firebird-mirror/doc/sql.extensions/README.profiler.md

Profiler (FB 5.0)

The profiler allows users to measure performance cost of SQL and PSQL code. It's implemented with a system package in the engine passing data to a profiler plugin.

This documentation treats the engine and plugin parts as a single thing, in the way the default profiler (Default_Profiler) is going to be used.

The RDB$PROFILER package can profile execution of PSQL code, collecting statistics of how many times each line was executed along with its minimum, maximum and accumulated execution times (with nanoseconds precision), as well as open and fetch statistics of implicit and explicit SQL cursors.

To collect profile data, a user must first start a profile session with RDB$PROFILER.START_SESSION. This function returns a profile session ID which is later stored in the profiler snapshot tables to be queried and analyzed by the user. A profiler session may be local (same attachment) or remote (another attachment).

Remote profiling just forwards commands to the remote attachment. So, it's possible that a client profiles multiple attachments simultaneously. It's also possible that a locally or remotely started profile session have commands issued by another attachment.

Remotely issued commands require that the target attachment is in an idle state, i.e. not executing others requests. When the target attachment is not idle, the call blocks waiting for that state.

If the remote attachment is from a different user, the calling user must have the system privilege PROFILE_ANY_ATTACHMENT.

After a session is started, PSQL and SQL statements statistics are collected in memory. A profile session collects data only of statements executed in the same attachment associated with the session. Data is aggregated and stored per requests (i.e. a statement execution). When querying snapshot tables, the user may do extra aggregation per statement, or use the auxiliary views that do that automatically.

A session may be paused to temporarily disable statistics collecting. It may be resumed later to return statistics collection in the same session.

A new session may be started when a session is already active. In that case, it has the same semantics of finishing the current session with RDB$PROFILER.FINISH_SESSION(FALSE), so snapshots tables are not updated.

To analyze the collected data, the user must flush the data to the snapshot tables, which can be done by finishing or pausing a session (with FLUSH parameter set to TRUE), or calling RDB$PROFILER.FLUSH. Data is flushed using an autonomous transaction (a transaction started and finished for the specific purpose of profiler data update).

Important

When the profiler is active, there is an overhead that makes everything slower. This overhead varies depending on OS, kernel version and CPU hardware and it's difficult to predict.

But sometimes this overhead may be very high, say, greater than 100%.

If this happens in Linux, you may see what clock source it's using with this command:

cat /sys/devices/system/clocksource/clocksource0/current_clocksource

If result is different than tsc, that may be the cause of this problem.

You can see here how to change clocksource, but you must understand it may have others consequences.

Another possible source of slowdown in Linux is this bug.

Example usage

Below is a sample profile session and queries for data analysis.

-- Preparation - create table and routines that will be analyzed

create table tab (
    id integer not null,
    val integer not null
);

set term !;

create or alter function mult(p1 integer, p2 integer) returns integer
as
begin
    return p1 * p2;
end!

create or alter procedure ins
as
    declare n integer = 1;
begin
    while (n <= 1000)
    do
    begin
        if (mod(n, 2) = 1) then
            insert into tab values (:n, mult(:n, 2));
        n = n + 1;
    end
end!

set term ;!

-- Start profiling

select rdb$profiler.start_session('Profile Session 1', null, null, null, 'DETAILED_REQUESTS') from rdb$database;

set term !;

execute block
as
begin
    execute procedure ins;
    delete from tab;
end!

set term ;!

execute procedure rdb$profiler.finish_session(true);

execute procedure ins;

select rdb$profiler.start_session('Profile Session 2') from rdb$database;

select mod(id, 5),
       sum(val)
  from tab
  where id <= 50
  group by mod(id, 5)
  order by sum(val);

execute procedure rdb$profiler.finish_session(true);

-- Data analysis

set transaction read committed;

select * from plg$prof_sessions;

select * from plg$prof_psql_stats_view;

select * from plg$prof_record_source_stats_view;

select preq.*
  from plg$prof_requests preq
  join plg$prof_sessions pses
    on pses.profile_id = preq.profile_id and
       pses.description = 'Profile Session 1';

select pstat.*
  from plg$prof_psql_stats pstat
  join plg$prof_sessions pses
    on pses.profile_id = pstat.profile_id and
       pses.description = 'Profile Session 1'
  order by pstat.profile_id,
           pstat.request_id,
           pstat.line_num,
           pstat.column_num;

select pstat.*
  from plg$prof_record_source_stats pstat
  join plg$prof_sessions pses
    on pses.profile_id = pstat.profile_id and
       pses.description = 'Profile Session 2'
  order by pstat.profile_id,
           pstat.request_id,
           pstat.cursor_id,
           pstat.record_source_id;

Function START_SESSION

RDB$PROFILER.START_SESSION starts a new profiler session, makes it the current session (of the given ATTACHMENT_ID) and returns its identifier.

