Execution and tracing

Execution model

SQLite does execution in two steps. First a statement is prepared, which verifies the syntax, tables and fields and converts the statement into an internal representation. The prepared statement is then run. Execution stops when a row is available, there is an error or the statement is complete.

The Cursor.execute() method automatically does the preparing and starts execution. If none of the statements return rows then execution will go to the end. If a row is returned then you use the cursor as an iterator. Execution will resume as necessary to return each result row.

However this means that if you don’t read the rows returned then the rest of your statements won’t be executed. APSW will detect unexecuted previous statements and generate an exception. For example:

>>> cursor.execute("select * from foo ; create table bar(x,y,z)")
>>> cursor.execute("create table bam(x,y,z)")
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
apsw.IncompleteExecutionError: Error: there are still remaining sql statements to execute

Because I didn’t read the results of select * from foo then the following create table command didn’t have a chance to get executed. On the next execute that condition is detected and an exception raised.

Multi-threading and re-entrancy

ASPW lets you use SQLite in multi-threaded programs and will let other threads execute while SQLite is working. It checks at start that SQLite was compiled in threadsafe mode which is the default. The GIL is released when when SQLite APIs are called, and re-acquired while running any Python code.

You cannot use the same cursor object in multiple threads concurrently to execute statements. APSW will detect this and raise an ThreadingViolationError. It is safe to use the object serially (eg calling Cursor.execute() in one thread and iterator in another. You also can’t do things like try to close() a Connection concurrently in two threads.

A cursor object can only be executing one query at a time. You cannot issue a new query from inside a trace function or from a user defined function or collation since these are called while executing a query. You can however make new cursors and use those without issue. You may want to remember the Connection object when you set your trace or user defined functions.

64 bit hosts

APSW is tested and works correctly on 32 and 64 bit hosts. SQLite is limited to 32 bit quantities for strings, blobs, number of columns etc even when compiled for 64 bit. You will get a TooBigError if trying to use strings and blobs larger than 1 gigabyte.

Statement Cache

Each Connection maintains a cache mapping SQL queries to a prepared statement to avoid the overhead of repreparing queries that are executed multiple times. This is a classic trade off using more memory to reduce CPU consumption.

By default there are up to 100 entries in the cache. Once the cache is full, the least recently used item is discarded to make space for new items.

You should pick a larger cache size if you have more than 100 unique queries that you run. For example if you have 101 different queries you run in order then the cache will not help.

If you are using authorizers then be aware authorizer callback is only called while statements are being prepared. You can specify zero which will disable the statement cache completely, use use can_cache = False flag to execute/executemany.

Tracing

You can install tracers on cursors or connections as an easy way of seeing exactly what gets executed and what is returned. The tracers can also abort execution and cause different values to be returned. This is very useful for diagnostics and testing without having to modify your main code.

Note

You cannot issue new execute statements against the cursor your tracer was called from. If you would like to make more queries in the tracer then do them from a new cursor object. For example:

def exec_tracer(cursor, sql, bindings):
  cursor.connection.cursor().execute("insert into log values(?,?)", (sql,str(bindings)))
  return True

Execution Tracer

The execution tracer is called after an SQL statement has been prepared. (ie syntax errors will have caused an exception during preparation so you won’t see them with a tracer). It is called with three arguments.

cursor
The cursor executing the statement

sql
The SQL text being executed

bindings
The bindings being used. This may be None, a dictionary or a tuple.

If the tracer return value is False then execution is aborted with an ExecTraceAbort exception. See the example.

Execution tracers can be installed on a specific cursor by setting Cursor.exec_trace or for all cursors by setting Connection.exec_trace, with the cursor tracer taking priority.

Row Tracer

The row tracer is called before each row is returned. It is called with two arguments.

cursor
The cursor returning the row

row
A tuple of the values about to be returned

Whatever you return from the tracer is what is actually returned to the caller of execute(). If you return None then the whole row is skipped. See the example.

Row tracers can be installed on a specific cursor by setting Cursor.row_trace or for all cursors by setting Connection.row_trace, with the cursor tracer taking priority.

APSW Trace

APSW includes a tracer that lets you easily trace SQL execution as well as providing a summary report without modifying your code.

$ python3 -m apsw.trace [apswtrace options] yourscript.py [your options]

The following options are available:

$ python3 -m apsw.trace --help
Usage: apswtrace.py [options] pythonscript.py [pythonscriptoptions]

This script runs a Python program that uses APSW and reports on SQL queries
without modifying the program.  This is done by using connection_hooks and
registering row and execution tracers.  See APSW documentation for more
details on the output.

