Skip to main content

TDengine connector for python

Project description

TDengine Connector for Python

Github Workflow PyPI Version PyPI Downloads CodeCov
workflow PyPI PyPI codecov

TDengine connector for Python enables python programs to access TDengine, using an API which is compliant with the Python DB API 2.0 (PEP-249). It contains two modules:

  1. The taos module. It uses TDengine C client library for client server communications.
  2. The taosrest module. It wraps TDengine RESTful API to Python DB API 2.0 (PEP-249). With this module, you do not need to install the TDengine C client library.

Install taospy

You can use pip to install the connector from PyPI:

pip install taospy

Or with git url:

pip install git+https://github.com/taosdata/taos-connector-python.git

Note: taospy v2.7.2 requirs Python 3.6+. The early versions of taospy from v2.5.0 to v2.7.1 require Python 3.7+.

Source Code

TDengine connector for Python source code is hosted on GitHub.

Install taos-ws-py

Install with taospy

pip install taospy[ws]

Install taos-ws-py only

pip install taos-ws-py

Note: The taosws module is provided by taos-ws-py package separately from v2.7.2. It is part of early version of taospy. taos-ws-py requires Python 3.7+.

Query with PEP-249 API using taosws

import taosws

# all parameters are optional
conn = taosws.connect("taosws://root:taosdata@localhost:6041")
cursor = conn.cursor()

cursor.execute("show databases")
results: list[tuple] = cursor.fetchall()
for row in results:
    print(row)

You can pass an optional req_id in the parameters.

import taosws

# all parameters are optional
conn = taosws.connect("taosws://root:taosdata@localhost:6041")
cursor = conn.cursor()

cursor.execute("show databases", req_id=1)
results: list[tuple] = cursor.fetchall()
for row in results:
    print(row)

Query with query method using taosws

from taosws import *

conn = connect("taosws://root:taosdata@localhost:6041")
result = conn.query("show databases")

num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

You can pass an optional req_id in the parameters.

from taosws import *

conn = connect("taosws://root:taosdata@localhost:6041")
result = conn.query("show databases", req_id=1)

num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

Read with Pandas using taosws

Method one

import pandas
import taosws

conn = taosws.connect("taosws://root:taosdata@localhost:6041")
df: pandas.DataFrame = pandas.read_sql("show databases", conn)
df

Method Two

import pandas
from sqlalchemy import create_engine

engine = create_engine("taosws://root:taosdata@localhost:6041")
df: pandas.DataFrame = pandas.read_sql("show databases", engine)
df

Examples for taosrest Module

Query with PEP-249 API

import taosrest

# all parameters are optional
conn = taosrest.connect(url="http://localhost:6041",
                        user="root",
                        password="taosdata")
cursor = conn.cursor()

cursor.execute("show databases")
results: list[tuple] = cursor.fetchall()
for row in results:
    print(row)

You can pass an optional req_id in the parameters.

import taosrest

# all parameters are optional
conn = taosrest.connect(url="http://localhost:6041",
                        user="root",
                        password="taosdata")
cursor = conn.cursor()

cursor.execute("show databases", req_id=1)
results: list[tuple] = cursor.fetchall()
for row in results:
    print(row)

Query with query method

from taosrest import connect, TaosRestConnection, Result

conn: TaosRestConnection = connect()
result: Result = conn.query("show databases")

num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

You can pass an optional req_id in the parameters.

from taosrest import connect, TaosRestConnection, Result

conn: TaosRestConnection = connect()
result: Result = conn.query("show databases", req_id=1)

num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

Read with Pandas

Method one

import pandas
import taos

conn = taos.connect()
df: pandas.DataFrame = pandas.read_sql("select * from log.logs", conn)

Method two

import pandas
import taosrest

conn = taosrest.connect()
df: pandas.DataFrame = pandas.read_sql("select * from log.logs", conn)

Method three

import pandas
from sqlalchemy import create_engine

engine = create_engine("taosrest://root:taosdata@localhost:6041")
df: pandas.DataFrame = pandas.read_sql("select * from log.logs", engine)

Examples for taos Module

Connect options

Supported config options:

  • config: TDengine client configuration directory, by default use "/etc/taos/".
  • host: TDengine server host, by default use "localhost".
  • user: TDengine user name, default is "root".
  • password: TDengine user password, default is "taosdata".
  • database: Default connection database name, empty if not set.
  • timezone: Timezone for timestamp type (which is datetime object with tzinfo in python) no matter what the host's timezone is.
import taos

