Output Modes¶

The output can be changed by use of the "--output" option. The default output is nagios, although this can be changed at the top of the script if you wish. The current option choices are nagios, mrtg, and simple. To avoid having to enter the output argument each time, the type of output is automatically set if no --output argument is given, and if the current directory has one of the output options in its name. For example, creating a directory named mrtg and populating it with symlinks via the --symlinks argument would ensure that any actions run from that directory will always default to an output of "mrtg" As a shortcut for --output=simple, you can enter --simple, which also overrides the directory naming trick.

Nagios output

The default output format is for Nagios, which is a single line of information, along with four specific exit codes:

0 (OK)

1 (WARNING)

2 (CRITICAL)

3 (UNKNOWN)

The output line is one of the words above, a colon, and then a short description of what was measured. Additional statistics information, as well as the total time the command took, can be output as well: see the documentation on the arguments --showperf, --perflimit, and --showtime.

MRTG output

The MRTG output is four lines, with the first line always giving a single number of importance. When possible, this number represents an actual value such as a number of bytes, but it may also be a 1 or a 0 for actions that only return "true" or "false", such as check_postgres_version. The second line is an additional stat and is only used for some actions. The third line indicates an "uptime" and is not used. The fourth line is a description and usually indicates the name of the database the stat from the first line was pulled from, but may be different depending on the action.

Some actions accept an optional --mrtg argument to further control the output.

See the documentation on each action for details on the exact MRTG output for each one.

Simple output

The simple output is simply a truncated version of the MRTG one, and simply returns the first number and nothing else. This is very useful when you just want to check the state of something, regardless of any threshold. You can transform the numeric output by appending KB, MB, GB, TB, or EB to the output argument, for example:

  --output=simple,MB

Cacti output

The Cacti output consists of one or more items on the same line, with a simple name, a colon, and then a number. At the moment, the only action with explicit Cacti output is 'dbstats', and using the --output option is not needed in this case, as Cacti is the only output for this action. For many other actions, using --simple is enough to make Cacti happy.

archive_ready¶

("symlink: check_postgres_archive_ready") Checks how many WAL files with extension .ready exist in the pg_xlog/archive_status directory (PostgreSQL 10 and later: pg_wal/archive_status), which is found off of your data_directory. If the --lsfunc option is not used then this action must be run as a superuser, in order to access the contents of the pg_xlog/archive_status directory. The minimum version to use this action is Postgres 8.1. The --warning and --critical options are simply the number of .ready files in the pg_xlog/archive_status directory. Usually, these values should be low, turning on the archive mechanism, we usually want it to archive WAL files as fast as possible.

If the archive command fail, number of WAL in your pg_xlog directory will grow until exhausting all the disk space and force PostgreSQL to stop immediately.

To avoid running as a database superuser, a wrapper function around pg_ls_dir() should be defined as a superuser with SECURITY DEFINER, and the --lsfunc option used. This example function, if defined by a superuser, will allow the script to connect as a normal user nagios with --lsfunc=ls_archive_status_dir

  BEGIN;
  CREATE FUNCTION ls_archive_status_dir()
      RETURNS SETOF TEXT
      AS $$ SELECT pg_ls_dir('pg_xlog/archive_status') $$
      LANGUAGE SQL
      SECURITY DEFINER;
  REVOKE ALL ON FUNCTION ls_archive_status_dir() FROM PUBLIC;
  GRANT EXECUTE ON FUNCTION ls_archive_status_dir() to nagios;
  COMMIT;

Example 1: Check that the number of ready WAL files is 10 or less on host "pluto", using a wrapper function "ls_archive_status_dir" to avoid the need for superuser permissions

  check_postgres_archive_ready --host=pluto --critical=10 --lsfunc=ls_archive_status_dir

For MRTG output, reports the number of ready WAL files on line 1.

autovac_freeze¶

("symlink: check_postgres_autovac_freeze") Checks how close each database is to the Postgres autovacuum_freeze_max_age setting. This action will only work for databases version 8.2 or higher. The --warning and --critical options should be expressed as percentages. The 'age' of the transactions in each database is compared to the autovacuum_freeze_max_age setting (200 million by default) to generate a rounded percentage. The default values are 90% for the warning and 95% for the critical. Databases can be filtered by use of the --include and --exclude options. See the "BASIC FILTERING" section for more details.

Example 1: Give a warning when any databases on port 5432 are above 97%

  check_postgres_autovac_freeze --port=5432 --warning="97%"

For MRTG output, the highest overall percentage is reported on the first line, and the highest age is reported on the second line. All databases which have the percentage from the first line are reported on the fourth line, separated by a pipe symbol.

backends¶

("symlink: check_postgres_backends") Checks the current number of connections for one or more databases, and optionally compares it to the maximum allowed, which is determined by the Postgres configuration variable max_connections. The --warning and --critical options can take one of three forms. First, a simple number can be given, which represents the number of connections at which the alert will be given. This choice does not use the max_connections setting. Second, the percentage of available connections can be given. Third, a negative number can be given which represents the number of connections left until max_connections is reached. The default values for --warning and --critical are '90%' and '95%'. You can also filter the databases by use of the --include and --exclude options. See the "BASIC FILTERING" section for more details.

To view only non-idle processes, you can use the --noidle argument. Note that the user you are running as (either connecting directly or switching via --role) must be a superuser for this to work properly.

Example 1: Give a warning when the number of connections on host quirm reaches 120, and a critical if it reaches 150.

  check_postgres_backends --host=quirm --warning=120 --critical=150

Example 2: Give a critical when we reach 75% of our max_connections setting on hosts lancre or lancre2.

  check_postgres_backends --warning='75%' --critical='75%' --host=lancre,lancre2

Example 3: Give a warning when there are only 10 more connection slots left on host plasmid, and a critical when we have only 5 left.

  check_postgres_backends --warning=-10 --critical=-5 --host=plasmid

Example 4: Check all databases except those with "test" in their name, but allow ones that are named "pg_greatest". Connect as port 5432 on the first two hosts, and as port 5433 on the third one. We want to always throw a critical when we reach 30 or more connections.

 check_postgres_backends --dbhost=hong,kong --dbhost=fooey --dbport=5432 --dbport=5433 --warning=30 --critical=30 --exclude="~test" --include="pg_greatest,~prod"

For MRTG output, the number of connections is reported on the first line, and the fourth line gives the name of the database, plus the current maximum_connections. If more than one database has been queried, the one with the highest number of connections is output.

bloat¶

("symlink: check_postgres_bloat") Checks the amount of bloat in tables and indexes. (Bloat is generally the amount of dead unused space taken up in a table or index. This space is usually reclaimed by use of the VACUUM command.) This action requires that stats collection be enabled on the target databases, and requires that ANALYZE is run frequently. The --include and --exclude options can be used to filter out which tables to look at. See the "BASIC FILTERING" section for more details.

