V4.98 13 Feb 2008 (c) 2000-2008 John Lim (jlim#natsoft.com.my)
This software is dual licensed using BSD-Style and LGPL. This means you can use it in compiled proprietary and commercial products.
Useful ADOdb links: Download Other Docs
This module, part of the ADOdb package, provides both CLI and HTML interfaces for viewing key performance indicators of your database. This is very useful because web apps such as the popular phpMyAdmin currently do not provide effective database health monitoring tools. The module provides the following:
$perf->HealthCheck()
or $perf->HealthCheckCLI()
. $perf->UI()
.
This UI displays:
$perf->DBParameter('data cache hit
ratio')
returns this very important statistic in a database
independant manner. ADOdb also has the ability to log all SQL executed, using LogSQL. All SQL logged can be analyzed through the performance monitor UI. In the View SQL mode, we categorize the SQL into 3 types:
Each query is hyperlinked to a description of the query plan, and every PHP script that executed that query is also shown.
Please note that the information presented is a very basic database health check, and does not provide a complete overview of database performance. Although some attempt has been made to make it work across multiple databases in the same way, it is impossible to do so. For the health check, we do try to display the following key database parameters for all drivers:
You will need to connect to the database as an administrator to view most of the parameters.
Code improvements as very welcome, particularly adding new database parameters and automated tuning hints.
Currently, the following drivers: mysql, postgres, oci8, mssql, informix and db2 are supported. To create a new performance monitor, call NewPerfMonitor( ) as demonstrated below:
<?php
include_once('adodb.inc.php');
session_start(); # session variables required for monitoring
$conn = ADONewConnection($driver);
$conn->Connect($server,$user,$pwd,$db);
$perf =& NewPerfMonitor($conn);
$perf->UI($pollsecs=5);
?>
It is also possible to retrieve a single database parameter:
$size = $perf->DBParameter('data cache size');
Thx to Fernando Ortiz for the informix module.
function UI($pollsecs=5)
Creates a web-based user interface for performance monitoring. When you click on Poll, server statistics will be displayed every $pollsecs seconds. See Usage above.
Since 4.11, we allow users to enter and run SQL interactively via the "Run SQL" link. To disable this for security reasons, set this constant before calling $perf->UI().
define('ADODB_PERF_NO_RUN_SQL',1);
Sample output follows below:
ADOdb
Performance Monitor for localhost, db=test PostgreSQL 7.3.2 on i686-pc-cygwin, compiled by GCC gcc (GCC) 3.2 20020927 (prerelease) |
Performance Stats View SQL View Tables Poll Stats |
postgres7 |
||
Parameter | Value | Description |
Ratios | ||
statistics collector | TRUE | Value must be TRUE to enable hit ratio statistics (stats_start_collector,stats_row_level and stats_block_level must be set to true in postgresql.conf) |
data cache hit ratio | 99.7967555299239 | |
IO | ||
data reads | 125 | |
data writes | 21.78125000000000000 | Count of inserts/updates/deletes * coef |
Data Cache | ||
data cache buffers | 640 | Number of cache buffers. Tuning |
cache blocksize | 8192 | (estimate) |
data cache size | 5M | |
operating system cache size | 80M | (effective cache size) |
Memory Usage | ||
sort buffer size | 1M | Size of sort buffer (per query) |
Connections | ||
current connections | 0 | |
max connections | 32 | |
Parameters | ||
rollback buffers | 8 | WAL buffers |
random page cost | 4 | Cost of doing a seek (default=4). See random_page_cost |
function HealthCheck()
Returns database health check parameters as a HTML table. You will need to echo or print the output of this function,
function HealthCheckCLI()
Returns database health check parameters formatted for a command line interface. You will need to echo or print the output of this function. Sample output for mysql:
-- Ratios --
MyISAM cache hit ratio => 56.5635738832
InnoDB cache hit ratio => 0
sql cache hit ratio => 0
-- IO --
data reads => 2622
data writes => 2415.5
-- Data Cache --
MyISAM data cache size => 512K
BDB data cache size => 8388600
InnoDB data cache size => 8M
-- Memory Pools --
read buffer size => 131072
sort buffer size => 65528
table cache => 4
-- Connections --
current connections => 3
max connections => 100
function Poll($pollSecs=5)
Run in infinite loop, displaying the following information every $pollSecs. This will not work properly if output buffering is enabled. In the example below, $pollSecs=3:
Accumulating statistics...
