pyxllib 0.3.96__py3-none-any.whl → 0.3.197__py3-none-any.whl

pyxllib/data/pglib.py

@@ -34,24 +34,46 @@ import json
 import json
 import textwrap
 import datetime
+import re
 
 from tqdm import tqdm
 
 import psycopg
 import psycopg.rows
 
-from pyxllib.prog.newbie import round_int
-from pyxllib.prog.pupil import utc_now, utc_timestamp
+from pyxllib.prog.newbie import round_int, human_readable_number
+from pyxllib.prog.pupil import utc_now, utc_timestamp, is_valid_identifier
 from pyxllib.prog.specialist import XlOsEnv
-from pyxllib.file.specialist import get_etag
-from pyxllib.data.sqlite import SqlBase
+from pyxllib.algo.pupil import ValuesStat2
+from pyxllib.file.specialist import get_etag, StreamJsonlWriter
+from pyxllib.data.sqlite import SqlBase, SqlBuilder
 
 
 class Connection(psycopg.Connection, SqlBase):
 
+    def __init__(self, *args, **kwargs):
+        psycopg.Connection.__init__(self, *args, **kwargs)
+        SqlBase.__init__(self, *args, **kwargs)
+
+    def __del__(self):
+        self.close()
+
     def __1_库(self):
         pass
 
+    def get_db_activities(self, datname=None):
+        """
+        检索当前数据库的活动信息。
+
+        :param datname: 这个字段理论上应该要能自动检测出来才对，但这会急着没空钻研，先手动输入吧
+        """
+        sql = SqlBuilder('pg_stat_activity')
+        sql.select('pid', 'datname', 'usename', 'state', 'query', 'age(now(), query_start) AS "query_age"')
+        sql.where("state = 'active'")
+        if datname:
+            sql.where(f"datname = '{datname}'")
+        return self.exec2dict(sql.build_select()).fetchall()
+
     def __2_表格(self):
         pass
 
@@ -84,6 +106,49 @@ class Connection(psycopg.Connection, SqlBase):
         self.execute(f"COMMENT ON COLUMN {table_name}.{col_name} IS $zyzf${comment}$zyzf$")
         self.commit()
 
+    def reset_table_item_id(self, table_name, item_id_name=None, counter_name=None):
+        """ 重置表格的数据的id值，也重置计数器
+
+        :param item_id_name: 表格的自增id字段名，有一套我自己风格的自动推算算法
+        :param counter_name: 计数器字段名，如果有的话，也会重置
+        """
+        # 1 重置数据的编号
+        if item_id_name is None:
+            m = re.match(r'(.+?)(_table)?$', table_name)
+            item_id_name = m.group(1) + '_id'
+
+        sql = f"""WITH cte AS (
+    SELECT {item_id_name}, ROW_NUMBER() OVER (ORDER BY {item_id_name}) AS new_{item_id_name}
+    FROM {table_name}
+)
+UPDATE {table_name}
+SET {item_id_name} = cte.new_{item_id_name}
+FROM cte
+WHERE {table_name}.{item_id_name} = cte.{item_id_name}"""
+        self.execute(sql)  # todo 这种sql写法好像偶尔有bug会出问题
+        self.commit()
+
+        # 2 重置计数器
+        if counter_name is None:
+            counter_name = f'{table_name}_{item_id_name}_seq'
+
+        # 找到目前最大的id值
+        max_id = self.exec2one(f'SELECT MAX({item_id_name}) FROM {table_name}')
+        # self.execute(f'ALTER SEQUENCE {counter_name} RESTART WITH {max_id + 1}')
+        # 检查序列是否存在，如果不存在则创建序列，然后重置序列
+        # 检查序列是否存在，如果不存在则创建序列，并将其关联到指定的表和字段
+        self.execute(f"""
+DO $$
+BEGIN
+    IF NOT EXISTS (SELECT FROM pg_class WHERE relkind = 'S' AND relname = '{counter_name}') THEN
+        EXECUTE format('CREATE SEQUENCE %I', '{counter_name}');
+        EXECUTE format('ALTER TABLE %I ALTER COLUMN %I SET DEFAULT nextval(''%I'')', '{table_name}', '{item_id_name}', '{counter_name}');
+    END IF;
+    EXECUTE format('ALTER SEQUENCE %I RESTART WITH %s', '{counter_name}', {max_id + 1});
+END $$;
+""")
+        self.commit()
+
     def __3_execute(self):
         pass
 