# 1. with empty options, connect TDengine by default options
#   that means:
#     - use /etc/taos/taos.cfg as default configuration file
#     - use "localhost" if not set in config file
#     - use "root" as default username
#     - use "taosdata" as default password
#     - use 6030 as default port if not set in config file
#     - use local timezone datetime as timestamp
conn = taos.connect()
# 2. with full set options, default db: log, use UTC datetime.
conn = taos.connect(host='localhost',
                    user='root',
                    password='taosdata',
                    database='log',
                    config='/etc/taos',
                    timezone='UTC')

Note that, the datetime formatted string will contain timezone information when timezone set. For example:

# without timezone(local timezone depends on your machine)
'1969-12-31 16:00:00'
# with timezone UTC
'1969-12-31 16:00:00+00:00'

Query with PEP-249 API

import taos

conn = taos.connect()
cursor = conn.cursor()

cursor.execute("show databases")
results = cursor.fetchall()
for row in results:
    print(row)

cursor.close()
conn.close()

You can pass an optional req_id in the parameters.

import taos

conn = taos.connect()
cursor = conn.cursor()

cursor.execute("show databases", req_id=1)
results = cursor.fetchall()
for row in results:
    print(row)

cursor.close()
conn.close()

Execute many

Method execute_many is supported:

There are two ways to use execute_many.

The first way is to pass a data_set as a list of dictionaries, where each dictionary will be expanded into a complete statement.

import taos

conn = taos.connect()
cursor = conn.cursor()

db_name = "test_db"

cursor.execute(f"DROP DATABASE IF EXISTS {db_name}")
cursor.execute(f"CREATE DATABASE {db_name}")
cursor.execute(f"USE {db_name}")

cursor.execute("create stable stb (ts timestamp, v1 int) tags(t1 int)")

create_table_data = [
    {
        "name": "tb1",
        "t1": 1,
    },
    {
        "name": "tb2",
        "t1": 2,
    },
    {
        "name": "tb3",
        "t1": 3,
    }
]

cursor.execute_many(
    "create table {name} using stb tags({t1})",
    create_table_data,
)

The second way is to pass a data_set as a list of tuples, where each tuple represents a different row of the same table, and each value within the tuple will become a data row in the SQL statement. All the data together will form a complete SQL statement.

import taos

conn = taos.connect()
cursor = conn.cursor()

db_name = "test_db"

cursor.execute(f"USE {db_name}")

data_set = [
    ('2018-10-03 14:38:05.100', 219),
    ('2018-10-03 14:38:15.300', 218),
    ('2018-10-03 14:38:16.800', 221),
]

table_name = "tb1"

cursor.execute_many(f"insert into {table_name} values", data_set)

Example: Insert many lines in one execute

Query with objective API

import taos

conn = taos.connect()
conn.execute("create database if not exists pytest")

result = conn.query("show databases")
num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

result.close()
conn.execute("drop database pytest")
conn.close()

You can pass an optional req_id in the parameters.

import taos

conn = taos.connect()
conn.execute("create database if not exists pytest", req_id=1)

result = conn.query("show databases")
num_of_fields = result.field_count
for field in result.fields:
    print(field)

for row in result:
    print(row)

result.close()
conn.execute("drop database pytest")
conn.close()

Query with async API

from taos import *
from ctypes import *
import time


def fetch_callback(p_param, p_result, num_of_rows):
    print("fetched ", num_of_rows, "rows")
    p = cast(p_param, POINTER(Counter))
    result = TaosResult(p_result)

    if num_of_rows == 0:
        print("fetching completed")
        p.contents.done = True
        result.close()
        return

    if num_of_rows < 0:
        p.contents.done = True
        result.check_error(num_of_rows)
        result.close()
        return None

    for row in result.rows_iter(num_of_rows):
        # print(row)
        pass

    p.contents.count += result.row_count
    result.fetch_rows_a(fetch_callback, p_param)


def query_callback(p_param, p_result, code):
    # type: (c_void_p, c_void_p, c_int) -> None
    if p_result is None:
        return

    result = TaosResult(p_result)
    if code == 0:
        result.fetch_rows_a(fetch_callback, p_param)

    result.check_error(code)


class Counter(Structure):
    _fields_ = [("count", c_int), ("done", c_bool)]

    def __str__(self):
        return "{ count: %d, done: %s }" % (self.count, self.done)


def test_query(conn):
    # type: (TaosConnection) -> None
    counter = Counter(count=0)
    conn.query_a("select * from log.log", query_callback, byref(counter))

    while not counter.done:
        print("wait query callback")
        time.sleep(1)

    print(counter)
    conn.close()


if __name__ == "__main__":
    test_query(connect())