Time WS-CPU% Hit% Sess Reads/s Writes/s
11:08:30 0.7 56.56 1 0.0000 0.0000
11:08:33 1.8 56.56 2 0.0000 0.0000
11:08:36 11.1 56.55 3 2.5000 0.0000
11:08:39 9.8 56.55 2 3.1121 0.0000
11:08:42 2.8 56.55 1 0.0000 0.0000
11:08:45 7.4 56.55 2 0.0000 1.5000
WS-CPU% is the Web Server CPU load of the server that PHP is running from (eg. the database client), and not the database. The Hit% is the data cache hit ratio. Sess is the current number of sessions connected to the database. If you are using persistent connections, this should not change much. The Reads/s and Writes/s are synthetic values to give the viewer a rough guide to I/O, and are not to be taken literally.
function SuspiciousSQL($numsql=10)
Returns SQL which have high average execution times as a HTML table. Each sql statement is hyperlinked to a new window which details the execution plan and the scripts that execute this SQL.
The number of statements returned is determined by $numsql. Data is taken from the adodb_logsql table, where the sql statements are logged when $connection->LogSQL(true) is enabled. The adodb_logsql table is populated using $conn->LogSQL.
For Oracle, Ixora Suspicious SQL returns a list of SQL statements that are most cache intensive as a HTML table. These are data intensive SQL statements that could benefit most from tuning.
function ExpensiveSQL($numsql=10)
Returns SQL whose total execution time (avg time * #executions) is high as a HTML table. Each sql statement is hyperlinked to a new window which details the execution plan and the scripts that execute this SQL.
The number of statements returned is determined by $numsql. Data is taken from the adodb_logsql table, where the sql statements are logged when $connection->LogSQL(true) is enabled. The adodb_logsql table is populated using $conn->LogSQL.
For Oracle, Ixora Expensive SQL returns a list of SQL statements that are taking the most CPU load when run.
function InvalidSQL($numsql=10)
Returns a list of invalid SQL as an HTML table.
Data is taken from the adodb_logsql table, where the sql statements are logged when $connection->LogSQL(true) is enabled.
function Tables($orderby=1)
Returns information on all tables in a database, with the first two fields containing the table name and table size, the remaining fields depend on the database driver. If $orderby is set to 1, it will sort by name. If $orderby is set to 2, then it will sort by table size. Some database drivers (mssql and mysql) will ignore the $orderby clause. For postgresql, the information is up-to-date since the last vacuum. Not supported currently for db2.
Raw functions return values without any formatting.
function DBParameter($paramname)
Returns the value of a database parameter, such as $this->DBParameter("data cache size").
function CPULoad()
Returns the CPU load of the database client (NOT THE SERVER) as a percentage. Only works for Linux and Windows. For Windows, WMI must be available.
New in adodb 4.97/5.03 is this global variable, which controls whether sql timings which are too small are not saved. Currently it defaults to 0.05 (seconds). This means that all sql's which are faster than 0.05 seconds to execute are not saved.
To create new database parameters, you need to understand $settings. The $settings data structure is an associative array. Each element of the array defines a database parameter. The key is the name of the database parameter. If no key is defined, then it is assumed to be a section break, and the value is the name of the section break. If this is too confusing, looking at the source code will help a lot!