@@ -99,6 +164,43 @@ class Connection(psycopg.Connection, SqlBase):
         # cur.close()
         return data
 
+    def exec2dict_batch(self, sql, batch_size=1000, use_offset=None, **kwargs):
+        """ 分批返回数据的版本
+
+        :param use_offset: 是否使用offset分页，会根据sql中是否含有where自动判断，但有时候最好明确指定以防错误
+            如果外部sql每次操作，会改变数据库的情况，导致sql的where规则虽然没变，但是数据本身发生变化，则offset应该要关闭
+                每次取对应的满足条件的数据即可
+                这种情况，也需要本函数内部主动执行commit_all的
+            否则，只是一种遍历查询，没有where或者where获取的数据情况是不会变化的，则要使用offset
+        :return:
+            第1个值，是一个迭代器，看起来仍然能一条一条返回，实际后台是按照batch_size打包获取的
+            第2个值，是数据总数
+        """
+        if not isinstance(sql, SqlBuilder):
+            raise ValueError('暂时只能搭配SQLBuilder使用')
+
+        if use_offset is None:
+            use_offset = not sql._where
+
+        num = self.exec2one(sql.build_count())
+        offset = 0
+
+        def yield_row():
+            nonlocal offset
+            while True:
+                sql2 = sql.copy()
+                if not use_offset:  # 如果不使用offset，那么缓存的sql操作需要全部提交，确保数据都更新后，再提取数据
+                    self.commit_all()
+                sql2.limit(batch_size, offset)
+                rows = self.exec2dict(sql2.build_select(), **kwargs).fetchall()
+                if use_offset:
+                    offset += len(rows)
+                if not rows:
+                    break
+                yield from rows
+
+        return yield_row(), num
+
     exec_dict = exec2dict
 
     def __4_数据类型(self):
@@ -107,7 +209,7 @@ class Connection(psycopg.Connection, SqlBase):
     @classmethod
     def cvt_type(cls, val):
         if isinstance(val, (dict, list)):
-            val = json.dumps(val, ensure_ascii=False)
+            val = json.dumps(val, ensure_ascii=False, default=str)
         # 注意list数组类型读、写都会自动适配py
         return val
 
@@ -121,15 +223,17 @@ class Connection(psycopg.Connection, SqlBase):
             return 'boolean'
         elif isinstance(val, float):
             return 'float4'
-        elif isinstance(val, dict):
+        elif isinstance(val, (dict, list)):
             return 'jsonb'
+        elif isinstance(val, datetime.datetime):
+            return 'timestamp'
         else:  # 其他list等类型，可以用json.dumps或str转文本存储
             return 'text'
 
     def __5_增删改查(self):
         pass
 
-    def insert_row(self, table_name, cols, *, on_conflict='DO NOTHING', commit=True):
+    def insert_row(self, table_name, cols, *, on_conflict='DO NOTHING', commit=False):
         """ 【增】插入新数据
 
         :param dict cols: 用字典表示的要插入的值
@@ -140,6 +244,10 @@ class Connection(psycopg.Connection, SqlBase):
                 也可以写复杂的处理算法规则，详见 http://postgres.cn/docs/12/sql-insert.html
                 比如这里是插入的id重复的话，就把host_name替换掉，还可以指定nick_name替换为'abc'
                 注意前面的(id)是必须要输入的
+
+        注意：有个常见需求，是想插入后返回对应的id，但是这样就需要知道这张表自增的id字段名
+            以及还是很难获得插入后的id值，可以默认刚插入的id是最大的，但是这样并不安全，有风险
+            建议还是外部自己先计算全表最大的id值，自己实现自增，就能知道插入的这条数据的id了
         """
         ks = ','.join(cols.keys())
         vs = ','.join(['%s'] * (len(cols.keys())))
@@ -153,11 +261,9 @@ class Connection(psycopg.Connection, SqlBase):
             on_conflict = f'ON CONFLICT {on_conflict}'
         query += f' {on_conflict}'
 
