提交 88d1d8fa 编写于 作者: haoranc's avatar haoranc

test:add test case of tsbs query

上级 ebb2f34d
......@@ -76,9 +76,19 @@ class TDTestCase:
tdSql.query("insert into testsnode(ts,c1,c2,c3,c4) SELECT ts,avg(velocity) as mean_velocity,name,driver,fleet FROM readings WHERE ts > 1451606400000 AND ts <= 1451606460000 partition BY name,driver,fleet,ts interval(10m);")
# test interval fill
tdSql.query("SELECT name,floor(avg(velocity)/10)/floor(avg(velocity)/10) AS mv FROM readings WHERE name!='' AND ts > '2016-01-01T00:00:00Z' AND ts < '2016-01-05T00:00:01Z' partition by name interval(10m) fill(value,0) ;")
# test paitition interval fill
# tdSql.query("SELECT name,floor(avg(velocity)/10)/floor(avg(velocity)/10) AS mv FROM readings WHERE name!='' AND ts > '2016-01-01T00:00:00Z' AND ts < '2016-01-05T00:00:01Z' partition by name interval(10m) fill(value,0) ;")
# # test partition interval limit
# tdSql.query("SELECT ts,model,floor(2*(sum(nzs)/count(nzs)))/floor(2*(sum(nzs)/count(nzs))) AS broken_down FROM (SELECT ts,model, status/status AS nzs FROM diagnostics WHERE ts >= '2016-01-01T00:00:00Z' AND ts < '2016-01-05T00:00:01Z' ) WHERE ts >= '2016-01-01T00:00:00Z' AND ts < '2016-01-05T00:00:01Z' partition BY model,ts interval(10m) limit 10;")
# tdSql.checkRows(10)
# test partition interval Pseudo time-column
tdSql.query("SELECT count(ms1)/144 FROM (SELECT _wstartts as ts1,model, fleet,avg(status) AS ms1 FROM diagnostics WHERE ts >= '2016-01-01T00:00:00Z' AND ts < '2016-01-05T00:00:01Z' partition by model, fleet interval(10m)) WHERE ts1 >= '2016-01-01T00:00:00Z' AND ts1 < '2016-01-05T00:00:01Z' AND ms1<1;")
# test
def run(self): # sourcery skip: extract-duplicate-method, remove-redundant-fstring
tdLog.printNoPrefix("==========step1:create database and table,insert data ==============")
self.prepareData()
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册