The --warning and --critical options can be specified as sizes, percents, or both. Valid size units are bytes, kilobytes, megabytes, gigabytes, terabytes, exabytes, petabytes, and zettabytes. You can abbreviate all of those with the first letter. Items without units are assumed to be 'bytes'. The default values are '1 GB' and '5 GB'. The value represents the number of "wasted bytes", or the difference between what is actually used by the table and index, and what we compute that it should be.

Note that this action has two hard-coded values to avoid false alarms on smaller relations. Tables must have at least 10 pages, and indexes at least 15, before they can be considered by this test. If you really want to adjust these values, you can look for the variables $MINPAGES and $MINIPAGES at the top of the "check_bloat" subroutine. These values are ignored if either --exclude or --include is used.

Only the top 10 most bloated relations are shown. You can change this number by using the --perflimit option to set your own limit.

The schema named 'information_schema' is excluded from this test, as the only tables it contains are small and do not change.

Please note that the values computed by this action are not precise, and should be used as a guideline only. Great effort was made to estimate the correct size of a table, but in the end it is only an estimate. The correct index size is even more of a guess than the correct table size, but both should give a rough idea of how bloated things are.

Example 1: Warn if any table on port 5432 is over 100 MB bloated, and critical if over 200 MB

  check_postgres_bloat --port=5432 --warning='100 M' --critical='200 M'

Example 2: Give a critical if table 'orders' on host 'sami' has more than 10 megs of bloat

  check_postgres_bloat --host=sami --include=orders --critical='10 MB'

Example 3: Give a critical if table 'q4' on database 'sales' is over 50% bloated

  check_postgres_bloat --db=sales --include=q4 --critical='50%'

Example 4: Give a critical any table is over 20% bloated and has over 150 MB of bloat:

  check_postgres_bloat --port=5432 --critical='20% and 150 M'

Example 5: Give a critical any table is over 40% bloated or has over 500 MB of bloat:

  check_postgres_bloat --port=5432 --warning='500 M or 40%'

For MRTG output, the first line gives the highest number of wasted bytes for the tables, and the second line gives the highest number of wasted bytes for the indexes. The fourth line gives the database name, table name, and index name information. If you want to output the bloat ratio instead (how many times larger the relation is compared to how large it should be), just pass in "--mrtg=ratio".

checkpoint¶

("symlink: check_postgres_checkpoint") Determines how long since the last checkpoint has been run. This must run on the same server as the database that is being checked (e.g. the -h flag will not work). This check is meant to run on a "warm standby" server that is actively processing shipped WAL files, and is meant to check that your warm standby is truly 'warm'. The data directory must be set, either by the environment variable "PGDATA", or passing the "--datadir" argument. It returns the number of seconds since the last checkpoint was run, as determined by parsing the call to "pg_controldata". Because of this, the pg_controldata executable must be available in the current path. Alternatively, you can specify "PGBINDIR" as the directory that it lives in. It is also possible to use the special options --assume-prod or --assume-standby-mode, if the mode found is not the one expected, a CRITICAL is emitted.

At least one warning or critical argument must be set.

This action requires the Date::Parse module.

For MRTG or simple output, returns the number of seconds.

cluster_id¶

("symlink: check_postgres_cluster-id") Checks that the Database System Identifier provided by pg_controldata is the same as last time you checked. This must run on the same server as the database that is being checked (e.g. the -h flag will not work). Either the --warning or the --critical option should be given, but not both. The value of each one is the cluster identifier, an integer value. You can run with the special "--critical=0" option to find out an existing cluster identifier.

Example 1: Find the initial identifier

  check_postgres_cluster_id --critical=0 --datadir=/var//lib/postgresql/9.0/main

Example 2: Make sure the cluster is the same and warn if not, using the result from above.

  check_postgres_cluster_id  --critical=5633695740047915135

For MRTG output, returns a 1 or 0 indicating success of failure of the identifier to match. A identifier must be provided as the "--mrtg" argument. The fourth line always gives the current identifier.

commitratio¶

("symlink: check_postgres_commitratio") Checks the commit ratio of all databases and complains when they are too low. There is no need to run this command more than once per database cluster. Databases can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details. They can also be filtered by the owner of the database with the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The warning and critical options should be specified as percentages. There are not defaults for this action: the warning and critical must be specified. The warning value cannot be greater than the critical value. The output returns all databases sorted by commitratio, smallest first.

Example: Warn if any database on host flagg is less than 90% in commitratio, and critical if less then 80%.

  check_postgres_database_commitratio --host=flagg --warning='90%' --critical='80%'

For MRTG output, returns the percentage of the database with the smallest commitratio on the first line, and the name of the database on the fourth line.

connection¶

("symlink: check_postgres_connection") Simply connects, issues a 'SELECT version()', and leaves. Takes no --warning or --critical options.

For MRTG output, simply outputs a 1 (good connection) or a 0 (bad connection) on the first line.

custom_query¶

("symlink: check_postgres_custom_query") Runs a custom query of your choosing, and parses the results. The query itself is passed in through the "query" argument, and should be kept as simple as possible. If at all possible, wrap it in a view or a function to keep things easier to manage. The query should return one or two columns. It is required that one of the columns be named "result" and is the item that will be checked against your warning and critical values. The second column is for the performance data and any name can be used: this will be the 'value' inside the performance data section.

At least one warning or critical argument must be specified. What these are set to depends on the type of query you are running. There are four types of custom_queries that can be run, specified by the "valtype" argument. If none is specified, this action defaults to 'integer'. The four types are:

integer: Does a simple integer comparison. The first column should be a simple integer, and the warning and critical values should be the same.

string: The warning and critical are strings, and are triggered only if the value in the first column matches it exactly. This is case-sensitive.

time: The warning and the critical are times, and can have units of seconds, minutes, hours, or days. Each may be written singular or abbreviated to just the first letter. If no units are given, seconds are assumed. The first column should be an integer representing the number of seconds to check.

size: The warning and the critical are sizes, and can have units of bytes, kilobytes, megabytes, gigabytes, terabytes, or exabytes. Each may be abbreviated to the first letter. If no units are given, bytes are assumed. The first column should be an integer representing the number of bytes to check.

Normally, an alert is triggered if the values returned are greater than or equal to the critical or warning value. However, an option of --reverse will trigger the alert if the returned value is lower than or equal to the critical or warning value.

Example 1: Warn if any relation over 100 pages is named "rad", put the number of pages inside the performance data section.

  check_postgres_custom_query --valtype=string -w "rad" --query=
    "SELECT relname AS result, relpages AS pages FROM pg_class WHERE relpages > 100"

Example 2: Give a critical if the "foobar" function returns a number over 5MB:

  check_postgres_custom_query --critical='5MB'--valtype=size --query="SELECT foobar() AS result"

Example 2: Warn if the function "snazzo" returns less than 42:

  check_postgres_custom_query --critical=42 --query="SELECT snazzo() AS result" --reverse

If you come up with a useful custom_query, consider sending in a patch to this program to make it into a standard action that other people can use.