-        self.execute(query, params)
-        if commit:
-            self.commit()
+        self.commit_base(commit, query, params)
 
-    def insert_rows(self, table_name, keys, ls, *, on_conflict='DO NOTHING', commit=True):
+    def insert_rows(self, table_name, keys, ls, *, on_conflict='DO NOTHING', commit=False):
         """ 【增】插入新数据
 
         :param str keys: 要插入的字段名，一个字符串，逗号,隔开属性值
@@ -179,9 +285,199 @@ class Connection(psycopg.Connection, SqlBase):
             on_conflict = f'ON CONFLICT {on_conflict}'
         query += f' {on_conflict}'
 
-        self.execute(query, params)
-        if commit:
-            self.commit()
+        self.commit_base(commit, query, params)
+
+    def __6_高级统计(self):
+        pass
+
+    def get_column_valuesstat(self, table_name, column, filter_condition=None,
+                              percentile_count=5,
+                              by_data=False, data_type=None):
+        """ 获得指定表格的某个字段的统计特征ValuesStat2对象
+
+        :param table_name: 表名
+        :param column: 用于计算统计数据的字段名
+            不一定是标准的字段名
+        :param percentile_count: 分位数的数量，例如 3 表示只计算中位数
+        :param by_data: 是否获得原始数据
+            默认只获得统计特征，不获得原始数据
+        """
+
+        def init_from_db_data():
+            sql = SqlBuilder(table_name)
+            if filter_condition:
+                sql.where(filter_condition)
+            values = self.exec2col(sql.build_select(column))
+            if data_type == 'numeric':
+                values = [x and float(x) for x in values]
+            vs = ValuesStat2(raw_values=values, data_type=data_type)
+
+            if data_type == 'text' and is_valid_identifier(column):
+                vs0 = self.get_column_valuesstat(table_name, column, filter_condition=filter_condition,
+                                                 percentile_count=percentile_count, by_data=False)
+                vs.n = vs0.n
+                vs.dist = vs0.dist
+
+            return vs
+
+        def init_from_db():
+            # 1 构建基础的 SQL 查询
+            sql = SqlBuilder(table_name)
+            sql.select("COUNT(*) AS total_count")
+            sql.select(f"COUNT({column}) AS non_null_count")
+            sql.select(f"MIN({column}) AS min_value")
+            sql.select(f"MAX({column}) AS max_value")
+            if data_type and 'timestamp' in data_type:
+                percentile_type = 'PERCENTILE_DISC'
+                # todo 其实时间类也可以"泛化"一种平均值、标准差算法的，这需要获取全量数据，然后自己计算
+            elif data_type == 'text':
+                percentile_type = 'PERCENTILE_DISC'
+            else:  # 默认是正常的数值类型
+                sql.select(f"SUM({column}) AS total_sum")
+                sql.select(f"AVG({column}) AS average")
+                sql.select(f"STDDEV({column}) AS standard_deviation")
+                percentile_type = 'PERCENTILE_CONT'
+
+            percentiles = []
+            # 根据分位点的数量动态添加分位数计算
+            if percentile_count > 2:
+                step = 1 / (percentile_count - 1)
+                percentiles = [(i * step) for i in range(1, percentile_count - 1)]
+                for p in percentiles:
+                    sql.select(f"{percentile_type}({p:.2f}) WITHIN GROUP (ORDER BY {column}) "
+                               f"AS percentile_{int(p * 100)}")
+
+            if filter_condition:
+                sql.where(filter_condition)
+
+            row = self.exec2dict(sql.build_select()).fetchone()
+
+            # 2 统计展示
+            x = ValuesStat2(data_type=data_type)
+            x.raw_n = row['total_count']
+            x.n = row['non_null_count']
+            if not x.n:
+                return x
+
+            x.sum = row.get('total_sum', None)
+            x.mean = row.get('average', None)
+            x.std = row.get('standard_deviation', None)
+
+            # 如果计算了分位数，填充相应属性
+            x.dist = [row['min_value']] + [row[f"percentile_{int(p * 100)}"] for p in percentiles] + [row['max_value']]
+            if data_type == 'numeric':
+                x.dist = [float(x) for x in x.dist]
+
