Installation from package¶

Installation of pgCluu can be done through the PostgreSQL Global Development Group (PGDG) repositories. See how to install the PGDG repositories at the following URLs for Debian and Ubuntu:

        https://wiki.postgresql.org/wiki/Apt

and Redhat, CentOs, Fedora, Scientific Linux and Oracle Enterprise Linux:

        https://yum.postgresql.org/

Once it is done you can simply install pgCluu with commands:

        sudo apt install pgcluu
        sudo yum install pgcluu

Look at the package information to know where files are specifically installed. See next two chapters to see which default installation paths are used.

Installation from sources¶

Download the tarball from SourceForge and unpack the archive:

        tar xzf pgcluu-3.x.tar.gz
        cd pgcluu-3.x/
        perl Makefile.PL
        make && sudo make install

This will copy the Perl scripts pgcluu_collectd and pgcluu into /usr/local/bin directory and the man page to /usr/local/share/man/man1/pgcluu.1p.gz. Those are the default installation directories for 'site' install on some well-known distribution but the path could change.

If you want to install all under /usr location, use INSTALLDIRS='vendor' as an argument of Makefile.PL. The script will be installed into /usr/bin/pgcluu and the manpage into /usr/share/man/man1/pgcluu.1p.gz.

For example, to install everything just like Debian does, proceed as follows:

        perl Makefile.PL INSTALLDIRS=vendor

By default INSTALLDIRS is set to site.

The directory where statistics will be saved is /var/lib/pgcluu/data and the default directory for reports is /var/lib/pgcluu/report.

The CGI script is installed into /var/lib/cgi-bin/pgcluu.cgi and the Apache configuration file into /etc/apache/conf-available/pgcluu.conf. This file allow acces to resources files (CSS and JS files) from installation directory /usr/local/share/pgcluu/rsc/. Access is granted to local user only by default.

All scripts (pgcluu_collectd, pgcluu and pgcluu.cgi) are reading configuration file from /usr/local/etc/pgcluu.conf. This file is mainly use by the CGI script but some directives are dedicated to pgcluu_collectd and pgcluu script to define the retention days for example.

Custom installation¶

The installation of pgCluu can be fully customized through environment variables (RSCDIR,CGIDIR,CONFDIR,PIDDIR,STATDIR,APACHECONF,MANDIR, DOCDIR,SYSTEMDDIR,RETENTION). These variables are passed to Makefile.PL as command line arguments or can be exported as environment variables before running "perl Makefile.PL".

The default values for these variables are:

    DESTDIR     => /usr/local
    INSTALLDIRS => site
    CONFDIR     => DESTDIR/etc
    PIDDIR      => /var/run/postgres
    STATDIR     => /var/lib/pgcluu/data
    REPORTDIR   => /var/lib/pgcluu/data
    RSCDIR      => DESTDIR/share/pgcluu
    CGIDIR      => /usr/lib/cgi-bin
    APACHECONF  => /etc/apache2/conf-available
    MANDIR      => DESTDIR/share/man
    DOCDIR      => DESTDIR/share/doc
    SYSTEMDIR   => DESTDIR/lib/systemd/system
    RETENTION   => 0

If INSTALLDIRS is set to 'vendor':

    CONFDIR     => /etc
    DESTDIR     => /usr

The configuration file is auto-generated by the Makefile.PL script and saved into CONFDIR/pgcluu.conf. If the destination file exists it is not overridden. The file is also saved as example in the directory DOCDIR/pgcluu/examples/pgcluu.conf.dist

The directory where pgcluu_colletd will store statistics is defined with STATDIR which is by default /var/lib/pgcluu/data. The directory where pgcluu will generate static html reports is defined by REPORTDIR. The default is /var/lib/pgcluu/report. Both directory must be owned by the postgres user.

The man page is saved as DESTDIR/share/man/pgcluu.1p.gz and a symbolic link pgcluu_collectd.1p.gz is created to this file. The documentation, README, changelog.gz, LICENSE files are saved under DESTDIR/share/doc/.

For the CGI mode, the resources (css and js files from the cgi-bin/rsc) are saved under the DESTDIR/share/pgcluu/rsc directory. The CGI script is saved under /usr/lib/cgi-bin/pgcluu.cgi. The Apache configuration file under /etc/apache2/conf-available/pgcluu.conf with a symbolic link /etc/apache2/conf-enabled/pgcluu.conf created to this file. Its content:

    Alias /pgcluu RSCDIR/
    <Directory RSCDIR/>
        Options FollowSymLinks MultiViews
        AllowOverride None
        Require local
        #Require ip 192.1.168.0/24
    </Directory>

The systemd files (pgcluu_collectd.service,pgcluu.service,pgcluu.timer) are saved as examples into DOCDIR/pgcluu/examples/ and into the systemd directory SYSTEMDDIR/

The right path to the configuration file is set into all scripts pgcluu, pgcluu_collectd and pgcluu.cgi. The path where the pid file must be saved is replaced into pgcluu_collectd with the value of PIDFILE variable.