This action does not support MRTG or simple output yet.

database_size¶

("symlink: check_postgres_database_size") Checks the size of all databases and complains when they are too big. There is no need to run this command more than once per database cluster. Databases can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details. They can also be filtered by the owner of the database with the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The warning and critical options can be specified as bytes, kilobytes, megabytes, gigabytes, terabytes, or exabytes. Each may be abbreviated to the first letter as well. If no unit is given, the units are assumed to be bytes. There are not defaults for this action: the warning and critical must be specified. The warning value cannot be greater than the critical value. The output returns all databases sorted by size largest first, showing both raw bytes and a "pretty" version of the size.

Example 1: Warn if any database on host flagg is over 1 TB in size, and critical if over 1.1 TB.

  check_postgres_database_size --host=flagg --warning='1 TB' --critical='1.1 t'

Example 2: Give a critical if the database template1 on port 5432 is over 10 MB.

  check_postgres_database_size --port=5432 --include=template1 --warning='10MB' --critical='10MB'

Example 3: Give a warning if any database on host 'tardis' owned by the user 'tom' is over 5 GB

  check_postgres_database_size --host=tardis --includeuser=tom --warning='5 GB' --critical='10 GB'

For MRTG output, returns the size in bytes of the largest database on the first line, and the name of the database on the fourth line.

dbstats¶

("symlink: check_postgres_dbstats") Reports information from the pg_stat_database view, and outputs it in a Cacti-friendly manner. No other output is supported, as the output is informational and does not lend itself to alerts, such as used with Nagios. If no options are given, all databases are returned, one per line. You can include a specific database by use of the "--include" option, or you can use the "--dbname" option.

Eleven items are returned on each line, in the format name:value, separated by a single space. The items are:

backends

The number of currently running backends for this database.

commits

The total number of commits for this database since it was created or reset.

rollbacks

The total number of rollbacks for this database since it was created or reset.

read

The total number of disk blocks read.

hit

The total number of buffer hits.

ret

The total number of rows returned.

fetch

The total number of rows fetched.

ins

The total number of rows inserted.

upd

The total number of rows updated.

del

The total number of rows deleted.

dbname

The name of the database.

Note that ret, fetch, ins, upd, and del items will always be 0 if Postgres is version 8.2 or lower, as those stats were not available in those versions.

If the dbname argument is given, seven additional items are returned:

idxscan

Total number of user index scans.

idxtupread

Total number of user index entries returned.

idxtupfetch

Total number of rows fetched by simple user index scans.

idxblksread

Total number of disk blocks read for all user indexes.

idxblkshit

Total number of buffer hits for all user indexes.

seqscan

Total number of sequential scans against all user tables.

seqtupread

Total number of tuples returned from all user tables.

Example 1: Grab the stats for a database named "products" on host "willow":

  check_postgres_dbstats --dbhost willow --dbname products

The output returned will be like this (all on one line, not wrapped):

    backends:82 commits:58374408 rollbacks:1651 read:268435543 hit:2920381758 idxscan:310931294 idxtupread:2777040927
    idxtupfetch:1840241349 idxblksread:62860110 idxblkshit:1107812216 seqscan:5085305 seqtupread:5370500520
    ret:0 fetch:0 ins:0 upd:0 del:0 dbname:willow

disabled_triggers¶

("symlink: check_postgres_disabled_triggers") Checks on the number of disabled triggers inside the database. The --warning and --critical options are the number of such triggers found, and both default to "1", as in normal usage having disabled triggers is a dangerous event. If the database being checked is 8.3 or higher, the check is for the number of triggers that are in a 'disabled' status (as opposed to being 'always' or 'replica'). The output will show the name of the table and the name of the trigger for each disabled trigger.

Example 1: Make sure that there are no disabled triggers

  check_postgres_disabled_triggers

For MRTG output, returns the number of disabled triggers on the first line.

disk_space¶

("symlink: check_postgres_disk_space") Checks on the available physical disk space used by Postgres. This action requires that you have the executable "/bin/df" available to report on disk sizes, and it also needs to be run as a superuser (either connecting directly or switching via --role), so it can examine the data_directory setting inside of Postgres. The --warning and --critical options are given in either sizes or percentages or both. If using sizes, the standard unit types are allowed: bytes, kilobytes, gigabytes, megabytes, gigabytes, terabytes, or exabytes. Each may be abbreviated to the first letter only; no units at all indicates 'bytes'. The default values are '90%' and '95%'.

This command checks the following things to determine all of the different physical disks being used by Postgres.

data_directory - The disk that the main data directory is on.

log directory - The disk that the log files are on.

WAL file directory - The disk that the write-ahead logs are on (e.g. symlinked pg_xlog or pg_wal)

tablespaces - Each tablespace that is on a separate disk.

The output shows the total size used and available on each disk, as well as the percentage, ordered by highest to lowest percentage used. Each item above maps to a file system: these can be included or excluded. See the "BASIC FILTERING" section for more details.

Example 1: Make sure that no file system is over 90% for the database on port 5432.

  check_postgres_disk_space --port=5432 --warning='90%' --critical='90%'

Example 2: Check that all file systems starting with /dev/sda are smaller than 10 GB and 11 GB (warning and critical)

  check_postgres_disk_space --port=5432 --warning='10 GB' --critical='11 GB' --include="~^/dev/sda"

Example 4: Make sure that no file system is both over 50% and has over 15 GB

  check_postgres_disk_space --critical='50% and 15 GB'

Example 5: Issue a warning if any file system is either over 70% full or has more than 1T

  check_postgres_disk_space --warning='1T or 75'

For MRTG output, returns the size in bytes of the file system on the first line, and the name of the file system on the fourth line.

fsm_pages¶

("symlink: check_postgres_fsm_pages") Checks how close a cluster is to the Postgres max_fsm_pages setting. This action will only work for databases of 8.2 or higher, and it requires the contrib module pg_freespacemap be installed. The --warning and --critical options should be expressed as percentages. The number of used pages in the free-space-map is determined by looking in the pg_freespacemap_relations view, and running a formula based on the formula used for outputting free-space-map pageslots in the vacuum verbose command. The default values are 85% for the warning and 95% for the critical.

Example 1: Give a warning when our cluster has used up 76% of the free-space pageslots, with pg_freespacemap installed in database robert

  check_postgres_fsm_pages --dbname=robert --warning="76%"

While you need to pass in the name of the database where pg_freespacemap is installed, you only need to run this check once per cluster. Also, checking this information does require obtaining special locks on the free-space-map, so it is recommend you do not run this check with short intervals.

For MRTG output, returns the percent of free-space-map on the first line, and the number of pages currently used on the second line.

fsm_relations¶

("symlink: check_postgres_fsm_relations") Checks how close a cluster is to the Postgres max_fsm_relations setting. This action will only work for databases of 8.2 or higher, and it requires the contrib module pg_freespacemap be installed. The --warning and --critical options should be expressed as percentages. The number of used relations in the free-space-map is determined by looking in the pg_freespacemap_relations view. The default values are 85% for the warning and 95% for the critical.