+            return x
+
+        data_type = data_type or self.get_column_data_type(table_name, column)
+
+        # 如果不是标准的列名，强制获取数据
+        if not is_valid_identifier(column):
+            by_data = True
+
+        if by_data:
+            vs = init_from_db_data()
+        else:
+            vs = init_from_db()
+
+        return vs
+
+    def export_jsonl(self, file_path, table_name, key_col=None, batch_size=1000, print_mode=0):
+        """ 将某个表导出为本地jsonl文件
+
+        :param str|SqlBuilder table_name: 表名
+            支持传入SqlBuilder对象，这样可以更灵活的控制导出的数据规则
+        :param file_path: 导出的文件路径
+        :param batch_size: 每次读取的行数和保存的行数
+        :param key_col: 作为主键的列名，如果有的话，会自动去重
+            强烈推荐要设置
+            实际不一定要用主键，只要是有顺序值的列就行
+
+        todo 暴力最简单的版本不难写，我纠结的是缓存机制，还有bytes类型数据会有点大等问题
+            还需要先支持一个通用的缓存写文件功能
+        """
+        # 1 sql
+        if isinstance(table_name, str):
+            sql = SqlBuilder(table_name)
+            sql.select('*')
+        else:
+            sql = table_name
+            m = re.search(r'FROM (\w+)', sql.build_select())
+            table_name = m.group(1) if m else 'table'
+        assert isinstance(sql, SqlBuilder)
+
+        file_path = XlPath(file_path)
+        if key_col:
+            sql.order_by(key_col)
+            if file_path.is_file():
+                # 读取现有数据，找出主键最大值
+                data = file_path.read_jsonl(batch_size=1000)
+                if data:
+                    max_val = max([x[key_col] for x in data]) if data else None
+                    if max_val is not None:
+                        sql.where(f'{key_col} > {max_val}')
+
+        # 2 获取数据
+        file = StreamJsonlWriter(file_path, batch_size=batch_size)  # 流式存储
+        rows, total = self.exec2dict_batch(sql, batch_size=batch_size, use_offset=True)
+        for row in tqdm(rows, total=total, desc=f'从{table_name}表导出数据', disable=not print_mode):
+            file.append_line(row)
+        file.flush()
+
+    def check_db_tables_size(self, db_name=None):
+        """ 查看指定数据下所有表格的大小 """
+        from datetime import datetime
+        import pandas as pd
+
+        if db_name is None:
+            # 使用sql获取当前self所在数据库
+            db_name = self.exec2one("SELECT current_database()")
+
+        data = []
+        tables = self.exec2col("SELECT table_name FROM information_schema.tables WHERE table_schema='public'")
+        for table_name in tables:
+            row = {
+                'database': db_name,
+                'table_name': table_name,
+            }
+            sz = self.exec2one(f"SELECT pg_total_relation_size('public.{table_name}')")
+            if not sz:
+                continue
+            lines = self.exec2one(f"SELECT COUNT(*) FROM {table_name}")
+            row['size'], row['lines'] = sz, lines
+            row['readable_size'] = human_readable_number(sz, 'KB')
+            row['perline_size'] = human_readable_number(sz / lines, 'KB') if lines else -1
+            row['update_time'] = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
+            data.append(row)
+
+        df = pd.DataFrame.from_dict(data)
+        if len(df):
+            df.sort_values(['size'], ascending=False, inplace=True)
+        df.reset_index(drop=True, inplace=True)
+        return df
+
+    def check_multi_db_size(self, db_list):
+        """ 这个功能一般要用postgres账号，才有权限处理所有数据库 """
+        from datetime import datetime
+        import pandas as pd
+
+        data = []
+        for db in db_list:
+            row = {
+                'name': db,
+            }
+            sz = self.exec2one(f"SELECT pg_database_size('{db}')")
+            row['size'] = sz
+            row['readable_size'] = human_readable_number(sz, 'KB')
+            row['update_time'] = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
+
+            data.append(row)
+
+        df = pd.DataFrame.from_dict(data)
+        df.sort_values(['size'], ascending=False, inplace=True)
+        df.reset_index(drop=True, inplace=True)
+        return df
 
