monitoring database activity database activity monitoring A database administrator frequently wonders, What is the system doing right now? This chapter discusses how to find that out. Several tools are available for monitoring database activity and analyzing performance. Most of this chapter is devoted to describing PostgreSQL's statistics collector, but one should not neglect regular Unix monitoring programs such as ps, top, iostat, and vmstat. Also, once one has identified a poorly-performing query, further investigation may be needed using PostgreSQL's command. discusses EXPLAIN and other methods for understanding the behavior of an individual query. ps to monitor activity On most platforms, PostgreSQL modifies its command title as reported by ps, so that individual server processes can readily be identified. A sample display is $ ps auxww | grep ^postgres postgres 960 0.0 1.1 6104 1480 pts/1 SN 13:17 0:00 postmaster -i postgres 963 0.0 1.1 7084 1472 pts/1 SN 13:17 0:00 postgres: stats buffer process postgres 965 0.0 1.1 6152 1512 pts/1 SN 13:17 0:00 postgres: stats collector process postgres 998 0.0 2.3 6532 2992 pts/1 SN 13:18 0:00 postgres: tgl runbug 127.0.0.1 idle postgres 1003 0.0 2.4 6532 3128 pts/1 SN 13:19 0:00 postgres: tgl regression [local] SELECT waiting postgres 1016 0.1 2.4 6532 3080 pts/1 SN 13:19 0:00 postgres: tgl regression [local] idle in transaction (The appropriate invocation of ps varies across different platforms, as do the details of what is shown. This example is from a recent Linux system.) The first process listed here is the postmaster, the master server process. The command arguments shown for it are the same ones given when it was launched. The next two processes implement the statistics collector, which will be described in detail in the next section. (These will not be present if you have set the system not to start the statistics collector.) Each of the remaining processes is a server process handling one client connection. Each such process sets its command line display in the form postgres: user database host activity The user, database, and connection source host items remain the same for the life of the client connection, but the activity indicator changes. The activity may be idle (i.e., waiting for a client command), idle in transaction (waiting for client inside a BEGIN block), or a command type name such as SELECT. Also, waiting is attached if the server process is presently waiting on a lock held by another server process. In the above example we can infer that process 1003 is waiting for process 1016 to complete its transaction and thereby release some lock or other. Solaris requires special handling. You must use /usr/ucb/ps, rather than /bin/ps. You also must use two w flags, not just one. In addition, your original invocation of the postmaster command must have a shorter ps status display than that provided by each server process. If you fail to do all three things, the ps output for each server process will be the original postmaster command line. statistics PostgreSQL's statistics collector is a subsystem that supports collection and reporting of information about server activity. Presently, the collector can count accesses to tables and indexes in both disk-block and individual-row terms. It also supports determining the exact command currently being executed by other server processes. Since collection of statistics adds some overhead to query execution, the system can be configured to collect or not collect information. This is controlled by configuration parameters that are normally set in postgresql.conf. (See for details about setting configuration parameters.) The parameter must be set to true for the statistics collector to be launched at all. This is the default and recommended setting, but it may be turned off if you have no interest in statistics and want to squeeze out every last drop of overhead. (The savings is likely to be small, however.) Note that this option cannot be changed while the server is running. The parameters , , and control how much information is actually sent to the collector and thus determine how much run-time overhead occurs. These respectively determine whether a server process sends its current command string, disk-block-level access statistics, and row-level access statistics to the collector. Normally these parameters are set in postgresql.conf so that they apply to all server processes, but it is possible to turn them on or off in individual sessions using the command. (To prevent ordinary users from hiding their activity from the administrator, only superusers are allowed to change these parameters with SET.) Since the parameters stats_command_string, stats_block_level, and stats_row_level default to false, very few statistics are collected in the default configuration. Enabling one or more of these configuration variables will significantly enhance the amount of useful data produced by the statistics collector, at the expense of additional run-time overhead. Several predefined views, listed in , are available to show the results of statistics collection. Alternatively, one can build custom views using the underlying statistics functions. When using the statistics to monitor current activity, it is important to realize that the information does not update instantaneously. Each individual server process transmits new block and row access counts to the collector just before going idle; so a query or transaction still in progress does not affect the displayed totals. Also, the collector itself emits a new report at most once per PGSTAT_STAT_INTERVAL milliseconds (500 unless altered while building the server). So the displayed