Example 1: Give a warning when our cluster has used up 80% of the free-space relations, with pg_freespacemap installed in database dylan

  check_postgres_fsm_relations --dbname=dylan --warning="75%"

While you need to pass in the name of the database where pg_freespacemap is installed, you only need to run this check once per cluster. Also, checking this information does require obtaining special locks on the free-space-map, so it is recommend you do not run this check with short intervals.

For MRTG output, returns the percent of free-space-map on the first line, the number of relations currently used on the second line.

hitratio¶

("symlink: check_postgres_hitratio") Checks the hit ratio of all databases and complains when they are too low. There is no need to run this command more than once per database cluster. Databases can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details. They can also be filtered by the owner of the database with the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The warning and critical options should be specified as percentages. There are not defaults for this action: the warning and critical must be specified. The warning value cannot be greater than the critical value. The output returns all databases sorted by hitratio, smallest first.

Example: Warn if any database on host flagg is less than 90% in hitratio, and critical if less then 80%.

  check_postgres_hitratio --host=flagg --warning='90%' --critical='80%'

For MRTG output, returns the percentage of the database with the smallest hitratio on the first line, and the name of the database on the fourth line.

hot_standby_delay¶

("symlink: check_hot_standby_delay") Checks the streaming replication lag by computing the delta between the current xlog position of a master server and the replay location of a slave connected to it. The slave server must be in hot_standby (e.g. read only) mode, therefore the minimum version to use this action is Postgres 9.0. The --warning and --critical options are the delta between the xlog locations. Since these values are byte offsets in the WAL they should match the expected transaction volume of your application to prevent false positives or negatives.

The first "--dbname", "--host", and "--port", etc. options are considered the master; the second belongs to the slave.

Byte values should be based on the volume of transactions needed to have the streaming replication disconnect from the master because of too much lag, determined by the Postgres configuration variable wal_keep_segments. For units of time, valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. When specifying both, in the form 'bytes and time', both conditions must be true for the threshold to be met.

You must provide information on how to reach the databases by providing a comma separated list to the --dbhost and --dbport parameters, such as "--dbport=5432,5543". If not given, the action fails.

Example 1: Warn a database with a local replica on port 5433 is behind on any xlog replay at all

  check_hot_standby_delay --dbport=5432,5433 --warning='1'

Example 2: Give a critical if the last transaction replica1 receives is more than 10 minutes ago

  check_hot_standby_delay --dbhost=master,replica1 --critical='10 min'

Example 3: Allow replica1 to be 1 WAL segment behind, if the master is momentarily seeing more activity than the streaming replication connection can handle, or 10 minutes behind, if the master is seeing very little activity and not processing any transactions, but not both, which would indicate a lasting problem with the replication connection.

  check_hot_standby_delay --dbhost=master,replica1 --warning='1048576 and 2 min' --critical='16777216 and 10 min'

relation_size¶

index_size¶

table_size¶

indexes_size¶

total_relation_size¶

(symlinks: "check_postgres_relation_size", "check_postgres_index_size", "check_postgres_table_size", "check_postgres_indexes_size", and "check_postgres_total_relation_size")

The actions relation_size and index_size check for a relation (table, index, materialized view), respectively an index that has grown too big, using the pg_relation_size() function.

The action table_size checks tables and materialized views using pg_table_size(), i.e. including relation forks and TOAST table.

The action indexes_size checks tables and materialized views for the size of the attached indexes using pg_indexes_size().

The action total_relation_size checks relations using pg_total_relation_size(), i.e. including relation forks, indexes and TOAST table.

Relations can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details. Relations can also be filtered by the user that owns them, by using the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The values for the --warning and --critical options are file sizes, and may have units of bytes, kilobytes, megabytes, gigabytes, terabytes, or exabytes. Each can be abbreviated to the first letter. If no units are given, bytes are assumed. There are no default values: both the warning and the critical option must be given. The return text shows the size of the largest relation found.

If the --showperf option is enabled, all of the relations with their sizes will be given. To prevent this, it is recommended that you set the --perflimit option, which will cause the query to do a "ORDER BY size DESC LIMIT (perflimit)".

Example 1: Give a critical if any table is larger than 600MB on host burrick.

  check_postgres_table_size --critical='600 MB' --warning='600 MB' --host=burrick

Example 2: Warn if the table products is over 4 GB in size, and give a critical at 4.5 GB.

  check_postgres_table_size --host=burrick --warning='4 GB' --critical='4.5 GB' --include=products

Example 3: Warn if any index not owned by postgres goes over 500 MB.

  check_postgres_index_size --port=5432 --excludeuser=postgres -w 500MB -c 600MB

For MRTG output, returns the size in bytes of the largest relation, and the name of the database and relation as the fourth line.

last_analyze¶

last_vacuum¶

last_autoanalyze¶

last_autovacuum¶

(symlinks: "check_postgres_last_analyze", "check_postgres_last_vacuum", "check_postgres_last_autoanalyze", and "check_postgres_last_autovacuum") Checks how long it has been since vacuum (or analyze) was last run on each table in one or more databases. Use of these actions requires that the target database is version 8.3 or greater, or that the version is 8.2 and the configuration variable stats_row_level has been enabled. Tables can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details. Tables can also be filtered by their owner by use of the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The units for --warning and --critical are specified as times. Valid units are seconds, minutes, hours, and days; all can be abbreviated to the first letter. If no units are given, 'seconds' are assumed. The default values are '1 day' and '2 days'. Please note that there are cases in which this field does not get automatically populated. If certain tables are giving you problems, make sure that they have dead rows to vacuum, or just exclude them from the test.

The schema named 'information_schema' is excluded from this test, as the only tables it contains are small and do not change.

Note that the non-'auto' versions will also check on the auto versions as well. In other words, using last_vacuum will report on the last vacuum, whether it was a normal vacuum, or one run by the autovacuum daemon.

Example 1: Warn if any table has not been vacuumed in 3 days, and give a critical at a week, for host wormwood

  check_postgres_last_vacuum --host=wormwood --warning='3d' --critical='7d'

Example 2: Same as above, but skip tables belonging to the users 'eve' or 'mallory'

  check_postgres_last_vacuum --host=wormwood --warning='3d' --critical='7d' --excludeuser=eve,mallory

For MRTG output, returns (on the first line) the LEAST amount of time in seconds since a table was last vacuumed or analyzed. The fourth line returns the name of the database and name of the table.

listener¶

("symlink: check_postgres_listener") Confirm that someone is listening for one or more specific strings (using the LISTEN/NOTIFY system), by looking at the pg_listener table. Only one of warning or critical is needed. The format is a simple string representing the LISTEN target, or a tilde character followed by a string for a regular expression check. Note that this check will not work on versions of Postgres 9.0 or higher.

