Statistics Collector

Redrock Postgres’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 tracks the total number of rows in each table, and information about analyze actions for each table. It can also count calls to user-defined functions and the total time spent in each one.

Redrock Postgres also supports reporting dynamic information about exactly what is going on in the system right now, such as the exact command currently being executed by other server processes, and which other connections exist in the system. This facility is independent of the collector process.

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 Chapter 19 for details about setting configuration parameters.)

The parameter track_activities enables monitoring of the current command being executed by any server process.

The parameter track_counts controls whether statistics are collected about table and index accesses.

The parameter track_functions enables tracking of usage of user-defined functions.

The parameter track_io_timing enables monitoring of block read and write times.

The parameter track_wait_events controls whether statistics are collected about wait events.

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 SET command. (To prevent ordinary users from hiding their activity from the administrator, only superusers are allowed to change these parameters with SET.)

The statistics collector transmits the collected information to other PostgreSQL processes through temporary files. These files are stored in the directory named by the stats_temp_directory parameter, pg_tmp/stats by default. For better performance, stats_temp_directory can be pointed at a RAM-based file system, decreasing physical I/O requirements. When the server shuts down cleanly, a permanent copy of the statistics data is stored in the pg_stat subdirectory, so that statistics can be retained across server restarts. When recovery is performed at server start (e.g., after immediate shutdown, server crash, and point-in-time recovery), all statistics counters are reset.

Several predefined views, listed in Table 1, are available to show the current state of the system. There are also several other views, listed in Table 2, available to show the results of statistics collection. Alternatively, one can build custom views using the underlying statistics functions, as discussed in Statistics Functions.

When using the statistics to monitor collected data, it is important to realize that the information does not update instantaneously. Each individual server process transmits new statistical 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 ms unless altered while building the server). So the displayed information lags behind actual activity. However, current-query information collected by track_activities is always up-to-date.

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 show static information as long as you continue the current transaction. Similarly, information about the current queries of all sessions is collected when any such information is first requested within a transaction, and the same information will be displayed throughout the 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. Alternatively, you can invoke pg_stat_clear_snapshot(), which will discard the current transaction’s statistics snapshot (if any). The next use of statistical information will cause a new snapshot to be fetched.

A transaction can also see its own statistics (as yet untransmitted to the collector) in the views pg_stat_xact_all_tables, pg_stat_xact_all_undos, pg_stat_xact_sys_tables, pg_stat_xact_user_tables, and pg_stat_xact_user_functions. These numbers do not act as stated above; instead they update continuously throughout the transaction.

Some of the information in the dynamic statistics views shown in Table 1 is security restricted. Ordinary users can only see all the information about their own sessions (sessions belonging to a role that they are a member of). In rows about other sessions, many columns will be null. Note, however, that the existence of a session and its general properties such as its sessions user and database are visible to all users. Superusers and members of the built-in role pg_read_all_stats (see also Section 21.5) can see all the information about all sessions.

Table 1. Dynamic Statistics Views

Table 2. Collected Statistics Views

The per-index statistics are particularly useful to determine which indexes are being used and how effective they are.

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 might still reside in the kernel’s I/O cache, and might 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.

The pg_stat_activity view will have one row per server process, showing information related to the current activity of that process.

Table 3. pg_stat_activity View

The wait_event and state columns are independent. If a backend is in the active state, it may or may not be waiting on some event. If the state is active and wait_event is non-null, it means that a query is being executed, but is being blocked somewhere in the system.

The pg_stat_wait_event view will contain one row for each wait event in the current database, showing statistics about wait information of every occured wait events.

Table 4. pg_stat_wait_event View

The pg_stat_replication view will contain one row per WAL sender process, showing statistics about replication to that sender’s connected standby server. Only directly connected standbys are listed; no information is available about downstream standby servers.

Table 5. pg_stat_replication View

The lag times reported in the pg_stat_replication view are measurements of the time taken for recent WAL to be written, flushed and replayed and for the sender to know about it. These times represent the commit delay that was (or would have been) introduced by each synchronous commit level, if the remote server was configured as a synchronous standby. For an asynchronous standby, the replay_lag column approximates the delay before recent transactions became visible to queries. If the standby server has entirely caught up with the sending server and there is no more WAL activity, the most recently measured lag times will continue to be displayed for a short time and then show NULL.

