pganalyze collector

This is a Go-based daemon which collects various information about Postgres databases as well as queries run on it.

All data is converted to a protocol buffers structure which can then be used as data source for monitoring & graphing systems. Or just as reference on how to pull information out of PostgreSQL.

It currently collects information about

• Schema
  • Tables (including column, constraint and trigger definitions)
  • Indexes
• Statistics
  • Tables
  • Indexes
  • Database
  • Queries
• OS
  • CPU
  • Memory
  • Storage

Installation

The collector is available in multiple convenient options:

• APT/YUM packages: https://packages.pganalyze.com/
• Docker sidekick service, see details further down in this file

Configuration (APT/YUM Packages)

After the package was installed, you can find the configuration in /etc/pganalyze-collector.conf

Adjust the values in that file by adding your API key (found in the pganalyze dashboard, use one per database server), and database connection credentials.

You can repeat the configuration block with a different [name] if you have multiple servers to monitor.

See https://pganalyze.com/docs for further details.

Setting up a Restricted Monitoring User

By default pg_stat_statements does not allow viewing queries run by other users, unless you are a database superuser. Since you probably don't want monitoring to run as a superuser, you can setup a separate monitoring user like this:

CREATE SCHEMA pganalyze;
GRANT USAGE ON SCHEMA pganalyze TO pganalyze;

CREATE EXTENSION IF NOT EXISTS pg_stat_statements WITH SCHEMA public;

CREATE OR REPLACE FUNCTION pganalyze.get_stat_statements(showtext boolean = true) RETURNS SETOF pg_stat_statements AS
$$
  /* pganalyze-collector */ SELECT * FROM public.pg_stat_statements(showtext);
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

CREATE OR REPLACE FUNCTION pganalyze.get_stat_activity() RETURNS SETOF pg_stat_activity AS
$$
  /* pganalyze-collector */ SELECT * FROM pg_catalog.pg_stat_activity;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

CREATE OR REPLACE FUNCTION pganalyze.get_column_stats() RETURNS SETOF pg_stats AS
$$
  /* pganalyze-collector */ SELECT schemaname, tablename, attname, inherited, null_frac, avg_width,
  n_distinct, NULL::anyarray, most_common_freqs, NULL::anyarray, correlation, NULL::anyarray,
  most_common_elem_freqs, elem_count_histogram
  FROM pg_catalog.pg_stats;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

CREATE OR REPLACE FUNCTION pganalyze.get_stat_replication() RETURNS SETOF pg_stat_replication AS
$$
  /* pganalyze-collector */ SELECT * FROM pg_catalog.pg_stat_replication;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

CREATE USER pganalyze WITH PASSWORD 'mypassword' CONNECTION LIMIT 5;
REVOKE ALL ON SCHEMA public FROM pganalyze;
GRANT USAGE ON SCHEMA pganalyze TO pganalyze;

Note that these statements must be run as a superuser (to create the SECURITY DEFINER function), but from here onwards you can use the pganalyze user instead.

The collector will automatically use the helper methods if they exist in the pganalyze schema - otherwise data will be fetched directly.

If you are using the Buffer Cache report in pganalyze, you will also need to create this additional helper method:

CREATE EXTENSION IF NOT EXISTS pg_buffercache WITH SCHEMA public;
CREATE OR REPLACE FUNCTION pganalyze.get_buffercache() RETURNS SETOF public.pg_buffercache AS
$$
  /* pganalyze-collector */ SELECT * FROM public.pg_buffercache;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

If you are using the Sequence report in pganalyze, you will also need these helper methods:

CREATE OR REPLACE FUNCTION pganalyze.get_sequence_oid_for_column(table_name text, column_name text) RETURNS oid AS
$$
  /* pganalyze-collector */ SELECT pg_get_serial_sequence(table_name, column_name)::regclass::oid;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

--- The following is needed for Postgres 10+:

CREATE OR REPLACE FUNCTION pganalyze.get_sequence_state(schema_name text, sequence_name text) RETURNS TABLE(
  last_value bigint, start_value bigint, increment_by bigint,
  max_value bigint, min_value bigint, cache_size bigint, cycle boolean
) AS
$$
  /* pganalyze-collector */ SELECT last_value, start_value, increment_by, max_value, min_value, cache_size, cycle
    FROM pg_sequences WHERE schemaname = schema_name AND sequencename = sequence_name;
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

If you enabled the optional reset mode (usually not required), you will also need this helper method:

CREATE OR REPLACE FUNCTION pganalyze.reset_stat_statements() RETURNS SETOF void AS
$$
  /* pganalyze-collector */ SELECT * FROM public.pg_stat_statements_reset();
$$ LANGUAGE sql VOLATILE SECURITY DEFINER;

Setting up Log EXPLAIN helper

This section is relevant only if you use the enable_log_explain setting; if you use the recommended autoexplain extension, or if you do not plan to use EXPLAIN plan collection, you can skip it.