Example 1: Give a warning if nobody is listening for the string bucardo_mcp_ping on ports 5555 and 5556

  check_postgres_listener --port=5555,5556 --warning=bucardo_mcp_ping

Example 2: Give a critical if there are no active LISTEN requests matching 'grimm' on database oskar

  check_postgres_listener --db oskar --critical=~grimm

For MRTG output, returns a 1 or a 0 on the first, indicating success or failure. The name of the notice must be provided via the --mrtg option.

locks¶

("symlink: check_postgres_locks") Check the total number of locks on one or more databases. There is no need to run this more than once per database cluster. Databases can be filtered with the --include and --exclude options. See the "BASIC FILTERING" section for more details.

The --warning and --critical options can be specified as simple numbers, which represent the total number of locks, or they can be broken down by type of lock. Valid lock names are 'total', 'waiting', or the name of a lock type used by Postgres. These names are case-insensitive and do not need the "lock" part on the end, so exclusive will match 'ExclusiveLock'. The format is name=number, with different items separated by colons or semicolons (or any other symbol).

Example 1: Warn if the number of locks is 100 or more, and critical if 200 or more, on host garrett

  check_postgres_locks --host=garrett --warning=100 --critical=200

Example 2: On the host artemus, warn if 200 or more locks exist, and give a critical if over 250 total locks exist, or if over 20 exclusive locks exist, or if over 5 connections are waiting for a lock.

  check_postgres_locks --host=artemus --warning=200 --critical="total=250:waiting=5:exclusive=20"

For MRTG output, returns the number of locks on the first line, and the name of the database on the fourth line.

logfile¶

("symlink: check_postgres_logfile") Ensures that the logfile is in the expected location and is being logged to. This action issues a command that throws an error on each database it is checking, and ensures that the message shows up in the logs. It scans the various log_* settings inside of Postgres to figure out where the logs should be. If you are using syslog, it does a rough (but not foolproof) scan of /etc/syslog.conf. Alternatively, you can provide the name of the logfile with the --logfile option. This is especially useful if the logs have a custom rotation scheme driven be an external program. The --logfile option supports the following escape characters: "%Y %m %d %H", which represent the current year, month, date, and hour respectively. An error is always reported as critical unless the warning option has been passed in as a non-zero value. Other than that specific usage, the "--warning" and "--critical" options should not be used.

Example 1: On port 5432, ensure the logfile is being written to the file /home/greg/pg8.2.log

  check_postgres_logfile --port=5432 --logfile=/home/greg/pg8.2.log

Example 2: Same as above, but raise a warning, not a critical

  check_postgres_logfile --port=5432 --logfile=/home/greg/pg8.2.log -w 1

For MRTG output, returns a 1 or 0 on the first line, indicating success or failure. In case of a failure, the fourth line will provide more detail on the failure encountered.

new_version_bc¶

("symlink: check_postgres_new_version_bc") Checks if a newer version of the Bucardo program is available. The current version is obtained by running "bucardo_ctl --version". If a major upgrade is available, a warning is returned. If a revision upgrade is available, a critical is returned. (Bucardo is a master to slave, and master to master replication system for Postgres: see <https://bucardo.org/> for more information). See also the information on the "--get_method" option.

new_version_box¶

("symlink: check_postgres_new_version_box") Checks if a newer version of the boxinfo program is available. The current version is obtained by running "boxinfo.pl --version". If a major upgrade is available, a warning is returned. If a revision upgrade is available, a critical is returned. (boxinfo is a program for grabbing important information from a server and putting it into a HTML format: see <https://bucardo.org/Boxinfo/> for more information). See also the information on the "--get_method" option.

new_version_cp¶

("symlink: check_postgres_new_version_cp") Checks if a newer version of this program (check_postgres.pl) is available, by grabbing the version from a small text file on the main page of the home page for the project. Returns a warning if the returned version does not match the one you are running. Recommended interval to check is once a day. See also the information on the "--get_method" option.

new_version_pg¶

("symlink: check_postgres_new_version_pg") Checks if a newer revision of Postgres exists for each database connected to. Note that this only checks for revision, e.g. going from 8.3.6 to 8.3.7. Revisions are always 100% binary compatible and involve no dump and restore to upgrade. Revisions are made to address bugs, so upgrading as soon as possible is always recommended. Returns a warning if you do not have the latest revision. It is recommended this check is run at least once a day. See also the information on the "--get_method" option.

new_version_tnm¶

("symlink: check_postgres_new_version_tnm") Checks if a newer version of the tail_n_mail program is available. The current version is obtained by running "tail_n_mail --version". If a major upgrade is available, a warning is returned. If a revision upgrade is available, a critical is returned. (tail_n_mail is a log monitoring tool that can send mail when interesting events appear in your Postgres logs. See: <https://bucardo.org/tail_n_mail/> for more information). See also the information on the "--get_method" option.

pgb_pool_cl_active¶

pgb_pool_cl_waiting¶

pgb_pool_sv_active¶

pgb_pool_sv_idle¶

pgb_pool_sv_used¶

pgb_pool_sv_tested¶

pgb_pool_sv_login¶

pgb_pool_maxwait¶

(symlinks: "check_postgres_pgb_pool_cl_active", "check_postgres_pgb_pool_cl_waiting", "check_postgres_pgb_pool_sv_active", "check_postgres_pgb_pool_sv_idle", "check_postgres_pgb_pool_sv_used", "check_postgres_pgb_pool_sv_tested", "check_postgres_pgb_pool_sv_login", and "check_postgres_pgb_pool_maxwait")

Examines pgbouncer's pool statistics. Each pool has a set of "client" connections, referring to connections from external clients, and "server" connections, referring to connections to PostgreSQL itself. The related check_postgres actions are prefixed by "cl_" and "sv_", respectively. Active client connections are those connections currently linked with an active server connection. Client connections may also be "waiting", meaning they have not yet been allocated a server connection. Server connections are "active" (linked to a client), "idle" (standing by for a client connection to link with), "used" (just unlinked from a client, and not yet returned to the idle pool), "tested" (currently being tested) and "login" (in the process of logging in). The maxwait value shows how long in seconds the oldest waiting client connection has been waiting.

pgbouncer_backends¶

("symlink: check_postgres_pgbouncer_backends") Checks the current number of connections for one or more databases through pgbouncer, and optionally compares it to the maximum allowed, which is determined by the pgbouncer configuration variable max_client_conn. The --warning and --critical options can take one of three forms. First, a simple number can be given, which represents the number of connections at which the alert will be given. This choice does not use the max_connections setting. Second, the percentage of available connections can be given. Third, a negative number can be given which represents the number of connections left until max_connections is reached. The default values for --warning and --critical are '90%' and '95%'. You can also filter the databases by use of the --include and --exclude options. See the "BASIC FILTERING" section for more details.

To view only non-idle processes, you can use the --noidle argument. Note that the user you are running as (either connecting directly or switching via --role) must be a superuser for this to work properly.