Lag times work automatically for physical replication. Logical decoding plugins may optionally emit tracking messages; if they do not, the tracking mechanism will simply display NULL lag.

The reported lag times are not predictions of how long it will take for the standby to catch up with the sending server assuming the current rate of replay. Such a system would show similar times while new WAL is being generated, but would differ when the sender becomes idle. In particular, when the standby has caught up completely, pg_stat_replication shows the time taken to write, flush and replay the most recent reported WAL location rather than zero as some users might expect. This is consistent with the goal of measuring synchronous commit and transaction visibility delays for recent write transactions. To reduce confusion for users expecting a different model of lag, the lag columns revert to NULL after a short time on a fully replayed idle system. Monitoring systems should choose whether to represent this as missing data, zero or continue to display the last known value.

The pg_stat_database view will contain one row for each database in the instance, plus one for the shared objects, showing database-wide statistics.

Table 6. pg_stat_database View

The pg_stat_all_tables view will contain one row for each table in the current database (including TOAST tables), showing statistics about accesses to that specific table. The pg_stat_user_tables and pg_stat_sys_tables views contain the same information, but filtered to only show user and system tables respectively.

Table 7. pg_stat_all_tables View

The pg_stat_all_undos view will contain one row for each undo in the current database, showing statistics about accesses to that specific undo.

Table 8. pg_stat_all_undos View

The pg_stat_all_indexes view will contain one row for each index in the current database, showing statistics about accesses to that specific index. The pg_stat_user_indexes and pg_stat_sys_indexes views contain the same information, but filtered to only show user and system indexes respectively.

Table 9. pg_stat_all_indexes View

Indexes can be used by simple index scans, “bitmap” index scans, and the optimizer. 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. The optimizer also accesses indexes to check for supplied constants whose values are outside the recorded range of the optimizer statistics because the optimizer statistics might be stale.

The idx_tup_read and idx_tup_fetch counts can be different even without any use of 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, or if any heap fetches are avoided by means of an index-only scan.

The pg_statio_all_tables view will contain one row for each table in the current database (including TOAST tables), showing statistics about I/O on that specific table. The pg_statio_user_tables and pg_statio_sys_tables views contain the same information, but filtered to only show user and system tables respectively.

Table 10. pg_statio_all_tables View

The pg_statio_all_undos view will contain one row for each undo in the current database, showing statistics about I/O on that specific undo.

Table 11. pg_statio_all_undos View

The pg_statio_all_indexes view will contain one row for each index in the current database, showing statistics about I/O on that specific index. The pg_statio_user_indexes and pg_statio_sys_indexes views contain the same information, but filtered to only show user and system indexes respectively.

Table 12. pg_statio_all_indexes View

The pg_statio_all_sequences view will contain one row for each sequence in the current database, showing statistics about I/O on that specific sequence.

Table 13. pg_statio_all_sequences View

Other ways of looking at the statistics can be set up by writing queries that use the same underlying statistics access functions used by the standard views shown above. For details such as the functions’ names, consult the definitions of the standard views. (For example, in psql you could issue \d+ pg_stat_activity.) The access functions for per-database statistics take a database OID as an argument to identify which database to report on. The per-table and per-index functions take a table or index OID. The functions for per-function statistics take a function OID. Note that only tables, indexes, and functions in the current database can be seen with these functions.

Additional functions related to statistics collection are listed in Table 14.

Table 14. Additional Statistics Functions

pg_stat_get_activity, the underlying function of the pg_stat_activity view, returns a set of records containing all the available information about each backend process. Sometimes it may be more convenient to obtain just a subset of this information. In such cases, an older set of per-backend statistics access functions can be used; these are shown in Table 15. These access functions use a backend ID number, which ranges from one to the number of currently active backends. The function pg_stat_get_backend_idset provides a convenient way to generate one row for each active backend for invoking these functions. For example, to show the PIDs and current queries of all backends:

SELECT pg_stat_get_backend_pid(s.backendid) AS pid,
       pg_stat_get_backend_activity(s.backendid) AS query
    FROM (SELECT pg_stat_get_backend_idset() AS backendid) AS s;

Table 15. Per-Backend Statistics Functions