If you use enable_log_explain, use a superuser connection to create the pganalyze schema and this function on each database where EXPLAIN should run:

CREATE OR REPLACE FUNCTION pganalyze.explain(query text, params text[]) RETURNS text AS
$$
DECLARE
  prepared_query text;
  prepared_params text;
  result text;
BEGIN
  SELECT regexp_replace(query, ';+\s*\Z', '') INTO prepared_query;
  IF prepared_query LIKE '%;%' THEN
    RAISE EXCEPTION 'cannot run EXPLAIN when query contains semicolon';
  END IF;

  IF array_length(params, 1) > 0 THEN
    SELECT string_agg(quote_literal(param) || '::unknown', ',') FROM unnest(params) p(param) INTO prepared_params;

    EXECUTE 'PREPARE pganalyze_explain AS ' || prepared_query;
    BEGIN
      EXECUTE 'EXPLAIN (VERBOSE, FORMAT JSON) EXECUTE pganalyze_explain(' || prepared_params || ')' INTO STRICT result;
    EXCEPTION WHEN OTHERS THEN
      DEALLOCATE pganalyze_explain;
      RAISE;
    END;
    DEALLOCATE pganalyze_explain;
  ELSE
    EXECUTE 'EXPLAIN (VERBOSE, FORMAT JSON) ' || prepared_query INTO STRICT result;
  END IF;

  RETURN result;
END
$$ LANGUAGE plpgsql VOLATILE SECURITY DEFINER;

Note that this function contains a check for semicolons in the query text. This is to minimize collector access to your data: it ensures that the collector cannot piggyback other queries that could exfiltrate data.

Setting up log pg_read_file helper

Create the following helper as a superuser (note this needs to be an actual superuser, i.e. not available on systems like Amazon RDS), to read logs using the pg_read_file function with the restricted monitoring user:

CREATE OR REPLACE FUNCTION pganalyze.read_log_file(log_filename text, read_offset bigint, read_length bigint) RETURNS text AS
$$
DECLARE
  result text;
BEGIN
  IF log_filename !~ '\A[\w\.-]+\Z' THEN
    RAISE EXCEPTION 'invalid log filename';
  END IF;

  SELECT pg_catalog.pg_read_file(
    pg_catalog.current_setting('data_directory') || '/' || pg_catalog.current_setting('log_directory') || '/' || log_filename,
    read_offset,
    read_length
  ) INTO result;

  RETURN result;
END
$$ LANGUAGE plpgsql VOLATILE SECURITY DEFINER;

Example output

To get a feel for the data that is collected you can run the following command. This will show the data that would be sent (in JSON format), without sending it:

pganalyze-collector --dry-run

Don't hesitate to reach out to [email protected] if you have any questions about what gets sent, or how to adjust the collector data collection.

Docker Container (RDS)

If you are monitoring an RDS database and want to run the collector inside Docker, we recommend the following:

docker pull quay.io/pganalyze/collector:stable
docker run \
  --rm \
  --name pganalyze-mydb \
  -e DB_URL=postgres://username:[email protected]/mydb \
  -e PGA_API_KEY=YOUR_PGANALYZE_API_KEY \
  -e AWS_INSTANCE_ID=YOUR_RDS_DB_IDENTIFIER \
  quay.io/pganalyze/collector:stable

You'll need to set PGA_API_KEY, DB_URL, and AWS_INSTANCE_ID with the correct values.

Please also note that the EC2 instance running your Docker setup needs to have an IAM role that allows Cloudwatch access: https://pganalyze.com/docs/install/amazon_rds/03_setup_iam_policy

To get better data quality for server metrics, enable "Enhanced Monitoring" in your RDS dashboard. The pganalyze collector will automatically pick this up and get all the metrics.

We currently require one Docker container per RDS instance monitored.

If you have multiple databases on the same RDS instance, you can monitor them all by specifying DB_ALL_NAMES=1 as an environment variable.

Docker Container (non-RDS)

If the database you want to monitor is running inside a Docker environment you can use the Docker image:

docker pull quay.io/pganalyze/collector:stable
docker run \
  --name my-app-pga-collector \
  --link my-app-db:db \
  --env-file collector_config.env \
  quay.io/pganalyze/collector:stable

collector_config.env needs to look like this:

PGA_API_KEY=$YOUR_API_KEY
PGA_ALWAYS_COLLECT_SYSTEM_DATA=1
DB_NAME=your_database_name
DB_USERNAME=your_database_user
DB_PASSWORD=your_database_password

The only required arguments are PGA_API_KEY (found in the pganalyze dashboard) and DB_NAME. Only specify PGA_ALWAYS_COLLECT_SYSTEM_DATA if the database is running on the same host and you'd like the collector to gather system metrics (from inside the container).

Note: You can add -v /path/to/database/volume/on/host:/var/lib/postgresql/data in order to collect I/O statistics from your database (this requires that it runs on the same machine).

Heroku Monitoring

When monitoring a Heroku Postgres database, it is recommended you deploy the collector as its own app inside your Heroku account.

Follow the instructions in the pganalyze documentation to add your databases to the collector.

Success/Error Callbacks

In case you want to run a script based on data collection running successfully and/or failing, you can set the success_callback and error_callback options:

[pganalyze]
...
error_callback=/usr/local/bin/my_error_script.sh

[mydb]
...

Note that the callback is executed in a shell, so you can use shell expressions as well.

The script will also have the following environment variables set:

• PGA_CALLBACK_TYPE (type of callback, error or success)
• PGA_CONFIG_SECTION (server that was processed, mydb in this example)
• PGA_SNAPSHOT_TYPE (type of data that was processed, currently there are full snapshots, as well as logs snapshots which contain only log data)
• PGA_ERROR_MESSAGE (error message, in the case of the error callback)

Authors

• Lukas Fittl
• Michael Renner

License

pganalyze-collector is licensed under the 3-clause BSD license, see LICENSE file for details.