Example 1: Give a warning when the number of connections on host quirm reaches 120, and a critical if it reaches 150.

  check_postgres_pgbouncer_backends --host=quirm --warning=120 --critical=150 -p 6432 -u pgbouncer

Example 2: Give a critical when we reach 75% of our max_connections setting on hosts lancre or lancre2.

  check_postgres_pgbouncer_backends --warning='75%' --critical='75%' --host=lancre,lancre2 -p 6432 -u pgbouncer

Example 3: Give a warning when there are only 10 more connection slots left on host plasmid, and a critical when we have only 5 left.

  check_postgres_pgbouncer_backends --warning=-10 --critical=-5 --host=plasmid -p 6432 -u pgbouncer

For MRTG output, the number of connections is reported on the first line, and the fourth line gives the name of the database, plus the current max_client_conn. If more than one database has been queried, the one with the highest number of connections is output.

pgbouncer_checksum¶

("symlink: check_postgres_pgbouncer_checksum") Checks that all the pgBouncer settings are the same as last time you checked. This is done by generating a checksum of a sorted list of setting names and their values. Note that you shouldn't specify the database name, it will automatically default to pgbouncer. Either the --warning or the --critical option should be given, but not both. The value of each one is the checksum, a 32-character hexadecimal value. You can run with the special "--critical=0" option to find out an existing checksum.

This action requires the Digest::MD5 module.

Example 1: Find the initial checksum for pgbouncer configuration on port 6432 using the default user (usually postgres)

  check_postgres_pgbouncer_checksum --port=6432 --critical=0

Example 2: Make sure no settings have changed and warn if so, using the checksum from above.

  check_postgres_pgbouncer_checksum --port=6432 --warning=cd2f3b5e129dc2b4f5c0f6d8d2e64231

For MRTG output, returns a 1 or 0 indicating success of failure of the checksum to match. A checksum must be provided as the "--mrtg" argument. The fourth line always gives the current checksum.

pgbouncer_maxwait¶

("symlink: check_postgres_pgbouncer_maxwait") Checks how long the first (oldest) client in the queue has been waiting, in seconds. If this starts increasing, then the current pool of servers does not handle requests quick enough. Reason may be either overloaded server or just too small of a pool_size setting in pbouncer config file. Databases can be filtered by use of the --include and --exclude options. See the "BASIC FILTERING" section for more details. The values or the --warning and --critical options are units of time, and must be provided (no default). Valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. If no units are given, the units are assumed to be seconds.

This action requires Postgres 8.3 or better.

Example 1: Give a critical if any transaction has been open for more than 10 minutes:

  check_postgres_pgbouncer_maxwait -p 6432 -u pgbouncer --critical='10 minutes'

For MRTG output, returns the maximum time in seconds a transaction has been open on the first line. The fourth line gives the name of the database.

pgagent_jobs¶

("symlink: check_postgres_pgagent_jobs") Checks that all the pgAgent jobs that have executed in the preceding interval of time have succeeded. This is done by checking for any steps that have a non-zero result.

Either "--warning" or "--critical", or both, may be specified as times, and jobs will be checked for failures withing the specified periods of time before the current time. Valid units are seconds, minutes, hours, and days; all can be abbreviated to the first letter. If no units are given, 'seconds' are assumed.

Example 1: Give a critical when any jobs executed in the last day have failed.

  check_postgres_pgagent_jobs --critical=1d

Example 2: Give a warning when any jobs executed in the last week have failed.

  check_postgres_pgagent_jobs --warning=7d

Example 3: Give a critical for jobs that have failed in the last 2 hours and a warning for jobs that have failed in the last 4 hours:

  check_postgres_pgagent_jobs --critical=2h --warning=4h

prepared_txns¶

("symlink: check_postgres_prepared_txns") Check on the age of any existing prepared transactions. Note that most people will NOT use prepared transactions, as they are part of two-part commit and complicated to maintain. They should also not be confused with prepared STATEMENTS, which is what most people think of when they hear prepare. The default value for a warning is 1 second, to detect any use of prepared transactions, which is probably a mistake on most systems. Warning and critical are the number of seconds a prepared transaction has been open before an alert is given.

Example 1: Give a warning on detecting any prepared transactions:

  check_postgres_prepared_txns -w 0

Example 2: Give a critical if any prepared transaction has been open longer than 10 seconds, but allow up to 360 seconds for the database 'shrike':

  check_postgres_prepared_txns --critical=10 --exclude=shrike
  check_postgres_prepared_txns --critical=360 --include=shrike

For MRTG output, returns the number of seconds the oldest transaction has been open as the first line, and which database is came from as the final line.

query_runtime¶

("symlink: check_postgres_query_runtime") Checks how long a specific query takes to run, by executing a "EXPLAIN ANALYZE" against it. The --warning and --critical options are the maximum amount of time the query should take. Valid units are seconds, minutes, and hours; any can be abbreviated to the first letter. If no units are given, 'seconds' are assumed. Both the warning and the critical option must be given. The name of the view or function to be run must be passed in to the --queryname option. It must consist of a single word (or schema.word), with optional parens at the end.

Example 1: Give a critical if the function named "speedtest" fails to run in 10 seconds or less.

  check_postgres_query_runtime --queryname='speedtest()' --critical=10 --warning=10

For MRTG output, reports the time in seconds for the query to complete on the first line. The fourth line lists the database.

query_time¶

("symlink: check_postgres_query_time") Checks the length of running queries on one or more databases. There is no need to run this more than once on the same database cluster. Note that this already excludes queries that are "idle in transaction". Databases can be filtered by using the --include and --exclude options. See the "BASIC FILTERING" section for more details. You can also filter on the user running the query with the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The values for the --warning and --critical options are amounts of time, and at least one must be provided (no defaults). Valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. If no units are given, the unit is assumed to be seconds.

This action requires Postgres 8.1 or better.

Example 1: Give a warning if any query has been running longer than 3 minutes, and a critical if longer than 5 minutes.

  check_postgres_query_time --port=5432 --warning='3 minutes' --critical='5 minutes'

Example 2: Using default values (2 and 5 minutes), check all databases except those starting with 'template'.

  check_postgres_query_time --port=5432 --exclude=~^template

Example 3: Warn if user 'don' has a query running over 20 seconds

  check_postgres_query_time --port=5432 --includeuser=don --warning=20s

For MRTG output, returns the length in seconds of the longest running query on the first line. The fourth line gives the name of the database.

replicate_row¶

("symlink: check_postgres_replicate_row") Checks that master-slave replication is working to one or more slaves.

The first "--dbname", "--host", and "--port", etc. options are considered the master; subsequent uses are the slaves. The values or the --warning and --critical options are units of time, and at least one must be provided (no defaults). Valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. If no units are given, the units are assumed to be seconds.

This check updates a single row on the master, and then measures how long it takes to be applied to the slaves. To do this, you need to pick a table that is being replicated, then find a row that can be changed, and is not going to be changed by any other process. A specific column of this row will be changed from one value to another. All of this is fed to the "repinfo" option, and should contain the following options, separated by commas: table name, primary key, key id, column, first value, second value.

