README.org

December 6, 2018 ยท View on GitHub

  • pgcmp ** Purpose

    This toolset is intended to perform comparisons ("reconciliation") of schemas between databases to determine if they are /equivalent/. There are several contexts where this is useful, each with a somewhat different understanding of /equivalent/.

    1. Build process
      • The build process has database schema in two forms:
        • Build From Scratch :: A deployable database is built from scratch, from an empty database
        • Upgrade From Earlier Version :: An older version is upgraded to the latest version using a set of /upgrade scripts/.
      • We need to ensure that the results of building from scratch are /identical/ to those that come from upgrading the old version using upgrade scripts.
      • In this case, we do not expect /any/ variations, aside from the fact that OIDs will vary between databases. These databases will be built within the same smoke test environment, and thus have the same sets of users, roles, and such, and there should not be any additional components introduced (as might be the case in production)
      • It is therefore considered a build-breaking error to find /any/ variation between the two database schemas
    2. Validating upgrades
      • When DBAs are planning to upgrade an instance, using the /upgrade scripts/, it is essential that their starting point represents a schema that is /equivalent/ to the previous version. Otherwise, if the starting point is /wrong/, then the upgrade scripts may not provide a proper upgrade to the new version of the schema.
      • In this environment, there are a number of /acceptable/ kinds of differences that should be accommodated between the "production" schema and a "development" schema:
        • Replication :: Development does not include replication, but in production, Slony schema, tables, functions, and such will exist, and the reconciliation should be able to exclude these differences.
        • DBA Tooling :: DBAs add additional tooling into production databases, generally in their own schemas, to help with monitoring.
        • Supplementary Applications :: Some applications have add-ons, and, when considering the main application, we may wish to exclude add-ons.
        • Specialized Ownership :: Often, different users are used in production than are used in development.
      • As a result, when reconciling the "old" schema version against the version in production, it will be necessary to accept the need for /reconciling entries/ so that DBAs are not left puzzling over hundreds of differences that may be reasonably /expected/.
    3. Validating replicas in sync
      • When deploying an upgrade, it is not sufficient to know that the "master" node has an agreeable schema; it is essential that /all/ replicas have suitable schemas.
      • In the above analysis, it was necessary to accommodate there being considerable difference between the "development" schema and the "production" schema. When comparing replicas, differences should be considered rather more dangerous. Differences to accept/accommodate include:
        • Node-specific deployment :: In some cases, add-ons are only installed on certain nodes in the cluster.
        • Origin/Replica differences :: Slony defines some triggers that run only on origin nodes (/e.g./ - logtrigger()), and others that run only on replica nodes (/e.g./ - denyaccess()). ** The Comparison Script pgcmp tooling consists of pair of scripts:

    This script pulls data about the schema from the two databases that are to be compared, loads that data into a third database where that data is compared in order, and performs a reconciliation of the similarities and differences.

    • pgcmp-dump :: This tool extracts schema data from a database, putting it into a data file for later analysis
    • pgcmp :: This tool uses a pair of files generated by pgcmp-dump, and looks for differences.

*** Security Considerations This script requires /read-only/ access to schema data in INFORMATION_SCHEMA and pg_catalog in both databases that are to be compared.

This script requires /write/ access to a third database in which
it constructs a series of tables used to perform the
reconciliation.

*** Running pgcmp components

**** pgcmp-dump

pgcmp-dump uses a series of environment variables to control its activity.

  • Database connection information
    • PGDB :: Database name
    • PPGGHOST :: Database host
    • PGUSER :: DB User
    • PGPORT :: Port number
    • PGPGURI :: Since PostgreSQL 9.2, [[https://www.postgresql.org/docs/9.2/static/libpq-connect.html][libpq supports PGURIs]] as a singular connection string. pgcmp prefers the use of these URIs. If the individual parameters above are offered, they will be constructed into a URI; if PGURI is populated, it will be used instead.
  • PGCLABEL :: Indicates an identifer for the database's data to report in analysis output
  • PGCMPOUTPUT :: The name of the file in which to store output, that is, the set of metadata about the database schema that will be used later for comparison with the schema of another database.
  • PGCMPIGNORESLONYTRIGGERS :: defaults to false; if true, then ignore differences involving triggers of the form created by Slony-I

#+BEGIN_EXAMPLE $ PGURI=postgresql://postgres@localhost/test1 PGCMPOUTPUT=/tmp/test-pgcmp-file1 PGCLABEL=db1 ../pgcmp-dump pgcmp-dump - extracting schema data from database PGURI=[postgresql://postgres@localhost/test1]

Data file containing output: PGCMPOUTPUT[/tmp/test-pgcmp-file1] Label: PGCLABEL=[db1]

Extracted schema data for postgresql://postgres@localhost/test1 -rw-r--r-- 1 cbbrowne cbbrowne 7350 Nov 3 12:12 /tmp/test-pgcmp-file1 #+END_EXAMPLE

**** pgcmp

pgcmp uses a pair of input files as produced by pgcmp-dump, loads them into a comparison database, and checks for differences.