You can pass an optional req_id in the parameters.

from taos import *
from ctypes import *
import time


def fetch_callback(p_param, p_result, num_of_rows):
    print("fetched ", num_of_rows, "rows")
    p = cast(p_param, POINTER(Counter))
    result = TaosResult(p_result)

    if num_of_rows == 0:
        print("fetching completed")
        p.contents.done = True
        result.close()
        return

    if num_of_rows < 0:
        p.contents.done = True
        result.check_error(num_of_rows)
        result.close()
        return None

    for row in result.rows_iter(num_of_rows):
        # print(row)
        pass

    p.contents.count += result.row_count
    result.fetch_rows_a(fetch_callback, p_param)


def query_callback(p_param, p_result, code):
    # type: (c_void_p, c_void_p, c_int) -> None
    if p_result is None:
        return

    result = TaosResult(p_result)
    if code == 0:
        result.fetch_rows_a(fetch_callback, p_param)

    result.check_error(code)


class Counter(Structure):
    _fields_ = [("count", c_int), ("done", c_bool)]

    def __str__(self):
        return "{ count: %d, done: %s }" % (self.count, self.done)


def test_query(conn):
    # type: (TaosConnection) -> None
    counter = Counter(count=0)
    conn.query_a("select * from log.log", query_callback, byref(counter), req_id=1)

    while not counter.done:
        print("wait query callback")
        time.sleep(1)

    print(counter)
    conn.close()


if __name__ == "__main__":
    test_query(connect())

Statement API - Bind row after row

from taos import *

conn = connect()

dbname = "pytest_taos_stmt"
conn.execute("drop database if exists %s" % dbname)
conn.execute("create database if not exists %s" % dbname)
conn.select_db(dbname)

conn.execute(
    "create table if not exists log(ts timestamp, bo bool, nil tinyint, \
        ti tinyint, si smallint, ii int, bi bigint, tu tinyint unsigned, \
        su smallint unsigned, iu int unsigned, bu bigint unsigned, \
        ff float, dd double, bb binary(100), nn nchar(100), tt timestamp)",
)

stmt = conn.statement("insert into log values(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)")

params = new_bind_params(16)
params[0].timestamp(1626861392589)
params[1].bool(True)
params[2].tinyint(None)
params[3].tinyint(2)
params[4].smallint(3)
params[5].int(4)
params[6].bigint(5)
params[7].tinyint_unsigned(6)
params[8].smallint_unsigned(7)
params[9].int_unsigned(8)
params[10].bigint_unsigned(9)
params[11].float(10.1)
params[12].double(10.11)
params[13].binary("hello")
params[14].nchar("stmt")
params[15].timestamp(1626861392589)
stmt.bind_param(params)

params[0].timestamp(1626861392590)
params[15].timestamp(None)
stmt.bind_param(params)
stmt.execute()

assert stmt.affected_rows == 2

result = conn.query("select * from log")

for row in result:
    print(row)

Statement API - Bind multi rows

from taos import *

conn = connect()

dbname = "pytest_taos_stmt"
conn.execute("drop database if exists %s" % dbname)
conn.execute("create database if not exists %s" % dbname)
conn.select_db(dbname)

conn.execute(
    "create table if not exists log(ts timestamp, bo bool, nil tinyint, \
        ti tinyint, si smallint, ii int, bi bigint, tu tinyint unsigned, \
        su smallint unsigned, iu int unsigned, bu bigint unsigned, \
        ff float, dd double, bb binary(100), nn nchar(100), tt timestamp)",
)

stmt = conn.statement("insert into log values(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)")