 
 """
@@ -377,7 +673,7 @@ class XlprDb(Connection):
         self.insert_row('xlapi', {'input': input, 'output': output,
                                   'elapse_ms': elapse_ms, 'update_time': utc_timestamp(8)},
                         on_conflict=on_conflict)
-
+        self.commit()
         return self.execute('SELECT id FROM xlapi WHERE input=%s', (input,)).fetchone()[0]
 
     def insert_row2xlserver(self, request, xlapi_id=0, **kwargs):
@@ -387,7 +683,7 @@ class XlprDb(Connection):
               'xlapi_id': xlapi_id}
         kw.update(kwargs)
         print(kw)  # 监控谁在用api
-        self.insert_row('xlserver', kw)
+        self.insert_row('xlserver', kw, commit=True)
 
     def __3_host_trace相关可视化(self):
         """ TODO dbview 改名 host_trace """
@@ -419,15 +715,16 @@ class XlprDb(Connection):
                 if gpu:
                     status['gpu_memory'] = ssh.check_gpu_usage(print_mode=True)
                 if disk:
-                    # 检查磁盘空间会很慢，如果超时可以跳过。
-                    status['disk_memory'] = ssh.check_disk_usage(print_mode=True, timeout=7200)
+                    # 检查磁盘空间会很慢，如果超时可以跳过。（设置超时6小时）
+                    status['disk_memory'] = ssh.check_disk_usage(print_mode=True, timeout=60 * 60 * 6)
             except Exception as e:
                 status = {'error': f'{str(type(e))[8:-2]}: {e}'}
                 print(status)
 
             if status:
                 self.insert_row('host_trace',
-                                {'host_name': host_name, 'status': status, 'update_time': utc_timestamp(8)})
+                                {'host_name': host_name, 'status': status, 'update_time': utc_timestamp(8)},
+                                commit=True)
             print()
 
     def _get_host_trace_total(self, mode, title, yaxis_name, date_trunc, recent, host_attr):
@@ -499,6 +796,101 @@ class XlprDb(Connection):
         def pretty_val(v):
             return round_int(v) if v > 100 else round(v, 2)
 
