DBAPI-OpenTracing 0.0.5

OpenTracing support for Python DB API

This package enables distributed tracing in Python Database API 2.0 compatible-clients via The OpenTracing Project. Once a production system contends with real concurrency or splits into many services, crucial (and formerly easy) tasks become difficult: user-facing latency optimization, root-cause analysis of backend errors, communication about distinct pieces of a now-distributed system, etc. Distributed tracing follows a request on its journey from inception to completion from mobile/browser all the way to the microservices.

As core services and libraries adopt OpenTracing, the application builder is no longer burdened with the task of adding basic tracing instrumentation to their own code. In this way, developers can build their applications with the tools they prefer and benefit from built-in tracing instrumentation. OpenTracing implementations exist for major distributed tracing systems and can be bound or swapped with a one-line configuration change.

If you want to learn more about the underlying Python API, visit the Python source code.

Installation

Run the following command:

$ pip install dbapi-opentracing

Usage

This DB API extension allows the tracing of database queries using the OpenTracing API. All that it requires is for a ConnectionTracing tracer to be initialized using an instance of an OpenTracing 2.0-compatible tracer and a DB API Connection object. You can either trace all commands sent to your database, or use a Cursor to trace individual requests.

Initialize

ConnectionTracing wraps the Connection and Tracer instances that are supported by the Python DB API and OpenTracing, respectively. To create a ConnectionTracing object, you can either pass in a tracer object directly or default to the opentracing.tracer global tracer:

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection, opentracing_tracer)

or

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client
import opentracing

opentracing.tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection)

For expanded usage with Psycopg connection factories, a PsycopgConnectionTracing class is also available. This is necessary for directly using extensions and extras functions on traced connections. Its non-default usage requires a lambda proxy or closure to account for the expected tracing arguments, while using the global tracer and default allow it to be passed directly:

from dbapi_opentracing import PsycopgConnectionTracing
import opentracing
import psycopg2

opentracing_tracer = ## some OpenTracing tracer implementation
my_tags = dict(some='tag')
tracing = psycopg2.connect(
    ..., connection_factory=lambda dsn: PsycopgConnectionTracing(dsn, tracer=opentracing_tracer, span_tags=my_tags)
)

# or to use all defaults
opentracing.tracer = opentracing_tracer
tracing = psycopg2.connect(..., connection_factory=PsycopgConnectionTracing)

ConnectionTracing Configuration

Along with optionally providing an OpenTracing 2.0-compatible tracer, ConnectionTracing also accepts a span_tags named argument and several traced method disabling flags: trace_execute, trace_executemany, trace_callproc, trace_commit, and trace_rollback to specify the command types you’d like not to trace (all are True by default).

from dbapi_opentracing import ConnectionTracing
from opentracing.ext import tags
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection, opentracing_tracer,
                            # span_tags will be used for all generated spans
                            span_tags={'Custom': 'Tag', tags.DATABASE_TYPE: 'PostgreSQL',
                                       tags.DATABASE_INSTANCE='myDatabase'},
                            trace_callproc=False, trace_commit=False)
# Note that the default OpenTracing 'db.type' tag will have 'sql' as a value.
# If a more specific type is desired, you can set it with the span_tags dictionary argument as shown.

Trace All Cursor Commands

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection, opentracing_tracer,
                            span_tags={'Custom': 'Tag'})  # span_tags will be used for all generated spans

# Please note that the default OpenTracing 'db.type' tag will have 'sql' as a value.
# If a more specific type is desired, you can set it with the span_tags dictionary argument

with tracing.cursor() as cursor:
    cursor.execute('SELECT * FROM TABLE')
    vals = cursor.fetchall()
    cursor.executemany('INSERT INTO TABLE VALUES (%s, %s)',
                       [('one', 'two'), ('three', 'four')])
    cursor.callproc('MyStoredProcedure')
tracing.commit()

Trace Specific Cursor Command Types

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection, opentracing_tracer,
                            span_tags={'Custom': 'Tag'})  # span_tags will be used for all generated spans

# Provide False values for optional trace_execute, trace_executemany, and/or trace_callproc named arguments
with tracing.cursor(trace_executemany=False, trace_callproc=False) as cursor:
    # Traced query
    cursor.execute('SELECT * FROM TABLE')
    vals = cursor.fetchall()
    # Untraced command
    cursor.executemany('INSERT INTO TABLE VALUES (%s, %s)',
                       [('one', 'two'), ('three', 'four')])
    # Untraced command
    cursor.callproc('MyStoredProcedure')
tracing.commit()

Trace All Connection Commands (implicit commit() and rollback())

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
tracing = ConnectionTracing(connection, opentracing_tracer)

with tracing as cursor:  # If DB API client supports Connection as context manager
    cursor.execute('SELECT * FROM TABLE')
    vals = cursor.fetchall()
    cursor.executemany('INSERT INTO TABLE VALUES (%s, %s)',
                       [('one', 'two'), ('three', 'four')])
    cursor.callproc('MyStoredProcedure')

Trace Specific Connection Command Types

from dbapi_opentracing import ConnectionTracing
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)
# Provide False values for optional trace_execute, trace_executemany, trace_callproc, trace_commit,
# and/or trace_rollback named arguments
tracing = ConnectionTracing(connection, opentracing_tracer, trace_execute=False, trace_commit=False)

with tracing as cursor:  # If DB API client supports Connection as context manager
    # Untraced query
    cursor.execute('SELECT * FROM TABLE')
    vals = cursor.fetchall()
    # Traced command
    cursor.executemany('INSERT INTO TABLE VALUES (%s, %s)',
                       [('one', 'two'), ('three', 'four')])
    # Traced command
    cursor.callproc('MyStoredProcedure')
# Implicit commit() is not traced because of named argument value

Trace Individual Commands Without Named Arguments

from dbapi_opentracing import Cursor
import db_api_compatible_client

opentracing_tracer = ## some OpenTracing tracer implementation
connection = db_api_compatible_client.connect(...)

with connection.cursor() as cursor:
    # Traced query
    Cursor(cursor, opentracing_tracer).execute('SELECT * FROM TABLE_ONE')
    # Traced query using opentracing.tracer default
    Cursor(cursor).execute('SELECT * FROM TABLE_TWO')
    # Traced query with custom tags
    Cursor(cursor, span_tags={'Query': 'Tag', 'Another': 'Tag'}).execute('SELECT * FROM TABLE_THREE')
    # Untraced command by using unmodified cursor instance
    cursor.executemany('INSERT INTO TABLE VALUES (%s, %s)',
                       [('one', 'two'), ('three', 'four')])

Further Information

If you’re interested in learning more about the OpenTracing standard, please visit opentracing.io or join the mailing list. If you would like to implement OpenTracing in your project and need help, feel free to send us a note at [e-mail].