If FLUSH_INTERVAL is different than NULL, auto-flush is setup in the same way as manually calling RDB$PROFILER.SET_FLUSH_INTERVAL.

If PLUGIN_NAME is NULL (the default), it uses the database configuration DefaultProfilerPlugin.

PLUGIN_OPTIONS are plugin specific options and currently could be NULL or the string DETAILED_REQUESTS for Default_Profiler plugin.

When DETAILED_REQUESTS is used, PLG$PROF_REQUESTS will store detailed requests data, i.e., one record per each invocation of a statement. This may generate a lot of records, causing RDB$PROFILER.FLUSH to be slow.

When DETAILED_REQUESTS is not used (the default), PLG$PROF_REQUESTS stores an aggregated record per statement, using REQUEST_ID = 0.

Input parameters:

• DESCRIPTION type VARCHAR(255) CHARACTER SET UTF8 default NULL
• FLUSH_INTERVAL type INTEGER default NULL
• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)
• PLUGIN_NAME type VARCHAR(255) CHARACTER SET UTF8 default NULL
• PLUGIN_OPTIONS type VARCHAR(255) CHARACTER SET UTF8 default NULL

Return type: BIGINT NOT NULL.

Procedure PAUSE_SESSION

RDB$PROFILER.PAUSE_SESSION pauses the current profiler session (of the given ATTACHMENT_ID), so the next executed statements statistics are not collected.

If FLUSH is TRUE, the snapshot tables are updated with data up to the current moment, otherwise data remains only in memory for later update.

Calling RDB$PROFILER.PAUSE_SESSION(TRUE) has the same semantics of calling RDB$PROFILER.PAUSE_SESSION(FALSE) followed by RDB$PROFILER.FLUSH (using the same ATTACHMENT_ID).

Input parameters:

• FLUSH type BOOLEAN NOT NULL default FALSE
• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure RESUME_SESSION

RDB$PROFILER.RESUME_SESSION resumes the current profiler session (of the given ATTACHMENT_ID), if it was paused, so the next executed statements statistics are collected again.

Input parameters:

• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure FINISH_SESSION

RDB$PROFILER.FINISH_SESSION finishes the current profiler session (of the given ATTACHMENT_ID).

If FLUSH is TRUE, the snapshot tables are updated with data of the finished session (and old finished sessions not yet present in the snapshot), otherwise data remains only in memory for later update.

Calling RDB$PROFILER.FINISH_SESSION(TRUE) has the same semantics of calling RDB$PROFILER.FINISH_SESSION(FALSE) followed by RDB$PROFILER.FLUSH (using the same ATTACHMENT_ID).

Input parameters:

• FLUSH type BOOLEAN NOT NULL default TRUE
• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure CANCEL_SESSION

RDB$PROFILER.CANCEL_SESSION cancels the current profiler session (of the given ATTACHMENT_ID).

All session data present in the profiler plugin is discarded and will not be flushed.

Data already flushed is not deleted automatically.

Input parameters:

• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure DISCARD

RDB$PROFILER.DISCARD removes all sessions (of the given ATTACHMENT_ID) from memory, without flushing them.

If there is a active session, it is cancelled.

Input parameters:

• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure FLUSH

RDB$PROFILER.FLUSH updates the snapshot tables with data from the profile sessions (of the given ATTACHMENT_ID) in memory.

After flushing, the data is stored in tables PLG$PROF_SESSIONS, PLG$PROF_STATEMENTS, PLG$PROF_RECORD_SOURCES, PLG$PROF_REQUESTS, PLG$PROF_PSQL_STATS and PLG$PROF_RECORD_SOURCE_STATS and may be read and analyzed by the user.