Paths to scripts, pid directory, statistics and reports are replaced in all systemd service files following the values of the environment variable explain here.

pgcluu_collectd usage¶

usage: pgcluu_collectd [options] output_dir

        output_dir: full path to directory where pgcluu_collectd will
                    store statistics.

options:

  -B, --enable-buffercache enable buffercache statistics if pg_buffercache
                           extension is installed.
  -c, --capture            create a snapshot of the PostgreSQL installation
                           into tmp/pgcluu_capture.tar.gz.
  -C, --end-counter=NUM    terminate program after NUM reports.
  -d, --dbname=DATABASE    database name to connect to. Default to current user.
  -D, --daemonize          detach from console and enter in daemon mode.
  -E, --end-after=SECOND   self terminate program after a given number of seconds.
                           Can be written: 7200 or 120M or 2H, for days use 7D for
                           example to stop collecting data after seven days. 
  -f, --pid-file=FILE      path to pid file. Default: /var/run/postgresql/pgcluu_collectd.pid.
  -h, --host=HOSTNAME      database server host or socket directory
  -i, --interval=NUM       time to wait between runs
  -k, --kill               stop current pgcluu_collectd running daemon.
  -m, --metric=METRIC      set a coma separated list of metrics to perform.
  -M, --max-size=SIZE      self terminate program when the size of the output dir
                           exceed a given size. Can be written: 2GB or 2000MB. 
  -p, --port=PORT          database port(s) to connect to. Defaults to 5432.
  -P, --psql=BIN           path to the psql command. Default: psql.
  -Q, --no-statement       do not collect queries statistics from pg_stat_statements.
  -r, --rotate-daily       force daily rotation of data files.
  -R, --rotate-hourly      force hourly rotation of data files.
  -s, --sar=BIN            path to sar sysstat command. Default: sar.
  -S, --disable-sar        disable collect of system statistics with sar.
  -T, --no-tablespace      disable lookup at tablespace when the connect user
                           is not superuser to avoid printing an error message.
  -U, --dbuser=USERNAME    database user to connect as. Default to current user.
  -v, --verbose            Print out debug informations.
  -V, --version            Show pgcluu_collectd version and exit.
  -W, --password=pass      database password.
  -z, --compress           force compression of rotated data files.
  --included-db=DATABASE   collect statistics only for those databases present
                           in a comma separated list of database names.
  --list-metric            list available metrics actions that can be performed.
  --sysinfo                get operating system information and exit (sysinfo.txt).
  --no-sysinfo             do not collect operating system information at all. 
  --no-database            do not collect database statistics at all.
  --pgbouncer-args=OPTIONS Option to used to connect to the pgbouncer system
                           database. Ex: -p 6432 -U postgres -h 192.168.1.100
                           You must at least give one parameter to enable
                           pgbouncer monitoring.
  --sar-file=FILE          path to sar output data file for sysstat stats
                           Default to output_dir/sar_stats.dat.
  --stat-type all|user     Set stats tables to read. Values: 'all' or 'user' to
                           look at pg_stat_(all|user) tables. Default: user.
  --pgversion X.Y          force the PostgreSQL version to the given value.
  --pgservice NAME         Name of service inside of the pg_service.conf file.
  --exclude-time RANGE     exclude a laps of time by giving the start and end
                           hours.
  --cron-user=USERNAME     collect crontab settings for the given username (in
                           this case pgcluu_collectd need to be run as root).
                           Default is to use USERNAME environment variable or
                           postgres when it is not defined.
  --package-list=CMD       command to list PostgreSQL packages. Default is to
                           autodetect package type and using command 'rpm -qa'
                           or 'dpkg -l'. If you have an other system you can
                           set a custom command. A filter on keyword 'postgres'
                           is appended to the command: ' | grep postgres'.
  --retention NDAYS        number of rolling days to keep in data directory in
                           incremental mode. Default is to store indefinitely.
  --help                   print usage

Use those options to execute sar on the remote host defined by the -h option, otherwise it will be executed locally:

  --enable-ssh             activate the use of ssh to run sysstat remotely.
  --ssh-program ssh        path to the ssh program to use. Default: ssh.
  --ssh-user username      connection login name. Default to running user.
  --ssh-identity file      path to the identity file to use.
  --ssh-timeout second     timeout to ssh connection failure. Default 10 seconds.
  --ssh-options  options   list of -o options to use for the ssh connection. Options
                           always used:
                                 -o ConnectTimeout=$ssh_timeout
                                 -o PreferredAuthentications=hostbased,publickey

For example, as postgres user to monitor locally a full PostgreSQL cluster:

        mkdir /tmp/stat_db1/
        pgcluu_collectd -D -i 60 /tmp/stat_db1/

to collect statistics from pgbouncer too, and limit database statistics to a single database:

        pgcluu_collectd -D -i 60 /tmp/stat_db1/ -h 10.10.1.1 -U postgres -d mydb 
                --pgbouncer-args='-p 5342'

to disable statistics collect between 22:30 and 06:30 the next day:

        pgcluu_collectd -D -i 60 /tmp/stat_db1/ --exclude-time "22:30-06:30"

to collect statistics from a remote server:

        pgcluu_collectd -D -i 60 /tmp/stat_db1/ -h 10.0.0.1 -U postgres --disable-sar

the same but with collecting system statistics using remote sar calls:

        pgcluu_collectd -D -i 60 /tmp/stat_db1/ -h 10.0.0.1 -U postgres --enable-ssh 
                --ssh-user postgres --ssh-identity /var/lib/postgresql/.ssh/id_rsa.pub

You may need a .pgpass and be able to establish passwordless ssh connections to be able to collect statistics from remote hosts.

Then after some time and activities on the database, stop the daemon as follow:

        pgcluu_collectd -k

or by sending sigterm to the pgcluu_collectd's pid.

You can run the collector in incremental mode using a daily or a hourly statistics rotation:

        pgcluu_collectd -D -i 60 /tmp/stat_db1/ --rotate-daily

On a server with huge activity you may want to use --rotate-hourly and compression mode with --compress option. If you have limited disk space you can restrict the retention time of statistics files using option --retention with the storage day limit.

Statistics files¶

The output directory with all statistics collected should look likes:

        /tmp/stat_db1/
        |-- commit_memory.csv
        |-- end-pg_statio_user_indexes.csv
        |-- end-pg_statio_user_sequences.csv
        |-- end-pg_statio_user_tables.csv
        |-- end-pg_stat_user_functions.csv
        |-- end-pg_stat_user_indexes.csv
        |-- end-pg_stat_user_tables.csv
        |-- end-pg_stat_xact_user_functions.csv
        |-- end-pg_stat_xact_user_tables.csv
        |-- fs_stat_use.csv
        |-- pg_class_size.csv
        |-- pg_database_size.csv
        |-- pg_db_role_setting.csv
        |-- pg_hba.conf
        |-- pg_ident.conf
        |-- pg_nondefault_settings.csv
        |-- pg_prepared_xact.csv
        |-- pg_settings.csv
        |-- pg_stat_archiver.csv
        |-- pg_stat_bgwriter.csv
        |-- pg_stat_connections.csv
        |-- pg_stat_count_indexes.csv
        |-- pg_stat_database_conflicts.csv
        |-- pg_stat_database.csv
        |-- pg_stat_hash_indexes.csv
        |-- pg_stat_invalid_indexes.csv
        |-- pg_statio_user_indexes.csv
        |-- pg_statio_user_sequences.csv
        |-- pg_statio_user_tables.csv
        |-- pg_stat_locks.csv
        |-- pg_stat_missing_fkindexes.csv
        |-- pg_stat_redundant_indexes.csv
        |-- pg_stat_replication.csv
        |-- pg_stat_unlogged.csv
        |-- pg_stat_unused_indexes.csv
        |-- pg_stat_user_functions.csv
        |-- pg_stat_user_indexes.csv
        |-- pg_stat_user_tables.csv
        |-- pg_stat_xact_user_functions.csv
        |-- pg_stat_xact_user_tables.csv
        |-- pg_tablespace_size.csv
        |-- pg_xlog_stat.csv
        |-- postgresql.auto.conf
        |-- postgresql.conf
        |-- sar_stats.dat
        |-- sysinfo.txt

Then now you can proceed with pgcluu to generate reports.

Rotation and compression¶

When used the --rotate-daily or --rotate-hourly commands line option will force pgcluu_collectd to rotate daily or hourly all statistic's files. In this case, statistics files will be created in a subdirectory based on rotation frequency, output_dir/year/month/day[/hour]. This is called the incremental mode.

To save filesystem space it is possible to enable compression of all rotated files during the rotation process. Just activate the -z or --compress command line option. You can also use the --retention option to set the storage time limit in days.