params = new_multi_binds(16)
params[0].timestamp((1626861392589, 1626861392590, 1626861392591))
params[1].bool((True, None, False))
params[2].tinyint([-128, -128, None])  # -128 is tinyint null
params[3].tinyint([0, 127, None])
params[4].smallint([3, None, 2])
params[5].int([3, 4, None])
params[6].bigint([3, 4, None])
params[7].tinyint_unsigned([3, 4, None])
params[8].smallint_unsigned([3, 4, None])
params[9].int_unsigned([3, 4, None])
params[10].bigint_unsigned([3, 4, None])
params[11].float([3, None, 1])
params[12].double([3, None, 1.2])
params[13].binary(["abc", "dddafadfadfadfadfa", None])
params[14].nchar(["涛思数据", None, "a long string with 中文字符"])
params[15].timestamp([None, None, 1626861392591])
stmt.bind_param_batch(params)
stmt.execute()

assert stmt.affected_rows == 3

result = conn.query("select * from log")
for row in result:
    print(row)

Subscription

import taos
import random

conn = taos.connect()
dbname = "pytest_taos_subscribe"
conn.execute("drop database if exists %s" % dbname)
conn.execute("create database if not exists %s" % dbname)
conn.select_db(dbname)
conn.execute("create table if not exists log(ts timestamp, n int)")
for i in range(10):
    conn.execute("insert into log values(now, %d)" % i)

sub = conn.subscribe(False, "test", "select * from log", 1000)
print("# consume from begin")
for ts, n in sub.consume():
    print(ts, n)

print("# consume new data")
for i in range(5):
    conn.execute("insert into log values(now, %d)(now+1s, %d)" % (i, i))
    result = sub.consume()
    for ts, n in result:
        print(ts, n)

sub.close(True)
print("# keep progress consume")
sub = conn.subscribe(False, "test", "select * from log", 1000)
result = sub.consume()
rows = result.fetch_all()
# consume from latest subscription needs root privilege(for /var/lib/taos).
assert result.row_count == 0
print("## consumed ", len(rows), "rows")

print("# consume with a stop condition")
for i in range(10):
    conn.execute("insert into log values(now, %d)" % random.randint(0, 10))
    result = sub.consume()
    try:
        ts, n = next(result)
        print(ts, n)
        if n > 5:
            result.stop_query()
            print("## stopped")
            break
    except StopIteration:
        continue

sub.close()
# sub.close()

conn.execute("drop database if exists %s" % dbname)
# conn.close()

Subscription asynchronously with callback

from taos import *
from ctypes import *

import time


def subscribe_callback(p_sub, p_result, p_param, errno):
    # type: (c_void_p, c_void_p, c_void_p, c_int) -> None
    print("# fetch in callback")
    result = TaosResult(c_void_p(p_result))
    result.check_error(errno)
    for row in result.rows_iter():
        ts, n = row()
        print(ts, n)


def test_subscribe_callback(conn):
    # type: (TaosConnection) -> None
    dbname = "pytest_taos_subscribe_callback"
    try:
        print("drop if exists")
        conn.execute("drop database if exists %s" % dbname)
        print("create database")
        conn.execute("create database if not exists %s" % dbname)
        print("create table")
        # conn.execute("use %s" % dbname)
        conn.execute("create table if not exists %s.log(ts timestamp, n int)" % dbname)

        print("# subscribe with callback")
        sub = conn.subscribe(False, "test", "select * from %s.log" % dbname, 1000, subscribe_callback)

        for i in range(10):
            conn.execute("insert into %s.log values(now, %d)" % (dbname, i))
            time.sleep(0.7)

        sub.close()

        conn.execute("drop database if exists %s" % dbname)
        # conn.close()
    except Exception as err:
        conn.execute("drop database if exists %s" % dbname)
        # conn.close()
        raise err


if __name__ == "__main__":
    test_subscribe_callback(connect())

Insert with line protocol

import taos
from taos import SmlProtocol, SmlPrecision

conn = taos.connect()
dbname = "pytest_line"
conn.execute("drop database if exists %s" % dbname)
conn.execute("create database if not exists %s precision 'us'" % dbname)
conn.select_db(dbname)

lines = [
    'st,t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"pass",c2=false,c4=4f64 1626006833639000000',
]
conn.schemaless_insert(lines, taos.SmlProtocol.LINE_PROTOCOL, taos.SmlPrecision.NOT_CONFIGURED)
print("inserted")

conn.schemaless_insert(lines, taos.SmlProtocol.LINE_PROTOCOL, taos.SmlPrecision.NOT_CONFIGURED)

result = conn.query("show tables")
for row in result:
    print(row)

conn.execute("drop database if exists %s" % dbname)