Data is updated using an autonomous transaction, so if the procedure is called in a snapshot transaction, data will not be directly readable in the same transaction.

Once flush happens, finished sessions are removed from memory.

Input parameters:

• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Procedure SET_FLUSH_INTERVAL

RDB$PROFILER.SET_FLUSH_INTERVAL turns periodic auto-flush on (when FLUSH_INTERVAL is greater than 0) or off (when FLUSH_INTERVAL is equal to 0).

FLUSH_INTERVAL is interpreted as number of seconds.

Input parameters:

• FLUSH_INTERVAL type INTEGER NOT NULL
• ATTACHMENT_ID type BIGINT default NULL (meaning CURRENT_CONNECTION)

Snapshot tables

Snapshot tables (as well views and sequence) are automatically created in the first usage of the profiler. They are owned by the database owner, with read/write permissions for PUBLIC.

When a session is deleted, the related data in other profiler snapshot tables are automatically deleted too through foreign keys with DELETE CASCADE option.

Below is the list of tables that stores profile data.

Table PLG$PROF_SESSIONS

• PROFILE_ID type BIGINT - Profile session ID
• ATTACHMENT_ID type BIGINT - Attachment ID
• USER_NAME type CHAR(63) CHARACTER SET UTF8 - User name
• DESCRIPTION type VARCHAR(255) CHARACTER SET UTF8 - Description passed in RDB$PROFILER.START_SESSION
• START_TIMESTAMP type TIMESTAMP WITH TIME ZONE - Moment the profile session was started
• FINISH_TIMESTAMP type TIMESTAMP WITH TIME ZONE - Moment the profile session was finished (NULL when not finished)
• Primary key: PROFILE_ID

Table PLG$PROF_STATEMENTS

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• PARENT_STATEMENT_ID type BIGINT - Parent statement ID - related to sub routines
• STATEMENT_TYPE type VARCHAR(20) CHARACTER SET UTF8 - BLOCK, FUNCTION, PROCEDURE or TRIGGER
• PACKAGE_NAME type CHAR(63) CHARACTER SET UTF8 - Package of FUNCTION or PROCEDURE
• ROUTINE_NAME type CHAR(63) CHARACTER SET UTF8 - Routine name of FUNCTION, PROCEDURE or TRIGGER
• SQL_TEXT type BLOB subtype TEXT CHARACTER SET UTF8 - SQL text for BLOCK
• Primary key: PROFILE_ID, STATEMENT_ID

Table PLG$PROF_CURSORS

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• CURSOR_ID type INTEGER - Cursor ID
• NAME type CHAR(63) CHARACTER SET UTF8 - Name of explicit cursor
• LINE_NUM type INTEGER - Line number of the cursor
• COLUMN_NUM type INTEGER - Column number of the cursor
• Primary key: PROFILE_ID, STATEMENT_ID, CURSOR_ID

Table PLG$PROF_RECORD_SOURCES

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• CURSOR_ID type INTEGER - Cursor ID
• RECORD_SOURCE_ID type INTEGER - Record source ID
• PARENT_RECORD_SOURCE_ID type INTEGER - Parent record source ID
• LEVEL type INTEGER - Indentation level for the record source
• ACCESS_PATH type VARCHAR(255) CHARACTER SET UTF8 - Access path for the record source
• Primary key: PROFILE_ID, STATEMENT_ID, CURSOR_ID, RECORD_SOURCE_ID

Table PLG$PROF_REQUESTS

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• REQUEST_ID type BIGINT - Request ID
• CALLER_STATEMENT_ID type BIGINT - Caller statement ID
• CALLER_REQUEST_ID type BIGINT - Caller request ID
• START_TIMESTAMP type TIMESTAMP WITH TIME ZONE - Moment this request was first gathered profile data
• FINISH_TIMESTAMP type TIMESTAMP WITH TIME ZONE - Moment this request was finished
• TOTAL_ELAPSED_TIME type BIGINT - Accumulated elapsed time (in nanoseconds) of the request
• Primary key: PROFILE_ID, STATEMENT_ID, REQUEST_ID

