LKET\-B2A

Section: User Commands (1)
Updated: 2007-11-17
Index Return to Main Contents
 

NAME

lket-b2a - Converting and dumping utility for LKET binary trace data

 

SYNOPSIS


lket-b2a OPTIONS IN_FILENAME...

 

DESCRIPTION

The trace data generated by LKET is in binary format by default for better performance and smaller size. lket-b2a is used to convert the binary trace data into readable data in ascii format and save them into local file or MySQL database for off-line trace analysis. It uses the per-cpu binary trace data files (stpd_cpu*) as inputs. You can use "stap -bM" with LKET to get those per-cpu files before using it.

The database used by lket-b2a is MySQL. So MySQL must be properly installed and configured in order to make lket-b2a able to dump the trace data into MySQL database.

 

OPTIONS

lket-b2a supports the following two options. They can be used together or alone but at least one of them should be specified:

-m
convert and dump trace data into MySQL database. For more details, See the following section.
-f
convert and dump trace data into local file. For more details, See the following section.
-n
name_flag. name_flag set to 0 means not printing the event description string and 1 means printing. Only valid with -f option. name_flag is set to 1 by default.
-i
id_flag. id_flag set to 0 means not printing event groupid and hookid and 1 means printing. Only valid with -f option. id_flag is set to 0 by default.
-a
appname_flag. appname_flag set to 0 means not printing process name and 1 means printing. Only valid with -f option. appname_flag is set to 1 by default.

 

DUMP TRACE DATA INTO LOCAL FILE

The generated output file is named lket.out. The following is an example:


root:/home/root/data> lket-b2a -f stpd_cpu*
root:/home/root/data> cat lket.out
LKET Magic:     0xAEFCDB6B
InitHdrLen:     9
Version Major:  1
Version Minor:  1
Big endian:     YES
Timing method:  do_gettimeofday()
Bits width:     64
Initial CPU timebase:   1596405 (cycles per microsecond)
[...]
2.527880 CPU:0 PID:2450 APPNAME:gnome-panel EVT_NAME:iosyscall.read.entry fd:3,buff_addr:-1081126904,count:32,
2.527887 CPU:0 PID:2450 APPNAME:gnome-panel EVT_NAME:iosyscall.read.return return:32,
2.534343 CPU:0 PID:1450 APPNAME:kjournald EVT_NAME:scsi.ioentry major:8,minor:0,sdev_state:2,request_addr:3806880208,
2.534359 CPU:0 PID:1450 APPNAME:kjournald EVT_NAME:scsi.iodispatching host:0,channel:0,lun:0,dev_id:0,dev_state:2,data_dir:1,reqbuf_addr:3248315840,reqbuf_len:8192,request_addr:3806880208,
[...]

 

DUMP TRACE DATA INTO MYSQL DATABASE

To avoid either reading the complete trace data into internal data structures to process, or filter through the raw trace data again and again to calculate the interesting metrics, lket-b2a supports convert and save the binary trace data into MySQL database to facilitate the query and calculation on the trace data.

lket-b2a creates the database name based on current time. And lket-b2a will not only save the trace data, but also some meta data into MySQL database too, such as trace header, trace table description.

The following is an example of navigating the trace data in MySQL database created by lket-b2a:

Use lket-b2a to convert and dump the binary trace data of LKET into MySQL database:


root:/home/root/data> lket-b2a -m stpd_cpu*

list all databases in MySQL:


mysql> show databases;
+------------------+
| Database         |
+------------------+
| DB20061023161626 |
| mysql            |
| test             |
+------------------+
3 rows in set (0.00 sec)

DB20061023161626 is the newly created database by "lket-b2a -m stpd_cpu*". We can list all the tables contained in DB20061023161626:


mysql> use DB20061023161626
Database changed
mysql> show tables;
+----------------------------+
| Tables_in_DB20061023161626 |
+----------------------------+
| 2_1                        |
| 2_2                        |
| 3_1                        |
| 3_3                        |
| 3_5                        |
| 4_1                        |
| 4_3                        |
| 4_4                        |
| 8_1                        |
| 8_3                        |
| appNameMap                 |
| table_desc                 |
| trace_header               |
+----------------------------+
13 rows in set (0.00 sec)