It uses a series of environment variables to control its activity.

  • PGBINDIR :: Indicates where to find psql

  • Database connection information

    • PGDB :: Database name
    • PPGGHOST :: Database host
    • PGUSER :: DB User
    • PGPORT :: Port number
    • PGURI :: Since PostgreSQL 9.2, [[https://www.postgresql.org/docs/9.2/static/libpq-connect.html][libpq supports PGURIs]] as a singular connection string. pgcmp prefers the use of these URIs. If the individual parameters above are offered, they will be constructed into a URI; if PGURI is populated, it will be used instead.

  • PGCLABEL1 :: Label identifying data for the first database

  • PGCLABEL2 :: Label identifying data for the second database

  • PGCOMITSCHEMAS :: Indicates (as a SQL clause suitable to insert VALUES into a table), a set of schema labels from pg_catalog.pg_namespace that should be omitted from analysis. As identifiers, they may require doublequotes consistent with the function pg_catalog.quote_ident(). pg_catalog and information_schema were automatically left out in pgcmp-dump, so they do not need to be included.

  • Data file information

    • PGCWORKDIR :: Work directory where work files are placed/found, defaults to /tmp
    • PGCEXPLANATIONS :: Contains records intended to explain expected variations
      • Structure is a Postgres COPY of data for a table of expected differences, defined as follows: #+BEGIN_EXAMPLE create table expected_differences ( object_type text, object_name text, difference_type text, difference_cause text, primary key(object_type, object_name, difference_type, difference_cause) ); #+END_EXAMPLE
    • PGCFULLOUTPUT :: Contains full output of all records, those that match and those that do not.
      • Structure is a Postgres COPY of data from a table with the following structure: #+BEGIN_EXAMPLE create table fulloutput ( object_name text, object_type text, label1 text, object_definition1 text, label2 text, object_definition2 text, difference_type text, difference_cause text, importance integer ); #+END_EXAMPLE
    • PGCUNEXPLAINED :: Contains the set of records that did not match between schemas that did not have explanations provided.
      • Structure is a Postgres COPY of data from a table with the following structure: #+BEGIN_EXAMPLE create table unexplained_items ( object_name text, object_schema text, object_type text, label1 text, id1 integer, object_definition1 text, label2 text, id2 integer, object_definition2 text, difference_type text, importance integer ); #+END_EXAMPLE
    • PGCBADEXPLAIN :: Contains the set of records that did not match between schemas where explanations offered did not match the problem.
      • Structure is a Postgres COPY of data from a table with the following structure: #+BEGIN_EXAMPLE create table badexplanations_items ( object_type text, object_schema text, object_name text, difference_type text, difference_cause text, importance integer ); #+END_EXAMPLE
    • PGCMPINPUT1 :: Input file containing metadata about the schema for the first database
    • PGCMPINPUT2 :: Input file containing metadata about the schema for the second database

**** Interpretations of differences

  • object_type :: Indicates which kind of object had a difference
  • object_schema :: Indicates the namespace where the object occurs
  • object_name :: Fully qualified name of the object
  • difference_type :: Kind of difference, one of | type | description | |-------------------+-----------------------------------------| | match | perfect match across the databases | | mismatch | values differ between databases | | missing in 1st DB | exists in 2nd database, but not 1st one | | missing in 2nd DB | exists in 1st database, but not 2nd one |
  • difference_cause :: Indicates whether a difference has been explained satisfactorily via the explanations.txt data
    • Not Yet Properly Explained :: Indicates the difference has not been explained at all
    • Still Not Properly Explained :: Indicates that explanations.txt indicates that the problem was not explained. In effect, the administrator took the data straight out of an earlier pgcmp run, and tried to use it as an explanation, as opposed to analyzing it to determine what should be done about it.
    • Misexplained :: Indicates that the explanation in the explanations.txt file indicated a different sort of difference_type than was found by pgcmp.
  • importance :: Indicates how crucial a discrepancy is. Low importance values indicate the most crucial items. Generally:
    • NULL :: no discrepancy found; no difference to be examined
    • 1 :: mismatch between databases, likely indicating a problem to be fixed
    • 2 :: object is present in one database, absent in the other
    • 3 or more :: object is present in one database, absent in the other, but is a "child" of some other missing object. For instance, if a schema is missing, then that would be indicated as importance=2, the tables, views, sequences, and such, in that schema that are missing in one database would be marked with importance=3, and columns, indexes, and other attributes of tables/views/sequences would be marked with importance=4. The items with importance of 3 or 4 are the consequence of the importance=2 item; the focus of an analyst should be in the high importance (1,2) items.