Table PLG$PROF_PSQL_STATS

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• REQUEST_ID type BIGINT - Request ID
• LINE_NUM type INTEGER - Line number of the statement
• COLUMN_NUM type INTEGER - Column number of the statement
• COUNTER type BIGINT - Number of executed times of the line/column
• MIN_ELAPSED_TIME type BIGINT - Minimal elapsed time (in nanoseconds) of a line/column execution
• MAX_ELAPSED_TIME type BIGINT - Maximum elapsed time (in nanoseconds) of a line/column execution
• TOTAL_ELAPSED_TIME type BIGINT - Accumulated elapsed time (in nanoseconds) of the line/column executions
• Primary key: PROFILE_ID, STATEMENT_ID, REQUEST_ID, LINE_NUM, COLUMN_NUM

Table PLG$PROF_RECORD_SOURCE_STATS

• PROFILE_ID type BIGINT - Profile session ID
• STATEMENT_ID type BIGINT - Statement ID
• REQUEST_ID type BIGINT - Request ID
• CURSOR_ID type INTEGER - Cursor ID
• RECORD_SOURCE_ID type INTEGER - Record source ID
• OPEN_COUNTER type BIGINT - Number of open times of the record source
• OPEN_MIN_ELAPSED_TIME type BIGINT - Minimal elapsed time (in nanoseconds) of a record source open
• OPEN_MAX_ELAPSED_TIME type BIGINT - Maximum elapsed time (in nanoseconds) of a record source open
• OPEN_TOTAL_ELAPSED_TIME type BIGINT - Accumulated elapsed time (in nanoseconds) of the record source openings
• FETCH_COUNTER type BIGINT - Number of fetch times of the record source
• FETCH_MIN_ELAPSED_TIME type BIGINT - Minimal elapsed time (in nanoseconds) of a record source fetch
• FETCH_MAX_ELAPSED_TIME type BIGINT - Maximum elapsed time (in nanoseconds) of a record source fetch
• FETCH_TOTAL_ELAPSED_TIME type BIGINT - Accumulated elapsed time (in nanoseconds) of the record source fetches
• Primary key: PROFILE_ID, STATEMENT_ID, REQUEST_ID, CURSOR_ID, RECORD_SOURCE_ID

Auxiliary views

These views help profile data extraction aggregated at statement level.

They should be the preferred way to analyze the collected data. They can also be used together with the tables to get additional data not present on the views.

After hotspots are found, one can drill down in the data at the request level through the tables.

View PLG$PROF_STATEMENT_STATS_VIEW

select req.profile_id,
       req.statement_id,
       sta.statement_type,
       sta.package_name,
       sta.routine_name,
       sta.parent_statement_id,
       sta_parent.statement_type parent_statement_type,
       sta_parent.routine_name parent_routine_name,
       (select sql_text
          from plg$prof_statements
          where profile_id = req.profile_id and
                statement_id = coalesce(sta.parent_statement_id, req.statement_id)
       ) sql_text,
       count(*) counter,
       min(req.total_elapsed_time) min_elapsed_time,
       max(req.total_elapsed_time) max_elapsed_time,
       cast(sum(req.total_elapsed_time) as bigint) total_elapsed_time,
       cast(sum(req.total_elapsed_time) / count(*) as bigint) avg_elapsed_time
  from plg$prof_requests req
  join plg$prof_statements sta
    on sta.profile_id = req.profile_id and
       sta.statement_id = req.statement_id
  left join plg$prof_statements sta_parent
    on sta_parent.profile_id = sta.profile_id and
       sta_parent.statement_id = sta.parent_statement_id
  group by req.profile_id,
           req.statement_id,
           sta.statement_type,
           sta.package_name,
           sta.routine_name,
           sta.parent_statement_id,
           sta_parent.statement_type,
           sta_parent.routine_name
  order by sum(req.total_elapsed_time) desc

