koichi-szk / pg_orphaned_file

Find suspicious orphaned file in PostgreSQL data directory

This command, pg_find_orphaned_file compared all the database filenode and actual files in the database directory and finds suspicious orphaned files in the data directory.

The algorithm of this inspection is straight forward. Take a look at the command itself, pg_find_orphaned_file. This is actually bash script and there are nothing special in the command.

Command option

-h host specifies host name of the target database,

-p port specifies port number to conect to the target database,

-U username or -u username specifies database user name to connet to the database. This user must have privlledge to readd all pg_class tuples for all the database except for template0.

-d database Specifies database name to login to get list of databases. Please note that to obtain information of each database, this toll will use such dagtabase name to login. Database user as specified by -U or -u option must have proper priviledes to read pg_class catalog of each database.

If any of these options are omitted, default value will be used. If a password is associated with the user, you will be prompted for the password each time this tool connects to the database. You can bypass this by configuring .pgpass file as described in psql documentation of PosgtgreSQL.

Output from the command

Output from the command is lines of suspicious orphaned file found in $PGDATA directory. Each line contains single file path relative from $PGDATA directory.

Before removing these files, pleases make sure that you absolutely do not need these files. It will be a good idea to back up such files in other places to see this is not needed by the database.

If no such suspicious orphaned file is not detected, the command will write this message.

More to improve

This toold does not detect the file with valid name such as:

• [1-9][0-9]*.[1-9][0-9]* where the former part [1-9][0-9]* is valid relfilenode but the file itself is not actually a part of relfile.
• [1-9][0-9]*_fsm where the former part [1-9][0-9]* is valid relfilenode but the relation does not have free space map,
• [1-9][0-9]*_vm where the former part [1-9][0-9]* is valid relfilenode but the relation does not have visibiility map,

Matching is done using each pair of the file list and relfilenode output list, therefore the algorithm is $O(N^2)$. If databaes has many objects (huge tables and huge idnexes) it may take very long to find suspicious orphaned file. To improve, we need hash or other index.