Merge pull request #26533 from taosdata/enh/3.0/TD-26258

fix asc/desc fill windows
This commit is contained in:
dapan1121 2024-08-05 09:00:14 +08:00 committed by GitHub
commit 37fc4f5674
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 148 additions and 3 deletions

View File

@ -102,6 +102,7 @@ The detailed beaviors of `NULL`, `NULL_F`, `VALUE`, and VALUE_F are described be
1. A huge volume of interpolation output may be returned using `FILL`, so it's recommended to specify the time range when using `FILL`. The maximum number of interpolation values that can be returned in a single query is 10,000,000. 1. A huge volume of interpolation output may be returned using `FILL`, so it's recommended to specify the time range when using `FILL`. The maximum number of interpolation values that can be returned in a single query is 10,000,000.
2. The result set is in ascending order of timestamp when you aggregate by time window. 2. The result set is in ascending order of timestamp when you aggregate by time window.
3. If aggregate by window is used on STable, the aggregate function is performed on all the rows matching the filter conditions. If `PARTITION BY` is not used in the query, the result set will be returned in strict ascending order of timestamp; otherwise the result set will be returned in the order of ascending timestamp in each group. 3. If aggregate by window is used on STable, the aggregate function is performed on all the rows matching the filter conditions. If `PARTITION BY` is not used in the query, the result set will be returned in strict ascending order of timestamp; otherwise the result set will be returned in the order of ascending timestamp in each group.
4. The output windows of Fill are related with time range of WHERE Clause. For asc fill, the first output window is the first window that conains the start time of WHERE clause. The last output window is the last window that contains the end time of WHERE clause.
::: :::

View File

@ -97,6 +97,7 @@ NULL, NULL_F, VALUE, VALUE_F 这几种填充模式针对不同场景区别如下
1. 使用 FILL 语句的时候可能生成大量的填充输出,务必指定查询的时间区间。针对每次查询,系统可返回不超过 1 千万条具有插值的结果。 1. 使用 FILL 语句的时候可能生成大量的填充输出,务必指定查询的时间区间。针对每次查询,系统可返回不超过 1 千万条具有插值的结果。
2. 在时间维度聚合中,返回的结果中时间序列严格单调递增。 2. 在时间维度聚合中,返回的结果中时间序列严格单调递增。
3. 如果查询对象是超级表,则聚合函数会作用于该超级表下满足值过滤条件的所有表的数据。如果查询中没有使用 PARTITION BY 语句,则返回的结果按照时间序列严格单调递增;如果查询中使用了 PARTITION BY 语句分组,则返回结果中每个 PARTITION 内按照时间序列严格单调递增。 3. 如果查询对象是超级表,则聚合函数会作用于该超级表下满足值过滤条件的所有表的数据。如果查询中没有使用 PARTITION BY 语句,则返回的结果按照时间序列严格单调递增;如果查询中使用了 PARTITION BY 语句分组,则返回结果中每个 PARTITION 内按照时间序列严格单调递增。
4. Fill输出的起始和结束窗口与WHERE条件的时间范围有关, 如增序Fill时, 第一个窗口是包含WHERE条件开始时间的第一个窗口, 最后一个窗口是包含WHERE条件结束时间的最后一个窗口。
::: :::