Each database parameter is itself an array consisting of the following elements:
Example from MySQL, table_cache database parameter:
'table cache' => array('CACHE', # category code
array("show variables", 'table_cache'), # array (type 1b)
'Number of tables to keep open'), # description
db2 informix mysql mssql oci8 postgres
db2 |
||
Parameter | Value | Description |
Ratios | ||
data cache hit ratio | 0 | |
Data Cache | ||
data cache buffers | 250 | See tuning reference. |
cache blocksize | 4096 | |
data cache size | 1000K | |
Connections | ||
current connections | 2 |
informix |
||
Parameter | Val ue | Description |
Ratios | ||
data cache hit ratio | 95.89 | |
IO | ||
data reads | 1883884 | Page reads |
data writes | 1716724 | Page writes |
Connections | ||
current connections | 263.0 | Number of sessions |
mysql |
||
Parameter | Value | Description |
Ratios | ||
MyISAM cache hit ratio | 56.5658301822 | Cache ratio should be at least 90% |
InnoDB cache hit ratio | 0 | Cache ratio should be at least 90% |
sql cache hit ratio | 0 | |
IO | ||
data reads | 2622 | Number of selects (Key_reads is not accurate) |
data writes | 2415.5 | Number of inserts/updates/deletes * coef (Key_writes is not accurate) |
Data Cache | ||
MyISAM data cache size | 512K | |
BDB data cache size | 8388600 | |
InnoDB data cache size | 8M | |
Memory Pools | ||
read buffer size | 131072 | (per session) |
sort buffer size | 65528 | Size of sort buffer (per session) |
table cache | 4 | Number of tables to keep open |
Connections | ||
current connections | 3 | |
max connections | 100 |
mssql |
||
Parameter | Value | Description |
Ratios | ||
data cache hit ratio | 99.9999694824 | |
prepared sql hit ratio | 99.7738579828 | |
adhoc sql hit ratio | 98.4540169133 | |
IO | ||
data reads | 2858 | |
data writes | 1438 | |
Data Cache | ||
data cache size | 4362 | in K |
Connections | ||
current connections | 14 | |
max connections | 32767 |
oci8 |
||
Parameter | Value | Description |
Ratios | ||
data cache hit ratio | 96.98 | |
sql cache hit ratio | 99.96 | |
IO | ||
data reads | 842938 | |
data writes | 16852 | |
Data Cache | ||
data cache buffers | 3072 | Number of cache buffers |
data cache blocksize | 8192 | |
data cache size | 48M | shared_pool_size |
Memory Pools | ||
java pool size | 0 | java_pool_size |
sort buffer size | 512K | sort_area_size (per query) |
user session buffer size | 8M | large_pool_size |
Connections | ||
current connections | 1 | |
max connections | 170 | |
data cache utilization ratio | 88.46 | Percentage of data cache actually in use |
user cache utilization ratio | 91.76 | Percentage of user cache (large_pool) actually in use |
rollback segments | 11 | |
Transactions | ||
peak transactions | 24 | Taken from high-water-mark |
max transactions | 187 | max transactions / rollback segments < 3.5 (or transactions_per_rollback_segment) |
Parameters | ||
cursor sharing | EXACT | Cursor reuse strategy. Recommended is FORCE (8i+) or SIMILAR (9i+). See cursor_sharing. |
index cache cost | 0 | % of indexed data blocks expected in the cache. Recommended is 20-80. Default is 0. See optimizer_index_caching. |
random page cost | 100 | Recommended is 10-50 for TP, and 50 for data warehouses. Default is 100. See optimizer_index_cost_adj. |
LOAD | EXECUTES | SQL_TEXT |
.73% | 89 | select u.name, o.name, t.spare1, t.pctfree$ from sys.obj$ o, sys.user$ u, sys.tab$ t where (bitand(t.trigflag, 1048576) = 1048576) and o.obj#=t.obj# and o.owner# = u.user# select i.obj#, i.flags, u.name, o.name from sys.obj$ o, sys.user$ u, sys.ind$ i where (bitand(i.flags, 256) = 256 or bitand(i.flags, 512) = 512) and (not((i.type# = 9) and bitand(i.flags,8) = 8)) and o.obj#=i.obj# and o.owner# = u.user# |
