWebKit/537.36 (KHTML, ' 'like Gecko) Chrome/79.0.3945.117 Safari/537.36' } def get_secu_abbr(self, code): """ 从聚源数据库获取证券代码对应的中文简称 :param code: 无前缀的证券代码 :return: """ sql = f'''select SecuAbbr from secumain where secucode=%s;''' name = self._juyuan_conn.get(sql, code).get("SecuAbbr") return name def get_redit_link(self, link): """获取最终重定向后的网址""" resp = self.my_get(link) redit_list = resp.history try: redit_link = redit_list[len(redit_list) - 1].headers["location"] except IndexError: redit_link = link except: return None return redit_link @retry(stop_max_attempt_number=5) def my_get(self, link): # 请求 超时\被ban 时重试 logger.debug(f'get in {link} .. ') resp = requests.get(link, headers=self.headers, timeout=5) return resp def parse(self, url: str, day_string: str): items = [] resp = self.my_get(url) if resp and resp.status_code == 200: text = resp.text datas = json.loads(text) if not datas: print("{} 无公告数据".format(url)) else: for data in datas: item = {} # 需要保存的字段: 快讯的发布详细时间、类型、标题、地址、股票代码、股票名称 announcementTime = time.localtime( int(data.get("announcementTime") / 1000)) announcementTime = time.strftime("%Y-%m-%d %H:%M:%S", announcementTime) item['pub_date'] = announcementTime item['type'] = data.get("type") item['title'] = data.get("title") page_path = data.get("pagePath") if page_path is None: link = '''http://www.cninfo.com.cn/new/disclosure/detail?stock=&announcementId={}&announcementTime={}'''.format( data.get("textId"), day_string) else: try: link = self.get_redit_link(page_path) except: link = None if not link: continue item['link'] = link code = data.get("code") if code: item['code'] = code # 无前缀的证券代码 item['name'] = self.get_secu_abbr(code) items.append(item) return items def start(self, start_dt: datetime.datetime = None, end_dt: datetime.datetime = None): """启动入口""" # 使用传入的结束时间否则置为今天 if end_dt is not None: end_dt = datetime.datetime.combine(end_dt, datetime.time.min) else: end_dt = datetime.datetime.combine(datetime.datetime.now(), datetime.time.min) # 使用传入的开始时间 否则置为结束时间之前 3 天 if start_dt is not None: start_dt = datetime.datetime.combine(start_dt, datetime.time.min) else: start_dt = end_dt - datetime.timedelta(days=3) # 从结束时间向前 按天 遍历到开始时间 遍历列表包含起止时间 this_day = end_dt while this_day >= start_dt: day_string = this_day.strftime("%Y-%m-%d") items = self.parse(self.api_url.format(day_string), day_string) logger.info(f"{day_string}, {len(items)}") self._spider_conn.batch_insert(items, self.table_name, self.fields) this_day -= datetime.timedelta(days=1) time.sleep(3)
class SourceAnnouncementBase(object): """将两个爬虫表合并生成公告基础表""" def __init__(self): self.merge_table_name = 'announcement_base2' self.his_table = 'juchao_ant2' self.live_table = 'juchao_kuaixun' self.batch_number = 10000 self._r_spider_conn = Connection( host=R_SPIDER_MYSQL_HOST, port=R_SPIDER_MYSQL_PORT, user=R_SPIDER_MYSQL_USER, password=R_SPIDER_MYSQL_PASSWORD, database=R_SPIDER_MYSQL_DB, ) self._spider_conn = Connection( host=SPIDER_MYSQL_HOST, port=SPIDER_MYSQL_PORT, user=SPIDER_MYSQL_USER, password=SPIDER_MYSQL_PASSWORD, database=SPIDER_MYSQL_DB, ) self._juyuan_conn = Connection( host=JUY_HOST, port=JUY_PORT, user=JUY_USER, password=JUY_PASSWD, database=JUY_DB, ) def category_code_map(self): return { 'category_bcgz_szsh': ('补充更正', 19), 'category_bndbg_szsh': ('半年报', 2), 'category_cqdq_szsh': ('澄清致歉', 12), 'category_dshgg_szsh': ('董事会', 18), 'category_fxts_szsh': ('风险提示', 21), 'category_gddh_szsh': ('股东大会', 15), 'category_gqbd_szsh': ('股权变动', 16), 'category_gqjl_szsh': ('股权激励', 17), 'category_gszl_szsh': ('公司治理', 24), 'category_gszq_szsh': ('公司债', 25), 'category_jj_szsh': ('解禁', 9), 'category_jshgg_szsh': ('监事会', 14), 'category_kzzq_szsh': ('可转债', 22), 'category_ndbg_szsh': ('年报', 4), 'category_pg_szsh': ('配股', 7), 'category_qtrz_szsh': ('其他融资', 23), 'category_qyfpxzcs_szsh': ('权益分派', 11), 'category_rcjy_szsh': ('日常经营', 10), 'category_sf_szsh': ('首发', 8), 'category_sjdbg_szsh': ('三季报', 3), 'category_tbclts_szsh': ('特别处理和退市', 13), 'category_tszlq_szsh': ('退市整理期', 20), 'category_yjdbg_szsh': ('一季报', 1), 'category_yjygjxz_szsh': ('业绩预告', 5), 'category_zf_szsh': ('增发', 6), 'category_zj_szsh': ('中介报告', 26), 'category_others': ('其他', 27), } def spider_max_id(self, table_name: str): sql = f'''select max(id) as max_id from {table_name}; ''' max_id = self._r_spider_conn.get(sql).get("max_id") return max_id def check_ids(self): _delta = [] max_id = self.spider_max_id(self.his_table) for i in range(int(max_id / self.batch_number) + 1): begin_id = i * self.batch_number end_id = i * self.batch_number + self.batch_number print() print(begin_id, end_id) sql1 = '''select count(id) from {} where id between {} and {}; '''.format( self.his_table, begin_id, end_id) sql2 = '''select count(id) from {} where id between {} and {}; '''.format( self.merge_table_name, begin_id, end_id) count1 = self._r_spider_conn.get(sql1).get("count(id)") count2 = self._r_spider_conn.get(sql2).get("count(id)") print(count1, count2) if count1 != count2: sl1 = '''select id from {} where id between {} and {};'''.format( self.his_table, begin_id, end_id) sl2 = '''select id from {} where id between {} and {};'''.format( self.merge_table_name, begin_id, end_id) rs1 = self._r_spider_conn.query(sl1) ids1 = set([r.get("id") for r in rs1]) rs2 = self._r_spider_conn.query(sl2) ids2 = set([r.get("id") for r in rs2]) print(list(ids1 - ids2)) _delta.extend(list(ids1 - ids2)) print(_delta) return _delta def process_diff_ids(self, diff_lst): diff_lst = tuple(diff_lst) category_code_map = self.category_code_map() _sql = '''select id, SecuCode, CategoryCode, SecuAbbr, AntTime as PubDatetime1, \ AntTitle as Title1, AntDoc as PDFLink, CREATETIMEJZ as InsertDatetime1 from {} where id in {}; '''.format( self.his_table, diff_lst) datas = self._r_spider_conn.query(_sql) print(len(datas)) for data in datas: data['CategoryCode'] = category_code_map.get( data.get("CategoryCode"))[1] data = utils.process_secucode(data) # 暂时只要 0 3 6 if data: print(data) ret = self._spider_conn.table_insert( self.merge_table_name, data, [ 'SecuCode', 'SecuAbbr', 'CategoryCode', 'PubDatetime1', 'InsertDatetime1', 'Title1' ]) print(ret) def first_load(self): inner_map = self.load_innercode_map() category_code_map = self.category_code_map() load_sql = '''select id, SecuCode, CategoryCode, SecuAbbr, AntTime as PubDatetime1, \ AntTitle as Title1, AntDoc as PDFLink, CREATETIMEJZ as InsertDatetime1 from {} where \ id >= {} and id < {}; ''' max_id = self.spider_max_id(self.his_table) for i in range(int(max_id / self.batch_number) + 1): begin_id = i * self.batch_number end_id = i * self.batch_number + self.batch_number print(begin_id, end_id) _sql = load_sql.format(self.his_table, begin_id, end_id) print(_sql) datas = self._r_spider_conn.query(_sql) for data in datas: inner_code = inner_map.get(data.get("SecuCode")) if inner_code: data['InnerCode'] = inner_code # 爬虫库中的 secucode 无前缀 else: inner_code = self.get_old_inner_code(data.get("SecuCode")) if inner_code: data['InnerCode'] = inner_code inner_map.update({data.get("SecuCode"): inner_code}) else: continue data['CategoryCode'] = category_code_map.get( data.get("CategoryCode"))[1] data = utils.process_secucode(data) # 暂时只要 0 3 6 if data: self._spider_conn.table_insert( self.merge_table_name, data, [ 'SecuCode', 'SecuAbbr', 'CategoryCode', 'PubDatetime1', 'InsertDatetime1', 'Title1', 'InnerCode' ]) # 将巨潮快讯的数据补充进去 update_sql = '''select A.* from {} A, {} B \ where A.code = B.SecuCode and A.link = B.AntDoc and A.type = '公告' \ and A.id between {} and {}; ''' max_id = self.spider_max_id(self.live_table) for j in range(int(max_id / self.batch_number) + 1): begin_id = j * self.batch_number end_id = j * self.batch_number + self.batch_number print(begin_id, end_id) _sql = update_sql.format(self.live_table, self.his_table, begin_id, end_id) print(_sql) datas = self._r_spider_conn.query(_sql) for data in datas: item = { 'PubDatetime2': data.get("pub_date"), 'InsertDatetime2': data.get("CREATETIMEJZ"), 'Title2': data.get("title"), } self._spider_conn.table_update(self.merge_table_name, item, 'PDFLink', data.get("link")) def load_innercode_map(self): sql = '''select secucode, innercode from secumain; ''' _map = dict() ret = self._juyuan_conn.query(sql) for r in ret: _map[r.get("secucode")] = r.get("innercode") return _map def get_old_inner_code(self, secucode: str): sql = '''select InnerCode from LC_CodeChange where secucode = '{}';'''.format( secucode) r = self._juyuan_conn.get(sql) if r: inner_code = r.get("InnerCode") return inner_code
class SourceAnnouncementBase(object): """将两个爬虫表合并生成公告基础表""" def __init__(self): self.merge_table_name = 'announcement_base2' self.his_table = 'juchao_ant2' self.live_table = 'juchao_kuaixun' self.batch_number = 10000 self._r_spider_conn = Connection( host=R_SPIDER_MYSQL_HOST, port=R_SPIDER_MYSQL_PORT, user=R_SPIDER_MYSQL_USER, password=R_SPIDER_MYSQL_PASSWORD, database=R_SPIDER_MYSQL_DB, ) self._spider_conn = Connection( host=SPIDER_MYSQL_HOST, port=SPIDER_MYSQL_PORT, user=SPIDER_MYSQL_USER, password=SPIDER_MYSQL_PASSWORD, database=SPIDER_MYSQL_DB, ) self._juyuan_conn = Connection( host=JUY_HOST, port=JUY_PORT, user=JUY_USER, password=JUY_PASSWD, database=JUY_DB, ) def category_code_map(self): return { 'category_bcgz_szsh': ('补充更正', 19), 'category_bndbg_szsh': ('半年报', 2), 