View File

@ -59,6 +59,7 @@ static void revisedFillStartKey(SFillOperatorInfo* pInfo, SSDataBlock* pBlock, i
static void destroyFillOperatorInfo(void* param); static void destroyFillOperatorInfo(void* param);
static void doApplyScalarCalculation(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_t order, int32_t scanFlag); static void doApplyScalarCalculation(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_t order, int32_t scanFlag);
static void fillResetPrevForNewGroup(SFillInfo* pFillInfo); static void fillResetPrevForNewGroup(SFillInfo* pFillInfo);
static void reviseFillStartAndEndKey(SFillOperatorInfo* pInfo, int32_t order);
static void doHandleRemainBlockForNewGroupImpl(SOperatorInfo* pOperator, SFillOperatorInfo* pInfo, static void doHandleRemainBlockForNewGroupImpl(SOperatorInfo* pOperator, SFillOperatorInfo* pInfo,
SResultInfo* pResultInfo, int32_t order) { SResultInfo* pResultInfo, int32_t order) {
@ -74,7 +75,7 @@ static void doHandleRemainBlockForNewGroupImpl(SOperatorInfo* pOperator, SFillOp
blockDataCleanup(pInfo->pRes); blockDataCleanup(pInfo->pRes);
doApplyScalarCalculation(pOperator, pInfo->existNewGroupBlock, order, scanFlag); doApplyScalarCalculation(pOperator, pInfo->existNewGroupBlock, order, scanFlag);
revisedFillStartKey(pInfo, pInfo->existNewGroupBlock, order); reviseFillStartAndEndKey(pOperator->info, order);
int64_t ts = (order == TSDB_ORDER_ASC) ? pInfo->existNewGroupBlock->info.window.ekey int64_t ts = (order == TSDB_ORDER_ASC) ? pInfo->existNewGroupBlock->info.window.ekey
: pInfo->existNewGroupBlock->info.window.skey; : pInfo->existNewGroupBlock->info.window.skey;
@ -258,7 +259,7 @@ static SSDataBlock* doFillImpl(SOperatorInfo* pOperator) {
if (pInfo->curGroupId == 0 || (pInfo->curGroupId == pInfo->pRes->info.id.groupId)) { if (pInfo->curGroupId == 0 || (pInfo->curGroupId == pInfo->pRes->info.id.groupId)) {
if (pInfo->curGroupId == 0 && taosFillNotStarted(pInfo->pFillInfo)) { if (pInfo->curGroupId == 0 && taosFillNotStarted(pInfo->pFillInfo)) {
revisedFillStartKey(pInfo, pBlock, order); reviseFillStartAndEndKey(pInfo, order);
} }
pInfo->curGroupId = pInfo->pRes->info.id.groupId; // the first data block pInfo->curGroupId = pInfo->pRes->info.id.groupId; // the first data block
@ -549,3 +550,31 @@ _error:
taosMemoryFreeClear(pOperator); taosMemoryFreeClear(pOperator);
return code; return code;
} }
static void reviseFillStartAndEndKey(SFillOperatorInfo* pInfo, int32_t order) {
int64_t skey, ekey, next;
if (order == TSDB_ORDER_ASC) {
skey = taosTimeTruncate(pInfo->win.skey, &pInfo->pFillInfo->interval);
taosFillUpdateStartTimestampInfo(pInfo->pFillInfo, skey);
ekey = taosTimeTruncate(pInfo->win.ekey, &pInfo->pFillInfo->interval);
next = ekey;
while (next < pInfo->win.ekey) {
next = taosTimeAdd(ekey, pInfo->pFillInfo->interval.sliding, pInfo->pFillInfo->interval.slidingUnit,
pInfo->pFillInfo->interval.precision);
ekey = next > pInfo->win.ekey ? ekey : next;
}
pInfo->win.ekey = ekey;
} else {
assert(order == TSDB_ORDER_DESC);
skey = taosTimeTruncate(pInfo->win.skey, &pInfo->pFillInfo->interval);
next = skey;
while (next < pInfo->win.skey) {
next = taosTimeAdd(skey, pInfo->pFillInfo->interval.sliding, pInfo->pFillInfo->interval.slidingUnit,
pInfo->pFillInfo->interval.precision);
skey = next > pInfo->win.skey ? skey : next;
}
taosFillUpdateStartTimestampInfo(pInfo->pFillInfo, skey);
pInfo->win.ekey = taosTimeTruncate(pInfo->win.ekey, &pInfo->pFillInfo->interval);
}
}