Example 1: Slony is replicating a table named 'orders' from host 'alpha' to host 'beta', in the database 'sales'. The primary key of the table is named id, and we are going to test the row with an id of 3 (which is historical and never changed). There is a column named 'salesrep' that we are going to toggle from a value of 'slon' to 'nols' to check on the replication. We want to throw a warning if the replication does not happen within 10 seconds.

  check_postgres_replicate_row --host=alpha --dbname=sales --host=beta
  --dbname=sales --warning=10 --repinfo=orders,id,3,salesrep,slon,nols

Example 2: Bucardo is replicating a table named 'receipt' from host 'green' to hosts 'red', 'blue', and 'yellow'. The database for both sides is 'public'. The slave databases are running on port 5455. The primary key is named 'receipt_id', the row we want to use has a value of 9, and the column we want to change for the test is called 'zone'. We'll toggle between 'north' and 'south' for the value of this column, and throw a critical if the change is not on all three slaves within 5 seconds.

 check_postgres_replicate_row --host=green --port=5455 --host=red,blue,yellow
  --critical=5 --repinfo=receipt,receipt_id,9,zone,north,south

For MRTG output, returns on the first line the time in seconds the replication takes to finish. The maximum time is set to 4 minutes 30 seconds: if no replication has taken place in that long a time, an error is thrown.

replication_slots¶

("symlink: check_postgres_replication_slots") Check the quantity of WAL retained for any replication slots in the target database cluster. This is handy for monitoring environments where all WAL archiving and replication is taking place over replication slots.

Warning and critical are total bytes retained for the slot. E.g:

  check_postgres_replication_slots --port=5432 --host=yellow -warning=32M -critical=64M

Specific named slots can be monitored using --include/--exclude

same_schema¶

("symlink: check_postgres_same_schema") Verifies that two or more databases are identical as far as their schema (but not the data within). Unlike most other actions, this has no warning or critical criteria - the databases are either in sync, or are not. If they are different, a detailed list of the differences is presented.

You may want to exclude or filter out certain differences. The way to do this is to add strings to the "--filter" option. To exclude a type of object, use "noname", where 'name' is the type of object, for example, "noschema". To exclude objects of a certain type by a regular expression against their name, use "noname=regex". See the examples below for a better understanding.

The types of objects that can be filtered include:

user

schema

table

view

index

sequence

constraint

trigger

function

The filter option "noposition" prevents verification of the position of columns within a table.

The filter option "nofuncbody" prevents comparison of the bodies of all functions.

The filter option "noperm" prevents comparison of object permissions.

To provide the second database, just append the differences to the first one by a call to the appropriate connection argument. For example, to compare databases on hosts alpha and bravo, use "--dbhost=alpha,bravo". Also see the examples below.

If only a single host is given, it is assumed we are doing a "time-based" report. The first time this is run a snapshot of all the items in the database is saved to a local file. When you run it again, that snapshot is read in and becomes "database #2" and is compared to the current database.

To replace the old stored file with the new version, use the --replace argument.

If you need to write the stored file to a specific directory, use the --audit-file-dir argument.

To avoid false positives on value based checks caused by replication lag on asynchronous replicas, use the --assume-async option.

To enable snapshots at various points in time, you can use the "--suffix" argument to make the filenames unique to each run. See the examples below.

Example 1: Verify that two databases on hosts star and line are the same:

  check_postgres_same_schema --dbhost=star,line

Example 2: Same as before, but exclude any triggers with "slony" in their name

  check_postgres_same_schema --dbhost=star,line --filter="notrigger=slony"

Example 3: Same as before, but also exclude all indexes

  check_postgres_same_schema --dbhost=star,line --filter="notrigger=slony noindexes"

Example 4: Check differences for the database "battlestar" on different ports

  check_postgres_same_schema --dbname=battlestar --dbport=5432,5544

Example 5: Create a daily and weekly snapshot file

  check_postgres_same_schema --dbname=cylon --suffix=daily
  check_postgres_same_schema --dbname=cylon --suffix=weekly

Example 6: Run a historical comparison, then replace the file

  check_postgres_same_schema --dbname=cylon --suffix=daily --replace

Example 7: Verify that two databases on hosts star and line are the same, excluding value data (i.e. sequence last_val):

  check_postgres_same_schema --dbhost=star,line --assume-async

sequence¶

("symlink: check_postgres_sequence") Checks how much room is left on all sequences in the database. This is measured as the percent of total possible values that have been used for each sequence. The --warning and --critical options should be expressed as percentages. The default values are 85% for the warning and 95% for the critical. You may use --include and --exclude to control which sequences are to be checked. Note that this check does account for unusual minvalue and increment by values. By default it does not care if the sequence is set to cycle or not, and by passing --skipcycled sequenced set to cycle are reported with 0% usage.

The output for Nagios gives the name of the sequence, the percentage used, and the number of 'calls' left, indicating how many more times nextval can be called on that sequence before running into the maximum value.

The output for MRTG returns the highest percentage across all sequences on the first line, and the name of each sequence with that percentage on the fourth line, separated by a "|" (pipe) if there are more than one sequence at that percentage.

Example 1: Give a warning if any sequences are approaching 95% full.

  check_postgres_sequence --dbport=5432 --warning=95%

Example 2: Check that the sequence named "orders_id_seq" is not more than half full.

  check_postgres_sequence --dbport=5432 --critical=50% --include=orders_id_seq

settings_checksum¶

("symlink: check_postgres_settings_checksum") Checks that all the Postgres settings are the same as last time you checked. This is done by generating a checksum of a sorted list of setting names and their values. Note that different users in the same database may have different checksums, due to ALTER USER usage, and due to the fact that superusers see more settings than ordinary users. Either the --warning or the --critical option should be given, but not both. The value of each one is the checksum, a 32-character hexadecimal value. You can run with the special "--critical=0" option to find out an existing checksum.

This action requires the Digest::MD5 module.

Example 1: Find the initial checksum for the database on port 5555 using the default user (usually postgres)

  check_postgres_settings_checksum --port=5555 --critical=0

Example 2: Make sure no settings have changed and warn if so, using the checksum from above.

  check_postgres_settings_checksum --port=5555 --warning=cd2f3b5e129dc2b4f5c0f6d8d2e64231

For MRTG output, returns a 1 or 0 indicating success of failure of the checksum to match. A checksum must be provided as the "--mrtg" argument. The fourth line always gives the current checksum.

slony_status¶

("symlink: check_postgres_slony_status") Checks in the status of a Slony cluster by looking at the results of Slony's sl_status view. This is returned as the number of seconds of "lag time". The --warning and --critical options should be expressed as times. The default values are 60 seconds for the warning and 300 seconds for the critical.

The optional argument --schema indicated the schema that Slony is installed under. If it is not given, the schema will be determined automatically each time this check is run.