'category_cqdq_szsh': ('澄清致歉', 12), 'category_dshgg_szsh': ('董事会', 18), 'category_fxts_szsh': ('风险提示', 21), 'category_gddh_szsh': ('股东大会', 15), 'category_gqbd_szsh': ('股权变动', 16), 'category_gqjl_szsh': ('股权激励', 17), 'category_gszl_szsh': ('公司治理', 24), 'category_gszq_szsh': ('公司债', 25), 'category_jj_szsh': ('解禁', 9), 'category_jshgg_szsh': ('监事会', 14), 'category_kzzq_szsh': ('可转债', 22), 'category_ndbg_szsh': ('年报', 4), 'category_pg_szsh': ('配股', 7), 'category_qtrz_szsh': ('其他融资', 23), 'category_qyfpxzcs_szsh': ('权益分派', 11), 'category_rcjy_szsh': ('日常经营', 10), 'category_sf_szsh': ('首发', 8), 'category_sjdbg_szsh': ('三季报', 3), 'category_tbclts_szsh': ('特别处理和退市', 13), 'category_tszlq_szsh': ('退市整理期', 20), 'category_yjdbg_szsh': ('一季报', 1), 'category_yjygjxz_szsh': ('业绩预告', 5), 'category_zf_szsh': ('增发', 6), 'category_zj_szsh': ('中介报告', 26), 'category_others': ('其他', 27), } def load_innercode_map(self): sql = '''select secucode, innercode from secumain; ''' _map = dict() ret = self._juyuan_conn.query(sql) for r in ret: _map[r.get("secucode")] = r.get("innercode") return _map def get_old_inner_code(self, secucode: str): sql = '''select InnerCode from LC_CodeChange where secucode = '{}';'''.format(secucode) r = self._juyuan_conn.get(sql) if r: inner_code = r.get("InnerCode") return inner_code def daily_update(self, deadline: datetime.datetime = None): inner_map = self.load_innercode_map() category_code_map = self.category_code_map() if deadline is None: deadline = datetime.datetime.now() - datetime.timedelta(days=1) load_sql = '''select id, SecuCode, CategoryCode, SecuAbbr, AntTime as PubDatetime1, \ AntTitle as Title1, AntDoc as PDFLink, CREATETIMEJZ as InsertDatetime1 from {} where \ UPDATETIMEJZ > '{}'; '''.format(self.his_table, deadline) logger.info(f"his load sql is {load_sql}") datas = self._r_spider_conn.query(load_sql) logger.info(f"load count is {len(datas)} from his table.") items = [] for data in datas: data['CategoryCode'] = category_code_map.get(data.get("CategoryCode"))[1] inner_code = inner_map.get(data.get("SecuCode")) if inner_code: data['InnerCode'] = inner_code # 爬虫库中的 secucode 无前缀 else: inner_code = self.get_old_inner_code(data.get("SecuCode")) if inner_code: inner_map.update({data.get("SecuCode"): inner_code}) data['InnerCode'] = inner_code else: continue data = utils.process_secucode(data) # 暂时只要 0 3 6 if data: items.append(data) self._spider_conn.batch_insert(items, self.merge_table_name, ['SecuCode', 'SecuAbbr', 'CategoryCode', 'PubDatetime1', 'InsertDatetime1', 'Title1', 'InnerCode']) update_sql = '''select A.* from juchao_kuaixun A, juchao_ant B where B.UPDATETIMEJZ > '{}' \ and A.code = B.SecuCode and A.link = B.AntDoc and A.type = '公告'; '''.format(deadline) logger.info(f"live sql is {update_sql}") datas = self._r_spider_conn.query(update_sql) logger.info(f'load count {len(datas)} from live table.') for data in datas: item = { 'PubDatetime2': data.get("pub_date"), 'InsertDatetime2': data.get("CREATETIMEJZ"), 'Title2': data.get("title"), } self._spider_conn.table_update(self.merge_table_name, item, 'PDFLink', data.get("link"))