You can pass an optional req_id in the parameters.

import taos
from taos import SmlProtocol, SmlPrecision

conn = taos.connect()
dbname = "pytest_line"
conn.execute("drop database if exists %s" % dbname)
conn.execute("create database if not exists %s precision 'us'" % dbname)
conn.select_db(dbname)

lines = [
    'st,t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"pass",c2=false,c4=4f64 1626006833639000000',
]
conn.schemaless_insert(lines, taos.SmlProtocol.LINE_PROTOCOL, taos.SmlPrecision.NOT_CONFIGURED, req_id=1)
print("inserted")

conn.schemaless_insert(lines, taos.SmlProtocol.LINE_PROTOCOL, taos.SmlPrecision.NOT_CONFIGURED, req_id=2)

result = conn.query("show tables")
for row in result:
    print(row)

conn.execute("drop database if exists %s" % dbname)

Insert with schemaless raw data.

import taos
from taos import utils
from taos import TaosConnection
from taos.cinterface import *
from taos.error import OperationalError, SchemalessError

conn = taos.connect()
dbname = "taos_schemaless_insert"
try:
    conn.execute("drop database if exists %s" % dbname)

    if taos.IS_V3:
        conn.execute("create database if not exists %s schemaless 1 precision 'ns'" % dbname)
    else:
        conn.execute("create database if not exists %s update 2 precision 'ns'" % dbname)

    conn.select_db(dbname)

    lines = '''st,t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000
    st,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin, abc",c2=true,c4=5f64,c5=5f64,c6=7u64 1626006933640000000
    stf,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''

    res = conn.schemaless_insert_raw(lines, 1, 0)
    print("affected rows: ", res)
    assert (res == 3)

    lines = '''stf,t1=5i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''
    res = conn.schemaless_insert_raw(lines, 1, 0)
    print("affected rows: ", res)
    assert (res == 1)

    result = conn.query("select * from st")
    dict2 = result.fetch_all_into_dict()
    print(dict2)
    print(result.row_count)

    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()
    assert (result.row_count == 2)

    # error test
    lines = ''',t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000'''
    try:
        res = conn.schemaless_insert_raw(lines, 1, 0)
        print(res)
        # assert(False)
    except SchemalessError as err:
        print('**** error: ', err)
        # assert (err.msg == 'Invalid data format')

    result = conn.query("select * from st")
    print(result.row_count)
    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()

    conn.execute("drop database if exists %s" % dbname)
    conn.close()
except InterfaceError as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
except SchemalessError as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
except Exception as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
    raise err

Pass optional ttl in the parameters.

import taos
from taos import utils
from taos import TaosConnection
from taos.cinterface import *
from taos.error import OperationalError, SchemalessError

conn = taos.connect()
dbname = "taos_schemaless_insert"
try:
    conn.execute("drop database if exists %s" % dbname)

    if taos.IS_V3:
        conn.execute("create database if not exists %s schemaless 1 precision 'ns'" % dbname)
    else:
        conn.execute("create database if not exists %s update 2 precision 'ns'" % dbname)

    conn.select_db(dbname)

    lines = '''st,t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000
    st,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin, abc",c2=true,c4=5f64,c5=5f64,c6=7u64 1626006933640000000
    stf,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''

    ttl = 1000
    res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl)
    print("affected rows: ", res)
    assert (res == 3)

    lines = '''stf,t1=5i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''
    ttl = 1000
    res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl)
    print("affected rows: ", res)
    assert (res == 1)

    result = conn.query("select * from st")
    dict2 = result.fetch_all_into_dict()
    print(dict2)
    print(result.row_count)

    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()
    assert (result.row_count == 2)

    # error test
    lines = ''',t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000'''
    try:
        ttl = 1000
        res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl)
        print(res)
        # assert(False)
    except SchemalessError as err:
        print('**** error: ', err)
        # assert (err.msg == 'Invalid data format')

    result = conn.query("select * from st")
    print(result.row_count)
    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()

    conn.execute("drop database if exists %s" % dbname)
    conn.close()
except InterfaceError as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
except Exception as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
    raise err

Pass optional req_id in the parameters.