**** Input File: explanations.txt

The input file, explanations.txt, provides a set of explanation items that allow an administrator to indicate explanations for discrepancies that are considered acceptable.

For instance:

  • Production includes Slony :: If comparing a /development/ schema that does not include replication against a /production/ schema where [[http://www.slony.info/][Slony]] has been installed, it is to be expected that all of the Slony objects will comprise a set of "expected" differences.
  • Production Monitoring :: DBAs may add in additional components such as the pgstattuples contrib module, or even additional schemas and tables.
  • Production Users :: The production environment may be expected to have additional users and roles not found in the /development/ schema.

#+BEGIN_EXAMPLE create table expected_differences ( object_type text, object_schema text, object_name text, difference_type text, difference_cause text, importance integer, primary key(object_type, object_name, difference_type, difference_cause) ); #+END_EXAMPLE

***** How To Populate explanations.txt

  The easiest way to populate this file is by running
  ~pgcmp.sh~ with an /empty/ set of explanations, and
  then transforming the resulting set of unexplained items into
  "explained" differences.

#+BEGIN_EXAMPLE insert into expected_differences (object_type, object_schema, object_name, difference_type, difference_cause) select object_type, object_schema, object_name, difference_type, 'Slony objects only in production' from unexplained_items where object_name like '_oxrspro%';

insert into expected_differences (object_type, object_schema, object_name, difference_type, difference_cause) select object_type, object_schema, object_name, difference_type, 'contrib objects only in production' from unexplained_items where object_name like 'postgres_contrib%';

insert into expected_differences (object_type, object_schema, object_name, difference_type, difference_cause) select object_type, object_schema, object_name, difference_type, 'Conversion objects to be removed from production' from unexplained_items where object_name like 'dotpro_conversion%';

\copy expected_differences to '/tmp/expected_differences.txt'; #+END_EXAMPLE

  In subsequent runs, these differences become "expected"
  differences, so that a DBA or QA analyst does not need to spend
  their attention manually filtering out these expected
  differences.

*** Outputs The process has output in several forms: - Brief report to standard output - Files containing details - Return codes useful for determining success/failure **** Brief Report Here is an example of running a comparison between two schemas: - Parameters :: lists values for all the environment variables - Extraction Summary :: lists information about the files of extracted schema data - SQL messages :: lists commands run against the comparison database - Results Summary :: indicates, by object type, statistics on matches, differences, and explanations. This is a summary on the table fulloutput. - Inadequately Explained Items :: indicates specific objects that were inadequately explained by the EXPLANATIONS data

**** File Output The following files (based on contents of these environment variables) are created and populated via COPY: - FULLOUTPUT :: populated from table fulloutput - This contains a full list of all objects examined in both databases, complete with objects, respective definitions, and difference type and cause - UNEXPLAINED :: populated from table unexplained_items - This lists all objects where there was some difference, but no item found to explain the difference. - BADEXPLAIN :: populated from table badexplanations_items - This lists all objects where there was some difference, and an explanation, but the explanation did not properly explain the difference. For instance, an object was missing from the second database, but the explanation indicated that there should have been a different definition (which indicates that the object was expected to be found in both databases). **** Return Codes

  If errors are encountered, the script ~pgcmp.sh~
  will terminate with varying exit codes:

  - exit 1 :: If data could not be extracted from the either of the source databases
  - exit 1 :: If a connection is not established with the comparison database
  - exit 2 :: If the comparison script does not run successfully
  - exit 3 :: If not all object differences were adequately explained
  - exit 0 :: If all runs to completion, and differences /were/
              adequately explained
	  
  Thus, generally explaining this:
  0. Comparison ran successfully, found no troublesome differences
  1. Database connectivity problems
  2. Error in processing comparison
  3. Comparison ran, and found irreconcilable differences

  These return codes should be useful if running scripts to do
  automated schema analyses.

** Installation Requirements

pgcmp is implemented as a shell script (expects to be run using /bin/bash), and thus requires:

  • Bash :: Available as /bin/bash
  • psql :: Available from a Postgres installation

If building an RPM file, this documentation, in README.org, is transformed to HTML using a Ruby script that uses a Ruby "gem" called org-ruby.

There are several ways the org-ruby gem may be installed:

  • Manual Installation of Ruby Gem :: This Gem may be installed via the command gem install org-ruby. On many systems, this would need to be run by the root user, perhaps using sudo.

  • Packaged Installation :: On Debian (and perhaps derivatives such as Ubuntu), this gem is contained by the package ruby-org, and hence may be installed via apt-get install ruby-org.