Options:
  -h, --help            show this help message and exit
  -o OUTPUT, --output=OUTPUT
                        Where to send the output.  Use a filename, a single
                        dash for stdout, or the words stdout and stderr.
                        [stdout]
  -s, --sql             Log SQL statements as they are executed. [False]
  -r, --rows            Log returned rows as they are returned (turns on sql).
                        [False]
  -t, --timestamps      Include timestamps in logging
  -i, --thread          Include thread id in logging
  -l LENGTH, --length=LENGTH
                        Max amount of a string to print [30]
  --no-report           A summary report is normally generated at program
                        exit.  This turns off the report and saves memory.
  --report-items=N      How many items to report in top lists [15]
  --reports=REPORTS     Which reports to show
                        [summary,popular,aggregate,individual]

This is sample output with the following options: –sql, –rows, –timestamps, –thread

1e0e5a0 0.152 7fccea8456e0 OPEN: ":memory:" unix READWRITE|CREATE
1f72ac0 0.161 7fccea8456e0 OPEN: "testdb" unix READWRITE|CREATE
1f6b8d0 0.162 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.162 7fccea8456e0 SQL: create table foo(x,y,z)
1f6b8d0 0.239 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.239 7fccea8456e0 SQL: insert into foo values(?,?,?) BINDINGS: ("kjfhgk", "gkjlfdhgjkhsdfkjg", "gklsdfjgkldfjhnbnvc,mnxb,mnxcv..")
1f6b8d0 0.242 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.242 7fccea8456e0 SQL: insert into foo values(?,?,?) BINDINGS: ("gdfklhj", ":gjkhgfdsgfd", "gjkfhgjkhdfkjh")
1f6b8d0 0.244 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.245 7fccea8456e0 SQL: insert into foo values(?,?,?) BINDINGS: ("gdfjkhg", "gkjlfd", "")
1f6b8d0 0.247 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.247 7fccea8456e0 SQL: insert into foo values(?,?,?) BINDINGS: (1, 2, 30)
1f6b8d0 0.257 7fccea8456e0 CURSORFROM: 1f72ac0 DB: "testdb"
1f6b8d0 0.257 7fccea8456e0 SQL: select longest(x,y,z) from foo
1f6b8d0 0.257 7fccea8456e0 ROW: ("gklsdfjgkldfjhnbnvc,mnxb,mnxcv..")

Each row starts with the following fields:

id
This is the id of the Cursor or Connection. You can easily filter the log if you just want to find out what happened on a specific cursor or connection.

timestamp
This is time since the program started in seconds

threadid
The unique thread identifier

The remainder of the line has one of the following forms:

OPEN: “dbname” vfs open_flags
A Connection has been opened. The dbname is the filename exactly as given in the call to Connection. vfs is the name of the VFS used to open the database. open_flags is the set of flags supplied with the leading SQLITE_OPEN prefix omitted.

CURSORFROM: connectionid DB: “dbname”
A cursor has been allocated. The id at the beginning of this row is of the new cursor. connectionid is the id of the Connection it was created from. The dbname is provided for convenience. This message is logged the first time a cursor issues a query.

SQL: query BINDINGS: bindings
A query was issued on a cursor.

ROW: row
A result row was returned by a cursor.

A report is also generated by default. This is example output from running the test suite. When calculating time for queries, your code execution time is included as well. For example if your query returned 10 rows and you slept for 1 second on reading each row then the time for the query will be recorded as 10 seconds. Because you can have multiple queries active at the same time, as well as across multiple threads, the total processing time can be larger than the program run time. The processing time is only recorded for queries that have no results or where you read all the result rows. Processing time also includes waiting time on busy connections.

APSW TRACE SUMMARY REPORT

Program run time                    83.073 seconds
Total connections                   1308
Total cursors                       3082
Number of threads used for queries  21
Total queries                       127973
Number of distinct queries          578
Number of rows returned             2369
Time spent processing queries       120.530 seconds

This shows how many times each query was run.

MOST POPULAR QUERIES

 121451 insert into foo values(?)
insert into abc values(1,2,?)
select x from foo
select timesten(x) from foo where x=? order by x
select * from foo
update t1 set b=b||a||b
begin
create table foo(x,y)
insert into foo values(1,2)
rollback
pragma locking_mode=exclusive
insert into t1 values(2, 'abcdefghijklmnopqrstuvwxyz')
insert into t1 values(1, 'abcdefghijklmnopqrstuvwxyz')
insert into t1 select 4-a, b from t2
insert into foo values(date('now'), date('now'))

This shows how many times a query was run and the sum of the processing times in seconds. The begin immediate query illustrates how time spent busy waiting is included.

LONGEST RUNNING - AGGREGATE

 94.305 select timesten(x) from foo where x=? order by x
 120637   12.941 select * from foo
  4.115 begin immediate
 121449    2.179 insert into foo values(?)
  1.509 insert into abc values(1,2,?)
  1.380 create index foo_x on foo(x)
  0.715 update t1 set b=b||a||b
  0.420 insert into foo values(?,?)
  0.241 create table t1(a unique, b)
  0.206 create table foo(x,y)
  0.170 create table abc(a,b,c)
  0.165 insert into foo values(?,?,?)
  0.158 select row,x,snap(x) from foo
  0.150 insert into foo values(1,2)
  0.127 insert into foo values(date('now'), date('now'))

This shows the longest running queries with time in seconds.

LONGEST RUNNING - INDIVIDUAL

001 begin immediate
377 create index foo_x on foo(x)
102 begin immediate
944 select timesten(x) from foo where x=? order by x
893 select timesten(x) from foo where x=? order by x
817 select timesten(x) from foo where x=? order by x
816 select timesten(x) from foo where x=? order by x
786 select timesten(x) from foo where x=? order by x
783 select timesten(x) from foo where x=? order by x
713 select timesten(x) from foo where x=? order by x
701 select timesten(x) from foo where x=? order by x
651 select timesten(x) from foo where x=? order by x
646 select timesten(x) from foo where x=? order by x
631 select timesten(x) from foo where x=? order by x
620 select timesten(x) from foo where x=? order by x