The table trace_header is used to store the trace header info:


mysql> select * from trace_header;
+-----------+-----------+------------+-----------------+------------+
| Major_Ver | Minor_Ver | Big_Endian | Timing_Method   | Bits_Width |
+-----------+-----------+------------+-----------------+------------+
|         1 |         1 |          2 | do_gettimeofday |         64 |
+-----------+-----------+------------+-----------------+------------+
1 row in set (0.00 sec)

The table appNameMap is used to store the mapping between PID and process name:


mysql> select * from appNameMap;
+-------+-----------------+
| pid   | pname           |
+-------+-----------------+
| 10764 | systemtap/0     |
| 10765 | systemtap/1     |
| 10766 | systemtap/2     |
| 10767 | systemtap/3     |
| 10768 | systemtap/4     |
| 10769 | systemtap/5     |
| 10770 | systemtap/6     |
| 10771 | systemtap/7     |
|     0 | swapper         |
|     1 | init            |
|     2 | migration/0     |
|     3 | ksoftirqd/0     |
|     4 | watchdog/0      |
|     5 | migration/1     |
[...]

Each event corresponds to one table which is named as groupid_hookid. For example table 8_1 corresponds to addevent.netdev.receive whose groupid is 8 and hookid is 1:


mysql> select * from table_desc;
+------------+----------------------------------+
| table_name | table_desc                       |
+------------+----------------------------------+
| 2_1        | addevent.syscall.entry           |
| 2_2        | addevent.syscall.return          |
| 3_1        | process_snapshot                 |
| 3_3        | addevent.process.execve          |
| 3_5        | addevent.process.fork            |
| 4_3        | addevent.ioscheduler.elv_next_re |
| 4_4        | addevent.ioscheduler.elv_next_re |
| 4_1        | addevent.ioscheduler.elv_add_req |
| 8_1        | addevent.netdev.receive          |
| 8_3        | addevent.netdev.transmit         |
+------------+----------------------------------+
10 rows in set (0.00 sec)

The hookid of a return type event( addevent.*.return ) should be an even number and its value should be the corresponding entry event hookid +1. For example, the hookid of addevent.syscall.return is 2 and the hookid of addevent.syscall.entry is 1. And what's more, there will be a new column named entry_usec for all return type event tables which is the timestamp of the entry of that event. The column entry_usec is created and calculated by lket-b2a on the fly while processing the binary trace data.


mysql> select * from 2_1 limit 4;
+---------+--------+--------+------------+--------+--------------+
| groupid | hookid | usec   | process_id | cpu_id | syscall      |
+---------+--------+--------+------------+--------+--------------+
|       2 |      1 |  10727 |      20922 |      2 | read         |
|       2 |      1 |  10746 |      20922 |      2 | read         |
|       2 |      1 | 729066 |       3605 |      5 | gettimeofday |
|       2 |      1 | 729086 |       3605 |      5 | gettimeofday |
+---------+--------+--------+------------+--------+--------------+
4 rows in set (0.00 sec)

mysql> select * from 2_2 limit 4;
+---------+--------+--------+------------+--------+------------+--------------+
| groupid | hookid | usec   | process_id | cpu_id | entry_usec | syscall      |
+---------+--------+--------+------------+--------+------------+--------------+
|       2 |      2 |  10742 |      20922 |      2 |      10727 | read         |
|       2 |      2 | 729072 |       3605 |      5 |     729066 | gettimeofday |
|       2 |      2 | 729089 |       3605 |      5 |     729086 | gettimeofday |
|       2 |      2 | 729100 |       3605 |      5 |     729096 | poll         |
+---------+--------+--------+------------+--------+------------+--------------+
4 rows in set (0.00 sec)

 

SEE ALSO

stap(1), lket(5)


 

Index

NAME
SYNOPSIS
DESCRIPTION
OPTIONS
DUMP TRACE DATA INTO LOCAL FILE
DUMP TRACE DATA INTO MYSQL DATABASE
SEE ALSO

linux.jgfs.net manual pages