.84% | 3 | select /*+ RULE */ distinct tabs.table_name, tabs.owner , partitioned, iot_type , TEMPORARY, table_type, table_type_owner from DBA_ALL_TABLES tabs where tabs.owner = :own |
3.95% | 6 | SELECT round(count(1)*avg(buf.block_size)/1048576) FROM DBA_OBJECTS obj, V$BH bh, dba_segments seg, v$buffer_pool buf WHERE obj.object_id = bh.objd AND obj.owner != 'SYS' and obj.owner = seg.owner and obj.object_name = seg.segment_name and obj.object_type = seg.segment_type and seg.buffer_pool = buf.name and buf.name = 'DEFAULT' |
4.50% | 6 | SELECT round(count(1)*avg(tsp.block_size)/1048576) FROM DBA_OBJECTS obj, V$BH bh, dba_segments seg, dba_tablespaces tsp WHERE obj.object_id = bh.objd AND obj.owner != 'SYS' and obj.owner = seg.owner and obj.object_name = seg.segment_name and obj.object_type = seg.segment_type and seg.tablespace_name = tsp.tablespace_name |
57.34% | 9267 | select t.schema, t.name, t.flags, q.name from system.aq$_queue_tables t, sys.aq$_queue_table_affinities aft, system.aq$_queues q where aft.table_objno = t.objno and aft.owner_instance = :1 and q.table_objno = t.objno and q.usage = 0 and bitand(t.flags, 4+16+32+64+128+256) = 0 for update of t.name, aft.table_objno skip locked |
LOAD | EXECUTES | SQL_TEXT |
5.24% | 1 | select round(sum(bytes)/1048576) from dba_segments |
6.89% | 6 | SELECT round(count(1)*avg(buf.block_size)/1048576) FROM DBA_OBJECTS obj, V$BH bh, dba_segments seg, v$buffer_pool buf WHERE obj.object_id = bh.objd AND obj.owner != 'SYS' and obj.owner = seg.owner and obj.object_name = seg.segment_name and obj.object_type = seg.segment_type and seg.buffer_pool = buf.name and buf.name = 'DEFAULT' |
7.85% | 6 | SELECT round(count(1)*avg(tsp.block_size)/1048576) FROM DBA_OBJECTS obj, V$BH bh, dba_segments seg, dba_tablespaces tsp WHERE obj.object_id = bh.objd AND obj.owner != 'SYS' and obj.owner = seg.owner and obj.object_name = seg.segment_name and obj.object_type = seg.segment_type and seg.tablespace_name = tsp.tablespace_name |
33.69% | 89 | select u.name, o.name, t.spare1, t.pctfree$ from sys.obj$ o, sys.user$ u, sys.tab$ t where (bitand(t.trigflag, 1048576) = 1048576) and o.obj#=t.obj# and o.owner# = u.user# |
36.44% | 89 | select i.obj#, i.flags, u.name, o.name from sys.obj$ o, sys.user$ u, sys.ind$ i where (bitand(i.flags, 256) = 256 or bitand(i.flags, 512) = 512) and (not((i.type# = 9) and bitand(i.flags,8) = 8)) and o.obj#=i.obj# and o.owner# = u.user# |
postgres7 |
||
Parameter | Value | Description |
Ratios | ||
statistics collector | FALSE | Must be set to TRUE to enable hit ratio statistics (stats_start_collector,stats_row_level and stats_block_level must be set to true in postgresql.conf) |
data cache hit ratio | 99.9666031916603 | |
IO | ||
data reads | 15 | |
data writes | 0.000000000000000000 | Count of inserts/updates/deletes * coef |
Data Cache | ||
data cache buffers | 1280 | Number of cache buffers. Tuning |
cache blocksize | 8192 | (estimate) |
data cache size | 10M | |
operating system cache size | 80000K | (effective cache size) |
Memory Pools | ||
sort buffer size | 1M | Size of sort buffer (per query) |
Connections | ||
current connections | 13 | |
max connections | 32 | |
Parameters | ||
rollback buffers | 8 | WAL buffers |
random page cost | 4 | Cost of doing a seek (default=4). See random_page_cost |