information lags behind actual activity. Current-query information is reported to the collector immediately, but is still subject to the PGSTAT_STAT_INTERVAL delay before it becomes visible. Another important point is that when a server process is asked to display any of these statistics, it first fetches the most recent report emitted by the collector process and then continues to use this snapshot for all statistical views and functions until the end of its current transaction. So the statistics will appear not to change as long as you continue the current transaction. This is a feature, not a bug, because it allows you to perform several queries on the statistics and correlate the results without worrying that the numbers are changing underneath you. But if you want to see new results with each query, be sure to do the queries outside any transaction block. View Name Description pg_stat_activity One row per server process, showing database OID, database name, process ID, user OID, user name, current query, time at which the current query began execution, time at which the process was started, and client's address and port number. The columns that report data on the current query are only available if the parameter stats_command_string has been turned on. Furthermore, these columns read as null unless the user examining the view is a superuser or the same as the user owning the process being reported on. (Note that because of the collector's reporting delay, the current query will only be up-to-date for long-running queries.) pg_stat_database One row per database, showing database OID, database name, number of active server processes connected to that database, number of transactions committed and rolled back in that database, total disk blocks read, and total buffer hits (i.e., block read requests avoided by finding the block already in buffer cache). pg_stat_all_tables For each table in the current database (including TOAST tables), the table OID, schema and table name, number of sequential scans initiated, number of live rows fetched by sequential scans, number of index scans initiated (over all indexes belonging to the table), number of live rows fetched by index scans, and numbers of row insertions, updates, and deletions. pg_stat_sys_tables Same as pg_stat_all_tables, except that only system tables are shown. pg_stat_user_tables Same as pg_stat_all_tables, except that only user tables are shown. pg_stat_all_indexes For each index in the current database, the table and index OID, schema, table and index name, number of index scans initiated on that index, number of index entries returned by index scans, and number of live table rows fetched by simple index scans using that index. pg_stat_sys_indexes Same as pg_stat_all_indexes, except that only indexes on system tables are shown. pg_stat_user_indexes Same as pg_stat_all_indexes, except that only indexes on user tables are shown. pg_statio_all_tables For each table in the current database (including TOAST tables), the table OID, schema and table name, number of disk blocks read from that table, number of buffer hits, numbers of disk blocks read and buffer hits in all indexes of that table, numbers of disk blocks read and buffer hits from that table's auxiliary TOAST table (if any), and numbers of disk blocks read and buffer hits for the TOAST table's index. pg_statio_sys_tables Same as pg_statio_all_tables, except that only system tables are shown. pg_statio_user_tables Same as pg_statio_all_tables, except that only user tables are shown. pg_statio_all_indexes For each index in the current database, the table and index OID, schema, table and index name, numbers of disk blocks read and buffer hits in that index. pg_statio_sys_indexes Same as pg_statio_all_indexes, except that only indexes on system tables are shown. pg_statio_user_indexes Same as pg_statio_all_indexes, except that only indexes on user tables are shown. pg_statio_all_sequences For each sequence object in the current database, the sequence OID, schema and sequence name, numbers of disk blocks read and buffer hits in that sequence. pg_statio_sys_sequences Same as pg_statio_all_sequences, except that only system sequences are shown. (Presently, no system sequences are defined, so this view is always empty.) pg_statio_user_sequences Same as pg_statio_all_sequences, except that only user sequences are shown. The per-index statistics are particularly useful to determine which indexes are being used and how effective they are. Beginning in PostgreSQL 8.1, indexes can be used either directly or via bitmap scans. In a bitmap scan the output of several indexes can be combined via AND or OR rules; so it is difficult to associate individual heap row fetches with specific indexes when a bitmap scan is used. Therefore, a bitmap scan increments the pg_stat_all_indexes.idx_tup_read count(s) for the index(es) it uses, and it increments the pg_stat_all_tables.idx_tup_fetch count for the table, but it does not affect pg_stat_all_indexes.idx_tup_fetch. Before PostgreSQL 8.1, the idx_tup_read and idx_tup_fetch counts were essentially always equal. Now they can be different even without considering bitmap scans, because idx_tup_read counts index entries retrieved from the index while idx_tup_fetch counts live rows fetched from the table; the latter will be less if any dead or not-yet-committed rows are fetched using the index. The pg_statio_ views are primarily useful to determine the effectiveness of the buffer cache. When the number of actual disk reads is much smaller than the number of buffer hits, then the cache is satisfying most read requests without invoking a kernel call. However, these statistics do not give