Incremental mode¶

This mode is enabled when --rotate-daily or --rotate-hourly command line options are used. It allow pgcluu to build reports incrementally by days or hours. In this mode you don't have to build reports per day or hour pgcluu will do the work automatically, just give it the top statistics directory.

        pgcluu -o /var/www/pgcluu/reports/ /var/lib/pgcluu/data/

pgcluu will detect that --rotate-daily or --rotate-hourly have been used to collect data and generate reports for each day or hours stored. Previous directories already processed will not be processed again unless this was the the last one.

The daily or hourly basis statistic storage also allow the use of the CGI script pgcluu.cgi to have dynamic reports and temporal search. See "Using dynamic mode" chapter.

When used the --rotate-daily or --rotate-hourly commands line option will force pgcluu_collectd to rotate daily or hourly all statistic's files. In this case, statistics files will be created in a subdirectory based on rotation frequency, output_dir/year/month/day[/hour]. This is called the incremental mode.

To save filesystem space it is possible to enable compression of all rotated files during the rotation process. Just activate the -z or --compress command line option. You can also use the --retention option to set the storage time limit in days.

Capture mode¶

The goal of this mode is to be able to obtain a simple report about the PostgreSQL installation without collected metrics others than database and tablespace size. This report can be use by ITs to better understand the configuration and things that need to be tuned.

To enable this mode, just run pgcluu_collectd with the single option -c or --capture. Other command line options will not be taken in account. pgcluu_collectd will create e temporary directory /tmp/pgcluu_capture to store temporary data and will removed if after building a compressed tar archive: /tmp/pgcluu_capture.tar.gz. This is this archive that can be used with pgcluu to build a snapshot report of the instance. pgcluu will automatically detect this mode.

Static HTML reports¶

To generate a pgCluu report about a PostgreSQL Cluster you must, at least, have a directory that contains all data files generated by pgcluu_collectd or pgstats. In this directory, if you have a file named sar_stats.dat or sadc_stats.dat for binary sadc data file, it will be taken to build report about system utilization. If you just want to make a report from a sar file use the -i or -I options.

usage: pgcluu [options] [-i sar_file | -I sadc_file] [input_dir]

        input_dir: directory where pgcluu_collectd or pgstats and sar data
                   files are stored.

options:
-b, --begin datetime start date/time for the data to be parsed.
-C, --cache generate cache files only (.bin), no html output.
-d, --db-only dbname only report for the whole cluster and the given
database name. You can use it multiple time or
give a comma separated list of database name.
-D, --device-only dev only report I/O stats for a particular device
You can use it multiple time or give a comma
separated list of device name, ex: sda,sdc.
-e, --end datetime end date/time for the data to be parsed.
-i, --sar-file=FILE path to the sar text data file to read to generate
system reports. Default to input_dir/sar_stats.dat.
-I, --sadc-file=FILE sadc binary data file to read to generate system
reports. Default to input_dir/sadc_stats.dat.
-n, --top-number Top number of tables or indexes I/O stats to show.
Default is set to top 10. Set it to 0 to show all.
-N, --network-only iface only report stats for a particular network interface.
You can use it multiple times or give a comma separated
list of network interfaces, ex: eth0,eth1.
-o, --output=DIR output directory
-r, --reverse-date By default pgcluu look at mm/dd/yy format in sar file.
When enabled pgcluu will look at dd/mm/yy format.
-s, --sadf=BIN path to the sadf sysstat command used to read the
sadc binary data file. Default: /usr/bin/sadf.
-S, --disable-sar disable collect of system statistics with sar.
-t, --with-table table Only report for the whole tables and the given
table name. You can use it multiple time or
give a comma separated list of database name.
-T, --no-table Do not report statistics related to tables.
-v, --verbose Print out debug informations.
-V, --version Show pgcluu version and exit.
-x, --external-menu Save menu in menu.html and load it into each report
using w3-include-html attribut from w3.js. This will only work if acces to HTML reports is through a Web server, not using the file:// protocol.
-z, --timezone +/-XX Set the number of hour(s) from GMT of the timezone.
Use this to adjust date/time from the sar output,
pgcluu use GMT time to draw charts.
-Z, --stats-timezone +/-XX Set the number of hour(s) from GMT of the timezone.
Use this to adjust date/time from the cluster and
system stats output, pgcluu use GMT time.
--from-sa-file instruct pgcluu that file specified by the -i option
uses the standard system activity daily data file.
--charset used to set the HTML charset to be used.
Default: utf-8.
--retention NDAYS number of rolling days to keep in report directory.
Default is to store indefinitely.
--help print usage