+        try:
+            chart = Line()
+            chart.set_title(title)
+            chart.options['xAxis'][0].update({'min': ls[0][0], 'type': 'time',
+                                              # 'minInterval': 3600 * 1000 * 24,
+                                              'name': '时间', 'nameGap': 50, 'nameLocation': 'middle'})
+            chart.options['yAxis'][0].update({'name': yaxis_name, 'nameGap': 50, 'nameLocation': 'middle'})
+            # 目前是比较暴力的方法调整排版，后续要研究是不是能更自动灵活些
+            chart.options['legend'][0].update({'top': '6%', 'icon': 'pin'})
+            chart.options['grid'] = [{'top': 55 + len(all_users_usaged) * 4, 'containLabel': True}]
+            chart.options['tooltip'].opts.update({'axisPointer': {'type': 'cross'}, 'trigger': 'item'})
+
+            chart.add_series(f'total{pretty_val(ls[0][1]):g}', to_list([x[1] for x in ls]), areaStyle={})
+            for user, usaged in all_users_usaged.most_common():
+                usaged = usaged / ((ls[-1][0] - ls[0][0]).total_seconds() / 3600 + 1e-9)
+                chart.add_series(f'{user}{pretty_val(usaged):g}',
+                                 to_list([x[2].get(user, 0) for x in ls]),
+                                 areaStyle={}, stack='Total', emphasis={'focus': 'series'})
+
+            return '<body>' + chart.render_embed() + '</body>', sum(all_users_usaged.values())
+        except Exception as e:
+            return str(e), 0
+
+    #   cdx_edit
+    def _get_database_trace_total(self, title, yaxis_name, date_trunc, recent, link_name):
+        ls = self.execute(textwrap.dedent(f"""\
+            WITH cte1 AS(
+                SELECT link_name, jsonb_each(status::jsonb) AS db_data, date_trunc('{date_trunc}', update_time) ttime
+                FROM database_trace WHERE update_time > %s AND link_name = '{link_name}'
+            ), cte2 AS(
+                SELECT ttime, link_name, (db_data).key AS table_name, ((db_data).value->> '_total')::bigint AS total
+                FROM cte1
+            )
+            SELECT ttime, jsonb_object_agg(table_name, total) AS aggregated_json,SUM(total) as total
+            FROM cte2
+            GROUP BY ttime
+            ORDER BY ttime"""), ((utc_now(8) - recent).isoformat(timespec='seconds'),)).fetchall()
+        return self.database_create_stack_chart(title, ls, yaxis_name=yaxis_name)
+
+    def _get_database_trace_per_host(self, db, title, yaxis_name, date_trunc, recent, link_name):
+        ls = self.execute(textwrap.dedent(f"""\
+            WITH cte1 AS (
+                SELECT link_name, jsonb_each(status::jsonb) AS db_data, date_trunc('{date_trunc}', update_time) ttime
+                FROM database_trace WHERE update_time > %s AND link_name = '{link_name}'
+                ), cte2 AS (
+                    SELECT ttime, link_name, (db_data).key AS table_name, (db_data).value AS size_text
+                    FROM cte1
+                ), cte3 AS (
+                    SELECT ttime, table_name, each.key AS key, each.value AS value
+                    FROM cte2, jsonb_each_text(size_text) AS each(key, value)
+                )
+                SELECT ttime, jsonb_object_agg(key,
+                    CASE
+                        WHEN key = '_total' THEN NULL
+                        ELSE (value::jsonb ->> 'size')::bigint  -- Handle other keys as usual
+                    END
+                ) FILTER (WHERE key != '_total') AS aggregated_result,  -- 确保 _total 不在 aggregated_result 中
+                MAX(CASE WHEN key = '_total' THEN value::bigint ELSE NULL END) AS total  -- 单独提取 _total 的值
+                FROM cte3
+                WHERE (key = '_total' OR value::jsonb ? 'size')  -- Ensure that '_total' is included
+                AND table_name = '{db}'
+                GROUP BY ttime
+                ORDER BY ttime"""), ((utc_now(8) - recent).isoformat(timespec='seconds'),)).fetchall()
+        return self.database_create_stack_chart(title, ls, yaxis_name=yaxis_name)
+
+    def database_create_stack_chart(self, title, ls, *, yaxis_name=''):
+        """ 创建展示表
+    
+        :param title: 表格标题
+        :param list ls: n*3，第1列是时间，第3列是总值，第2列是每个用户具体的数据
+        """
+        from pyecharts.charts import Line
+        all_database_usaged = Counter()
+        last_time = None
+        for x in ls:
+            hours = 0 if last_time is None else ((x[0] - last_time).total_seconds() / 3600)
+            last_time = x[0]
+            for k, v in x[1].items():
+                if k == '_total':
+                    continue
+                all_database_usaged[k] += v * hours
+
+        for i, x in enumerate(ls):
+            ct = Counter()
+            for k, v in x[1].items():
+                ct[k] += v
+            ls[i] = (x[0], ct, int(x[2]))
+
+        # 2 转图表可视化
+        def to_list(values):
+            return [(x[0], v) for x, v in zip(ls, values)]
+
+        def pretty_val(v):
+            return round_int(v) if v > 100 else round(v, 2)
+
         chart = Line()
         chart.set_title(title)
         chart.options['xAxis'][0].update({'min': ls[0][0], 'type': 'time',
@@ -507,24 +899,63 @@ class XlprDb(Connection):
         chart.options['yAxis'][0].update({'name': yaxis_name, 'nameGap': 50, 'nameLocation': 'middle'})
         # 目前是比较暴力的方法调整排版，后续要研究是不是能更自动灵活些
         chart.options['legend'][0].update({'top': '6%', 'icon': 'pin'})
-        chart.options['grid'] = [{'top': 55 + len(all_users_usaged) * 4, 'containLabel': True}]
+        chart.options['grid'] = [{'top': 55 + len(all_database_usaged) * 4, 'containLabel': True}]
         chart.options['tooltip'].opts.update({'axisPointer': {'type': 'cross'}, 'trigger': 'item'})
 