View PLG$PROF_PSQL_STATS_VIEW

select pstat.profile_id,
       pstat.statement_id,
       sta.statement_type,
       sta.package_name,
       sta.routine_name,
       sta.parent_statement_id,
       sta_parent.statement_type parent_statement_type,
       sta_parent.routine_name parent_routine_name,
       (select sql_text
          from plg$prof_statements
          where profile_id = pstat.profile_id and
                statement_id = coalesce(sta.parent_statement_id, pstat.statement_id)
       ) sql_text,
       pstat.line_num,
       pstat.column_num,
       cast(sum(pstat.counter) as bigint) counter,
       min(pstat.min_elapsed_time) min_elapsed_time,
       max(pstat.max_elapsed_time) max_elapsed_time,
       cast(sum(pstat.total_elapsed_time) as bigint) total_elapsed_time,
       cast(sum(pstat.total_elapsed_time) / nullif(sum(pstat.counter), 0) as bigint) avg_elapsed_time
  from plg$prof_psql_stats pstat
  join plg$prof_statements sta
    on sta.profile_id = pstat.profile_id and
       sta.statement_id = pstat.statement_id
  left join plg$prof_statements sta_parent
    on sta_parent.profile_id = sta.profile_id and
       sta_parent.statement_id = sta.parent_statement_id
  group by pstat.profile_id,
           pstat.statement_id,
           sta.statement_type,
           sta.package_name,
           sta.routine_name,
           sta.parent_statement_id,
           sta_parent.statement_type,
           sta_parent.routine_name,
           pstat.line_num,
           pstat.column_num
  order by sum(pstat.total_elapsed_time) desc

View PLG$PROF_RECORD_SOURCE_STATS_VIEW

select rstat.profile_id,
       rstat.statement_id,
       sta.statement_type,
       sta.package_name,
       sta.routine_name,
       sta.parent_statement_id,
       sta_parent.statement_type parent_statement_type,
       sta_parent.routine_name parent_routine_name,
       (select sql_text
          from plg$prof_statements
          where profile_id = rstat.profile_id and
                statement_id = coalesce(sta.parent_statement_id, rstat.statement_id)
       ) sql_text,
       rstat.cursor_id,
       cur.name cursor_name,
       cur.line_num cursor_line_num,
       cur.column_num cursor_column_num,
       rstat.record_source_id,
       recsrc.parent_record_source_id,
       recsrc.level,
       recsrc.access_path,
       cast(sum(rstat.open_counter) as bigint) open_counter,
       min(rstat.open_min_elapsed_time) open_min_elapsed_time,
       max(rstat.open_max_elapsed_time) open_max_elapsed_time,
       cast(sum(rstat.open_total_elapsed_time) as bigint) open_total_elapsed_time,
       cast(sum(rstat.open_total_elapsed_time) / nullif(sum(rstat.open_counter), 0) as bigint) open_avg_elapsed_time,
       cast(sum(rstat.fetch_counter) as bigint) fetch_counter,
       min(rstat.fetch_min_elapsed_time) fetch_min_elapsed_time,
       max(rstat.fetch_max_elapsed_time) fetch_max_elapsed_time,
       cast(sum(rstat.fetch_total_elapsed_time) as bigint) fetch_total_elapsed_time,
       cast(sum(rstat.fetch_total_elapsed_time) / nullif(sum(rstat.fetch_counter), 0) as bigint) fetch_avg_elapsed_time,
       cast(coalesce(sum(rstat.open_total_elapsed_time), 0) + coalesce(sum(rstat.fetch_total_elapsed_time), 0) as bigint) open_fetch_total_elapsed_time
  from plg$prof_record_source_stats rstat
  join plg$prof_cursors cur
    on cur.profile_id = rstat.profile_id and
       cur.statement_id = rstat.statement_id and
       cur.cursor_id = rstat.cursor_id
  join plg$prof_record_sources recsrc
    on recsrc.profile_id = rstat.profile_id and
       recsrc.statement_id = rstat.statement_id and
       recsrc.cursor_id = rstat.cursor_id and
       recsrc.record_source_id = rstat.record_source_id
  join plg$prof_statements sta
    on sta.profile_id = rstat.profile_id and
       sta.statement_id = rstat.statement_id
  left join plg$prof_statements sta_parent
    on sta_parent.profile_id = sta.profile_id and
       sta_parent.statement_id = sta.parent_statement_id
  group by rstat.profile_id,
           rstat.statement_id,
           sta.statement_type,
           sta.package_name,
           sta.routine_name,
           sta.parent_statement_id,
           sta_parent.statement_type,
           sta_parent.routine_name,
           rstat.cursor_id,
           cur.name,
           cur.line_num,
           cur.column_num,
           rstat.record_source_id,
           recsrc.parent_record_source_id,
           recsrc.level,
           recsrc.access_path
  order by coalesce(sum(rstat.open_total_elapsed_time), 0) + coalesce(sum(rstat.fetch_total_elapsed_time), 0) desc