import taos
from taos import utils
from taos import TaosConnection
from taos.cinterface import *
from taos.error import OperationalError, SchemalessError

conn = taos.connect()
dbname = "taos_schemaless_insert"
try:
    conn.execute("drop database if exists %s" % dbname)

    if taos.IS_V3:
        conn.execute("create database if not exists %s schemaless 1 precision 'ns'" % dbname)
    else:
        conn.execute("create database if not exists %s update 2 precision 'ns'" % dbname)

    conn.select_db(dbname)

    lines = '''st,t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000
    st,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin, abc",c2=true,c4=5f64,c5=5f64,c6=7u64 1626006933640000000
    stf,t1=4i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''

    ttl = 1000
    req_id = utils.gen_req_id()
    res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl, req_id=req_id)
    print("affected rows: ", res)
    assert (res == 3)

    lines = '''stf,t1=5i64,t3="t4",t2=5f64,t4=5f64 c1=3i64,c3=L"passitagin_stf",c2=false,c5=5f64,c6=7u64 1626006933641000000'''
    ttl = 1000
    req_id = utils.gen_req_id()
    res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl, req_id=req_id)
    print("affected rows: ", res)
    assert (res == 1)

    result = conn.query("select * from st")
    dict2 = result.fetch_all_into_dict()
    print(dict2)
    print(result.row_count)

    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()
    assert (result.row_count == 2)

    # error test
    lines = ''',t1=3i64,t2=4f64,t3="t3" c1=3i64,c3=L"passit",c2=false,c4=4f64 1626006833639000000'''
    try:
        ttl = 1000
        req_id = utils.gen_req_id()
        res = conn.schemaless_insert_raw(lines, 1, 0, ttl=ttl, req_id=req_id)
        print(res)
        # assert(False)
    except SchemalessError as err:
        print('**** error: ', err)
        # assert (err.msg == 'Invalid data format')

    result = conn.query("select * from st")
    print(result.row_count)
    all = result.rows_iter()
    for row in all:
        print(row)
    result.close()

    conn.execute("drop database if exists %s" % dbname)
    conn.close()
except InterfaceError as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
except Exception as err:
    conn.execute("drop database if exists %s" % dbname)
    conn.close()
    print(err)
    raise err

Read with Pandas

Method one

import pandas
import taos

conn = taos.connect()
df: pandas.DataFrame = pandas.read_sql("select * from log.logs", conn)

Method Two

import pandas
from sqlalchemy import create_engine

engine = create_engine("taos://root:taosdata@localhost:6030/log")
df: pandas.DataFrame = pandas.read_sql("select * from logs", engine)

Limitation

  • taosrest is designed to use with taosAdapter. If your TDengine version is older than v2.4.0.0, taosAdapter may not be available.

License

We use MIT license for Python connector.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

taospy-2.7.9.tar.gz (42.7 kB view details)

Uploaded Source

Built Distribution

taospy-2.7.9-py3-none-any.whl (44.3 kB view details)

Uploaded Python 3

File details

Details for the file taospy-2.7.9.tar.gz.

File metadata

  • Download URL: taospy-2.7.9.tar.gz
  • Upload date:
  • Size: 42.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.2.0 CPython/3.10.6 Linux/5.15.0-1040-azure

File hashes

Hashes for taospy-2.7.9.tar.gz
Algorithm Hash digest
SHA256 3fe9d434d727433ff5e76b2020979824a830bdb4585f07bb2dfad5668e2435dd
MD5 95d42c06c739a58dd1ca6cd0d0161257
BLAKE2b-256 f794670a69737a044133c2a29a3207f792c2b8867d37ee68b1e14794e7a89fb3

See more details on using hashes here.

File details

Details for the file taospy-2.7.9-py3-none-any.whl.

File metadata

  • Download URL: taospy-2.7.9-py3-none-any.whl
  • Upload date:
  • Size: 44.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.2.0 CPython/3.10.6 Linux/5.15.0-1040-azure

File hashes

Hashes for taospy-2.7.9-py3-none-any.whl
Algorithm Hash digest
SHA256 5e92ab518fb69b6f57504e521d919f47909f2dcf8a07bf59e20d8a5011252050
MD5 a8584b38deac1e15a817802f35a215e6
BLAKE2b-256 500fd1b0a874e75ddc8c910894479b82853eed97e2ecd815e20e5cedab6c1803

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page