-        chart.add_series(f'total{pretty_val(ls[0][1]):g}', to_list([x[1] for x in ls]), areaStyle={})
-        for user, usaged in all_users_usaged.most_common():
+        # chart.add_series(f'total {pretty_val(ls[0][2] / 1024 / 1024 / 1024):g}',
+        #                 to_list([x[2] / 1024 / 1024 / 1024 for x in ls]), areaStyle={})
+        for database, usaged in all_database_usaged.most_common():
             usaged = usaged / ((ls[-1][0] - ls[0][0]).total_seconds() / 3600 + 1e-9)
-            chart.add_series(f'{user}{pretty_val(usaged):g}',
-                             to_list([x[2].get(user, 0) for x in ls]),
+            chart.add_series(f'{database} {pretty_val(usaged / 1024 / 1024 / 1024):g}',
+                             to_list([x[1].get(database, 0) / 1024 / 1024 / 1024 for x in ls]),
                              areaStyle={}, stack='Total', emphasis={'focus': 'series'})
+        return '<body>' + chart.render_embed() + '</body>'
+
+    def dbview_xldb1_memory(self, recent=datetime.timedelta(days=180), date_trunc='day'):
+        from pyxllib.data.echarts import render_echart_html
 
-        return '<body>' + chart.render_embed() + '</body>', sum(all_users_usaged.values())
+        db_list = ['stdata', 'xlpr', 'st', 'ckz']
+        args = ['数据库大小(GB)', date_trunc, recent, 'xldb1']
+        htmltexts = []
+
+        res = self._get_database_trace_total('xldb1数据库使用近况', *args)
+        htmltexts.append(res)
+
+        data_stats = []
+        for idx, db in enumerate(db_list, start=1):
+            data_stats.append(self._get_database_trace_per_host(db, f'{db}', *args))
+        htmltexts += data_stats
+
+        self.commit()
+        h = render_echart_html('database_cdx', body='<br/>'.join(htmltexts))
+        return h
+
+    def dbview_xldb2_memory(self, recent=datetime.timedelta(days=180), date_trunc='day'):
+        from pyxllib.data.echarts import render_echart_html
+
+        db_list = ['ragdata', 'kq5034']
+        args = ['数据库大小(GB)', date_trunc, recent, 'xldb2']
+        htmltexts = []
+
+        res = self._get_database_trace_total('xldb2数据库使用近况', *args)
+        htmltexts.append(res)
+
+        data_stats = []
+        for idx, db in enumerate(db_list, start=1):
+            data_stats.append(self._get_database_trace_per_host(db, f'{db}', *args))
+        htmltexts += data_stats
+
+        self.commit()
+        h = render_echart_html('database_cdx', body='<br/>'.join(htmltexts))
+        return h
 
     def dbview_cpu(self, recent=datetime.timedelta(days=1), date_trunc='hour'):
         from pyxllib.data.echarts import render_echart_html
 
         args = ['CPU核心数（比如4核显示是400%）', date_trunc, recent, 'sum(hosts.cpu_number)*100']
 
-        htmltexts = ['<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
+        htmltexts = [
+            '<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
         res = self._get_host_trace_total('cpu', 'XLPR服务器 CPU 使用近况', *args)
         htmltexts.append(res[0])
 