the entire story: due to the way in which PostgreSQL handles disk I/O, data that is not in the PostgreSQL buffer cache may still reside in the kernel's I/O cache, and may therefore still be fetched without requiring a physical read. Users interested in obtaining more detailed information on PostgreSQL I/O behavior are advised to use the PostgreSQL statistics collector in combination with operating system utilities that allow insight into the kernel's handling of I/O. Other ways of looking at the statistics can be set up by writing queries that use the same underlying statistics access functions as these standard views do. These functions are listed in . The per-database access functions take a database OID as argument to identify which database to report on. The per-table and per-index functions take a table or index OID. (Note that only tables and indexes in the current database can be seen with these functions.) The per-server-process access functions take a server process number, which ranges from one to the number of currently active server processes. Function Return Type Description pg_stat_get_db_numbackends(oid) integer Number of active server processes for database pg_stat_get_db_xact_commit(oid) bigint Transactions committed in database pg_stat_get_db_xact_rollback(oid) bigint Transactions rolled back in database pg_stat_get_db_blocks_fetched(oid) bigint Number of disk block fetch requests for database pg_stat_get_db_blocks_hit(oid) bigint Number of disk block fetch requests found in cache for database pg_stat_get_numscans(oid) bigint Number of sequential scans done when argument is a table, or number of index scans done when argument is an index pg_stat_get_tuples_returned(oid) bigint Number of rows read by sequential scans when argument is a table, or number of index entries returned when argument is an index pg_stat_get_tuples_fetched(oid) bigint Number of table rows fetched by bitmap scans when argument is a table, or table rows fetched by simple index scans using the index when argument is an index pg_stat_get_tuples_inserted(oid) bigint Number of rows inserted into table pg_stat_get_tuples_updated(oid) bigint Number of rows updated in table pg_stat_get_tuples_deleted(oid) bigint Number of rows deleted from table pg_stat_get_blocks_fetched(oid) bigint Number of disk block fetch requests for table or index pg_stat_get_blocks_hit(oid) bigint Number of disk block requests found in cache for table or index pg_stat_get_backend_idset() setof integer Set of currently active server process numbers (from 1 to the number of active server processes). See usage example in the text pg_backend_pid() integer Process ID of the server process attached to the current session pg_stat_get_backend_pid(integer) integer Process ID of the given server process pg_stat_get_backend_dbid(integer) oid Database ID of the given server process pg_stat_get_backend_userid(integer) oid User ID of the given server process pg_stat_get_backend_activity(integer) text Active command of the given server process (null if the current user is not a superuser nor the same user as that of the session being queried, or stats_command_string is not on) pg_stat_get_backend_activity_start(integer) timestamp with time zone The time at which the given server process' currently executing query was started (null if the current user is not a superuser nor the same user as that of the session being queried, or stats_command_string is not on) pg_stat_get_backend_start(integer) timestamp with time zone The time at which the given server process was started, or null if the current user is not a superuser nor the same user as that of the session being queried pg_stat_get_backend_client_addr(integer) inet The IP address of the client connected to the given server process. Null if the connection is over a Unix domain socket. Also null if the current user is not a superuser nor the same user as that of the session being queried pg_stat_get_backend_client_port(integer) integer The IP port number of the client connected to the given server process. -1 if the connection is over a Unix domain socket. Null if the current user is not a superuser nor the same user as that of the session being queried pg_stat_reset() boolean Reset all currently collected statistics blocks_fetched minus blocks_hit gives the number of kernel read() calls issued for the table, index, or database; but the actual number of physical reads is usually lower due to kernel-level buffering. The function pg_stat_get_backend_idset provides a convenient way to generate one row for each active server process. For example, to show the PIDs and current queries of all server processes: SELECT pg_stat_get_backend_pid(s.backendid) AS procpid, pg_stat_get_backend_activity(s.backendid) AS current_query FROM (SELECT pg_stat_get_backend_idset() AS backendid) AS s; lock monitoring Another useful tool for monitoring database activity is the pg_locks system table. It allows the database administrator to view information about the outstanding locks in the lock manager. For example, this capability can be used to: View all the locks currently outstanding, all the locks on relations in a particular database, all the locks on a particular relation, or all the locks held by a particular PostgreSQL session. Determine the relation in the current database with the most ungranted locks (which might be a source of contention among database clients). Determine the effect of lock contention on overall database performance, as well as the extent to which contention varies with overall database traffic. Details of the pg_locks view appear in . For more information on locking and managing concurrency with PostgreSQL, refer to .