Example 1: Give a warning if any Slony is lagged by more than 20 seconds

  check_postgres_slony_status --warning 20

Example 2: Give a critical if Slony, installed under the schema "_slony", is over 10 minutes lagged

  check_postgres_slony_status --schema=_slony --critical=600

timesync¶

("symlink: check_postgres_timesync") Compares the local system time with the time reported by one or more databases. The --warning and --critical options represent the number of seconds between the two systems before an alert is given. If neither is specified, the default values are used, which are '2' and '5'. The warning value cannot be greater than the critical value. Due to the non-exact nature of this test, values of '0' or '1' are not recommended.

The string returned shows the time difference as well as the time on each side written out.

Example 1: Check that databases on hosts ankh, morpork, and klatch are no more than 3 seconds off from the local time:

  check_postgres_timesync --host=ankh,morpork,klatch --critical=3

For MRTG output, returns one the first line the number of seconds difference between the local time and the database time. The fourth line returns the name of the database.

txn_idle¶

("symlink: check_postgres_txn_idle") Checks the number and duration of "idle in transaction" queries on one or more databases. There is no need to run this more than once on the same database cluster. Databases can be filtered by using the --include and --exclude options. See the "BASIC FILTERING" section below for more details.

The --warning and --critical options are given as units of time, signed integers, or integers for units of time, and at least one must be provided (there are no defaults). Valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. If no units are given and the numbers are unsigned, the units are assumed to be seconds.

This action requires Postgres 8.3 or better.

As of PostgreSQL 10, you can just GRANT pg_read_all_stats to an unprivileged user account. In all earlier versions, superuser privileges are required to see the queries of all users in the system; UNKNOWN is returned if queries cannot be checked. To only include queries by the connecting user, use --includeuser.

Example 1: Give a warning if any connection has been idle in transaction for more than 15 seconds:

  check_postgres_txn_idle --port=5432 --warning='15 seconds'

Example 2: Give a warning if there are 50 or more transactions

  check_postgres_txn_idle --port=5432 --warning='+50'

Example 3: Give a critical if 5 or more connections have been idle in transaction for more than 10 seconds:

  check_postgres_txn_idle --port=5432 --critical='5 for 10 seconds'

For MRTG output, returns the time in seconds the longest idle transaction has been running. The fourth line returns the name of the database and other information about the longest transaction.

txn_time¶

("symlink: check_postgres_txn_time") Checks the length of open transactions on one or more databases. There is no need to run this command more than once per database cluster. Databases can be filtered by use of the --include and --exclude options. See the "BASIC FILTERING" section for more details. The owner of the transaction can also be filtered, by use of the --includeuser and --excludeuser options. See the "USER NAME FILTERING" section for more details.

The values or the --warning and --critical options are units of time, and at least one must be provided (no default). Valid units are 'seconds', 'minutes', 'hours', or 'days'. Each may be written singular or abbreviated to just the first letter. If no units are given, the units are assumed to be seconds.

This action requires Postgres 8.3 or better.

Example 1: Give a critical if any transaction has been open for more than 10 minutes:

  check_postgres_txn_time --port=5432 --critical='10 minutes'

Example 1: Warn if user 'warehouse' has a transaction open over 30 seconds

  check_postgres_txn_time --port-5432 --warning=30s --includeuser=warehouse

For MRTG output, returns the maximum time in seconds a transaction has been open on the first line. The fourth line gives the name of the database.

txn_wraparound¶

("symlink: check_postgres_txn_wraparound") Checks how close to transaction wraparound one or more databases are getting. The --warning and --critical options indicate the number of transactions done, and must be a positive integer. If either option is not given, the default values of 1.3 and 1.4 billion are used. There is no need to run this command more than once per database cluster. For a more detailed discussion of what this number represents and what to do about it, please visit the page <https://www.postgresql.org/docs/current/static/routine-vacuuming.html#VACUUM-FOR-WRAPAROUND>

The warning and critical values can have underscores in the number for legibility, as Perl does.

Example 1: Check the default values for the localhost database

  check_postgres_txn_wraparound --host=localhost

Example 2: Check port 6000 and give a critical when 1.7 billion transactions are hit:

  check_postgres_txn_wraparound --port=6000 --critical=1_700_000_000

For MRTG output, returns the highest number of transactions for all databases on line one, while line 4 indicates which database it is.

version¶

("symlink: check_postgres_version") Checks that the required version of Postgres is running. The --warning and --critical options (only one is required) must be of the format X.Y or X.Y.Z where X is the major version number, Y is the minor version number, and Z is the revision.

Example 1: Give a warning if the database on port 5678 is not version 8.4.10:

  check_postgres_version --port=5678 -w=8.4.10

Example 2: Give a warning if any databases on hosts valley,grain, or sunshine is not 8.3:

  check_postgres_version -H valley,grain,sunshine --critical=8.3

For MRTG output, reports a 1 or a 0 indicating success or failure on the first line. The fourth line indicates the current version. The version must be provided via the "--mrtg" option.

wal_files¶

("symlink: check_postgres_wal_files") Checks how many WAL files exist in the pg_xlog directory (PostgreSQL 10 and later" pg_wal), which is found off of your data_directory, sometimes as a symlink to another physical disk for performance reasons. If the --lsfunc option is not used then this action must be run as a superuser, in order to access the contents of the pg_xlog directory. The minimum version to use this action is Postgres 8.1. The --warning and --critical options are simply the number of files in the pg_xlog directory. What number to set this to will vary, but a general guideline is to put a number slightly higher than what is normally there, to catch problems early.

Normally, WAL files are closed and then re-used, but a long-running open transaction, or a faulty archive_command script, may cause Postgres to create too many files. Ultimately, this will cause the disk they are on to run out of space, at which point Postgres will shut down.

To avoid connecting as a database superuser, a wrapper function around pg_ls_dir() should be defined as a superuser with SECURITY DEFINER, and the --lsfunc option used. This example function, if defined by a superuser, will allow the script to connect as a normal user nagios with --lsfunc=ls_xlog_dir

  BEGIN;
  CREATE FUNCTION ls_xlog_dir()
      RETURNS SETOF TEXT
      AS $$ SELECT pg_ls_dir('pg_xlog') $$
      LANGUAGE SQL
      SECURITY DEFINER;
  REVOKE ALL ON FUNCTION ls_xlog_dir() FROM PUBLIC;
  GRANT EXECUTE ON FUNCTION ls_xlog_dir() to nagios;
  COMMIT;

Example 1: Check that the number of ready WAL files is 10 or less on host "pluto", using a wrapper function "ls_xlog_dir" to avoid the need for superuser permissions

  check_postgres_archive_ready --host=pluto --critical=10 --lsfunc=ls_xlog_dir

For MRTG output, reports the number of WAL files on line 1.

rebuild_symlinks¶

rebuild_symlinks_force¶

This action requires no other arguments, and does not connect to any databases, but simply creates symlinks in the current directory for each action, in the form check_postgres_<action_name>. If the file already exists, it will not be overwritten. If the action is rebuild_symlinks_force, then symlinks will be overwritten. The option --symlinks is a shorter way of saying --action=rebuild_symlinks