@@ -546,7 +977,8 @@ class XlprDb(Connection):
 
         args = ['内存（单位：GB）', date_trunc, recent, 'sum(hosts.cpu_gb)']
 
-        htmltexts = ['<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
+        htmltexts = [
+            '<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
         res = self._get_host_trace_total('cpu_memory', 'XLPR服务器 内存 使用近况', *args)
         htmltexts.append(res[0])
 
@@ -563,14 +995,15 @@ class XlprDb(Connection):
         h = render_echart_html('cpu_memory', body='<br/>'.join(htmltexts))
         return h
 
-    def dbview_disk_memory(self, recent=datetime.timedelta(days=30), date_trunc='day'):
+    def dbview_disk_memory(self, recent=datetime.timedelta(days=360), date_trunc='day'):
         """ 查看disk硬盘使用近况
         """
         from pyxllib.data.echarts import render_echart_html
 
         args = ['硬盘（单位：GB）', date_trunc, recent, 'sum(hosts.disk_gb)']
 
-        htmltexts = ['<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
+        htmltexts = [
+            '<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
         res = self._get_host_trace_total('disk_memory', 'XLPR服务器 DISK硬盘 使用近况', *args)
         htmltexts.append(res[0])
         htmltexts.append('注：xlpr4（四卡）服务器使用du计算/home大小有问题，未统计在列<br/>')
@@ -597,7 +1030,8 @@ class XlprDb(Connection):
 
         args = ['显存（单位：GB）', date_trunc, recent, 'sum(hosts.gpu_gb)']
 
-        htmltexts = ['<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
+        htmltexts = [
+            '<a target="_blank" href="https://www.yuque.com/xlpr/data/hnpb2g?singleDoc#"> 《服务器监控》工具使用文档 </a>']
         res = self._get_host_trace_total('gpu_memory', 'XLPR八台服务器 GPU显存 使用近况', *args)
         htmltexts.append(res[0])
 
@@ -641,3 +1075,53 @@ class XlprDb(Connection):
                 self.update_row('files', {'dhash': computed_dhash}, {'id': file_id})
                 progress_bar.update(1)
             self.commit()
+
+    def append_history(self, table_name, where, backup_keys, *,
+                       can_merge=None,
+                       update_time=None,
+                       commit=False):
+        """ 为表格添加历史记录，请确保这个表有一个jsonb格式的historys字段
+
+        这里每次都会对关键字段进行全量备份，没有进行高级的优化。
+        所以只适用于一些历史记录功能场景。更复杂的还是需要另外自己定制。
+
+        :param table_name: 表名
+        :param where: 要记录的id的规则，请确保筛选后记录是唯一的
+        :param backup_keys: 需要备份的字段名
+        :param can_merge: 在某些情况下，history不需要非常冗余地记录，可以给定与上一条合并的规则
+            def can_merge(last, now):
+                "last是上一条字典记录，now是当前要记录的字典数据，
+                返回True，则用now替换last，并不新增记录"
+                ...
+
+        :param update_time: 更新时间，如果不指定则使用当前时间
+        """
+        # 1 获得历史记录
+        ops = ' AND '.join([f'{k}=%s' for k in where.keys()])
+        historys = self.exec2one(f'SELECT historys FROM {table_name} WHERE {ops}', list(where.values())) or []
+        if historys:
+            status1 = historys[-1]
+        else:
+            status1 = {}
+
+        # 2 获得新记录
+        if update_time is None:
+            update_time = utc_timestamp()
+        status2 = self.exec2dict(f'SELECT {",".join(backup_keys)} FROM {table_name} WHERE {ops}',
+                                 list(where.values())).fetchone()
+        status2['update_time'] = update_time
+
+        # 3 添加历史记录
+        if can_merge is None:
+            def can_merge(status1, status2):
+                for k in backup_keys:
+                    if status1.get(k) != status2.get(k):
+                        return False
+                return True
+
+        if historys and can_merge(status1, status2):
+            historys[-1] = status2
+        else:
+            historys.append(status2)
+
+        self.update_row(table_name, {'historys': historys}, where, commit=commit)