For example, you can generate all HTML reports from data files stored into /tmp/stat_db1/ with the following commands:

        mkdir /tmp/report_db1/
        pgcluu -o /tmp/report_db1/ /tmp/stat_db1/

If you just want reports of some databases, use the following:

        pgcluu -o /tmp/report_db1/ /tmp/stat_db1/ --db-only "db1,db2,db3"

If you just want to create a report from a sar output file:

        sar -p -A 10 60 > /root/my_sar_file.txt
        pgcluu -o /tmp/report_sar/ -i /root/my_sar_file.txt

or from a daily sa file:

        sar -p -A -f /var/log/sa/sa18 > /root/my_sar_file.txt
        pgcluu -o /tmp/report_sar/ -i /root/my_sar_file.txt --from-sa-file

and from a sa binary file:

        pgcluu -o /tmp/report_sar/ -i /var/log/sysstat/sa22

or the sa text file if you don't have the same version of sysstat:

        pgcluu -o /tmp/report_sar/ -i /var/log/sysstat/sar23 --from-sa-file

If pgcluu_collectd have been run in incremental mode you can limit the number of retention days used for the reports:

        pgcluu -o /tmp/report_sar/ /tmp/stat_db1/ --retention 30

A static report will be built for each day or hour following the rotation used with pgcluu_collectd.

Dynamic reports (CGI)¶

Dynamics reports are build by a CGI script named pgcluu.cgi that can be found in the cgi-bin repository of the source code. It allow you to select the time period to build reports and to look at all differents reports for this period just as with static HTML reports. After installation from sources or binary packages the CGI might be found in /usr/lib/cgi-bin/pgcluu.cgi or /var/www/cgi-bin/pgcluu.cgi following your distribution.

To use pgCluu in CGI mode, you need a Web server (here we use Apache) and a cron task to build the cache periodically.

First enable CGI mode.

        sudo a2enmod cgi
        sudo service apache2 restart

Then the CGI need to find the statistics directory where pgcluu_collectd mostly /var/lib/pgcluu/data/. The content of this repository must be readable by the Apache user (www-data).

Take care to restrict access to your server and the CGI as information about your database and server are exposed in the reports.

The CGI script, pgcluu.cgi, use resources files (CSS and javascript). They are stored in the cgi-bin/rsc/ directory of the source distribution. Install this repository onto the DocumentRoot of your Web server, for example:

        sudo mkdir /var/www/pgcluu/
        sudo cp -rf cgi-bin/rsc /var/www/pgcluu/

Then edit /etc/pgcluu.conf, copy it from sources cgi-bin/pgcluu.conf if it doesn't exists. Change the RSC_BASE and INPUT_DIR configuration directive to match your installation. Here in our example:

        RSC_BASE        /pgcluu/rsc/
        INPUT_DIR       /var/lib/pgcluu/data

Now we can start the data collection. For the moment pgcluu.conf is only read by the CGI pgcluu.cgi but this might change in the future.

To let pgCluu start collecting data, you can use the following command (you can change it to however you like). We will need postgres user for this.

        sudo su - postgres
        /usr/local/bin/pgcluu_collectd -D -i 60 --rotate-daily /var/lib/pgcluu/data

This will rotate you data daily. This is also the default in the systemd service file.

Plugging the CGI directly to the CSV statistics files will result in very slow generation reports. To improve speed caching must be used, you must execute periodically pgcluu in cache mode. Run it manually the first time

        /usr/local/bin/pgcluu --cache /var/lib/pgcluu/data

then add a cron task to execute the command each five or ten minutes:

        */5 * * * * /usr/local/bin/pgcluu --cache /var/lib/pgcluu/data

If you are using systemd pgcluu.timer service then caching can be enabled using STATS_REPORT_CACHING directive. Enabling this mode generate cache files (*.bin) in the statistics directory and disable static HTML reports generation.

To see your reports, go to the following URL:

        http://localhost/cgi-bin/pgcluu.cgi

Or change localhost by your fqdn server name.

Note that here pgcluu_collectd and pgcluu scripts are found in /usr/local/bin/ which is the default for an installation from sources but with an installation from binary package you might find them into /usr/bin/.

The CGI mode allow you to select the period of time used to generate the statistics reports. Using default pgcluu_collectd 60 seconds interval, a daily graph will have 1440 points. Having too much timeseries to render is not a good point for performances so we are limiting to the first seven days of the selected period. This mean a maximum of 10080 points for one week, this seems a safe limit. Feel free to increase or reduce the limit following the server and client resources. This is controlled by

        MAX_RENDERED_DAYS       7

configuration directive in pgcluu.conf