View File

@ -1,6 +1,11 @@
import queue
import random
from fabric2.runners import threading
from pandas._libs import interval
import taos import taos
import sys import sys
from util.common import TDCom
from util.log import * from util.log import *
from util.sql import * from util.sql import *
from util.cases import * from util.cases import *
@ -8,6 +13,7 @@ from util.cases import *
class TDTestCase: class TDTestCase:
updatecfgDict = {'asynclog': 0, 'ttlUnit': 1, 'ttlPushInterval': 5, 'ratioOfVnodeStreamThrea': 4, 'numOfVnodeQueryThreads': 80}
def init(self, conn, logSql, replicaVar=1): def init(self, conn, logSql, replicaVar=1):
self.replicaVar = int(replicaVar) self.replicaVar = int(replicaVar)
@ -15,7 +21,115 @@ class TDTestCase:
#tdSql.init(conn.cursor()) #tdSql.init(conn.cursor())
tdSql.init(conn.cursor(), logSql) # output sql.txt file tdSql.init(conn.cursor(), logSql) # output sql.txt file
def generate_fill_range(self, data_start: int, data_end: int, interval: int, step: int):
ret = []
begin = data_start - 10 * interval
end = data_end + 10 * interval
for i in range(begin, end, step):
for j in range(begin, end, step):
ret.append((i,j))
return ret
def check_fill_range(self, where_start, where_end, res_asc, res_desc, sql: str, interval):
if len(res_asc) != len(res_desc):
tdLog.exit(f"err, asc desc with different rows, asc: {len(res_asc)}, desc: {len(res_desc)} sql: {sql}")
if len(res_asc) == 0:
tdLog.info(f'from {where_start} to {where_end} no rows returned')
return
asc_first = res_asc[0]
asc_last = res_asc[-1]
desc_first = res_desc[0]
desc_last = res_desc[-1]
if asc_first[0] != desc_last[0] or asc_last[0] != desc_first[0]:
tdLog.exit(f'fill sql different row data {sql}: asc<{asc_first[0].timestamp()}, {asc_last[0].timestamp()}>, desc<{desc_last[0].timestamp()}, {desc_first[0].timestamp()}>')
else:
tdLog.info(f'from {where_start} to {where_end} same time returned asc<{asc_first[0].timestamp()}, {asc_last[0].timestamp()}>, desc<{desc_last[0].timestamp()}, {desc_first[0].timestamp()}> interval: {interval}')
def generate_partition_by(self):
val = random.random()
if val < 0.6:
return ""
elif val < 0.8:
return "partition by location"
else:
return "partition by tbname"
def generate_fill_interval(self):
ret = []
#intervals = [60, 90, 120, 300, 3600]
intervals = [120, 300, 3600]
for i in range(0, len(intervals)):
for j in range(0, i+1):
ret.append((intervals[i], intervals[j]))
return ret
def generate_fill_sql(self, where_start, where_end, fill_interval: tuple):
partition_by = self.generate_partition_by()
where = f'where ts >= {where_start} and ts < {where_end}'
sql = f'select first(_wstart), last(_wstart) from (select _wstart, _wend, count(*) from test.meters {where} {partition_by} interval({fill_interval[0]}s) sliding({fill_interval[1]}s) fill(NULL)'
sql_asc = sql + " order by _wstart asc) t"
sql_desc = sql + " order by _wstart desc) t"
return sql_asc, sql_desc
def fill_test_thread_routine(self, cli: TDSql, interval, data_start, data_end, step):
ranges = self.generate_fill_range(data_start, data_end, interval[0], step)
for range in ranges:
sql_asc, sql_desc = self.generate_fill_sql(range[0], range[1], interval)
cli.query(sql_asc, queryTimes=1)
asc_res = cli.queryResult
cli.query(sql_desc, queryTimes=1)
desc_res = cli.queryResult
self.check_fill_range(range[0], range[1], asc_res,desc_res , sql_asc, interval)
def fill_test_task_routine(self, tdCom: TDCom, queue: queue.Queue):
cli = tdCom.newTdSql()
while True:
m: list = queue.get()
if len(m) == 0:
break
interval = m[0]
range = m[1]
sql_asc, sql_desc = self.generate_fill_sql(range[0], range[1], interval)
cli.query(sql_asc, queryTimes=1)
asc_res = cli.queryResult
cli.query(sql_desc, queryTimes=1)
desc_res = cli.queryResult
self.check_fill_range(range[0], range[1], asc_res,desc_res , sql_asc, interval)
cli.close()
def schedule_fill_test_tasks(self):
num: int = 20
threads = []
tdCom = TDCom()
q: queue.Queue = queue.Queue()
for _ in range(num):
t = threading.Thread(target=self.fill_test_task_routine, args=(tdCom, q))
t.start()
threads.append(t)
data_start = 1500000000000
data_end = 1500319968000
step = 30000000
fill_intervals: list[tuple] = self.generate_fill_interval()
for interval in fill_intervals:
ranges = self.generate_fill_range(data_start, data_end, interval[0], step)
for r in ranges:
q.put([interval, r])
for _ in range(num):
q.put([])
for t in threads:
t.join()
def test_fill_range(self):
os.system('taosBenchmark -t 10 -n 10000 -v 8 -S 32000 -y')
self.schedule_fill_test_tasks()
tdSql.execute('drop database test')
def run(self): def run(self):
self.test_fill_range()
dbname = "db" dbname = "db"
tbname = "tb" tbname = "tb"

View File

@ -176,7 +176,7 @@ class TDTestCase:
def test_query_with_window(self): def test_query_with_window(self):
# time window # time window
tdSql.query("select sum(c_int_empty) from st where ts > '2024-01-01 00:00:00.000' and ts <= '2024-01-01 14:00:00.000' interval(5m) sliding(1m) fill(value, 10);") tdSql.query("select sum(c_int_empty) from st where ts > '2024-01-01 00:00:00.000' and ts <= '2024-01-01 14:00:00.000' interval(5m) sliding(1m) fill(value, 10);")
tdSql.checkRows(841) tdSql.checkRows(845)
tdSql.checkData(0, 0, 10) tdSql.checkData(0, 0, 10)
tdSql.query("select _wstart, _wend, sum(c_int) from st where ts > '2024-01-01 00:00:00.000' and ts <= '2024-01-01 14:00:00.000' interval(5m) sliding(1m);") tdSql.query("select _wstart, _wend, sum(c_int) from st where ts > '2024-01-01 00:00:00.000' and ts <= '2024-01-01 14:00:00.000' interval(5m) sliding(1m);")