gprecoverseg

Recovers a primary or mirror segment instance that has been marked as down (if mirroring is enabled).

Synopsis

gprecoverseg [-p new_recover_host[,...]] | -i recover_config_file] [-d master_data_directory] 
             [-B parallel_processes] [-F [-s]] [-a] [-q] 
             [--no-progress] [-l logfile_directory]
gprecoverseg -r 
gprecoverseg -o output_recover_config_file 
             [-p new_recover_host[,...]]
gprecoverseg -? | --help
gprecoverseg --version

Description

In a system with mirrors enabled, the gprecoverseg utility reactivates a failed segment instance and identifies the changed database files that require resynchronization. Once gprecoverseg completes this process, the system goes into resyncronizing mode until the recovered segment is brought up to date. The system is online and fully operational during resyncronization.

During an incremental recovery (the -F option is not specified), if gprecoverseg detects a segment instance with mirroring disabled in a system with mirrors enabled, the utility reports that mirroring is disabled for the segment, does not attempt to recover that segment instance, and continues the recovery process.

A segment instance can fail for several reasons, such as a host failure, network failure, or disk failure. When a segment instance fails, its status is marked as down in the Greenplum Database system catalog, and its mirror is activated in change tracking mode. In order to bring the failed segment instance back into operation again, you must first correct the problem that made it fail in the first place, and then recover the segment instance in Greenplum Database using gprecoverseg.

Segment recovery using gprecoverseg requires that you have an active mirror to recover from. For systems that do not have mirroring enabled, or in the event of a double fault (a primary and mirror pair both down at the same time) — you must take manual steps to recover the failed segment instances and then perform a system restart to bring the segments back online. For example, this command restarts a system.

gpstop -r

By default, a failed segment is recovered in place, meaning that the system brings the segment back online on the same host and data directory location on which it was originally configured. In this case, use the following format for the recovery configuration file (using -i).

<failed_host_address>|<port>|<data_directory>

In some cases, this may not be possible (for example, if a host was physically damaged and cannot be recovered). In this situation, gprecoverseg allows you to recover failed segments to a completely new host (using -p), on an alternative data directory location on your remaining live segment hosts (using -s), or by supplying a recovery configuration file (using -i) in the following format. The word SPACE indicates the location of a required space. Do not add additional spaces.

<failed_host_address>|<port>|<data_directory>SPACE
<recovery_host_address>|<port>|<data_directory>

See the -i option below for details and examples of a recovery configuration file.

The gp_segment_configuration system catalog table can help you determine your current segment configuration so that you can plan your mirror recovery configuration. For example, run the following query:

=# SELECT dbid, content, address, port, datadir 
   FROM gp_segment_configuration
   ORDER BY dbid;

The new recovery segment host must be pre-installed with the Greenplum Database software and configured exactly the same as the existing segment hosts. A spare data directory location must exist on all currently configured segment hosts and have enough disk space to accommodate the failed segments.

The recovery process marks the segment as up again in the Greenplum Database system catalog, and then initiates the resyncronization process to bring the transactional state of the segment up-to-date with the latest changes. The system is online and available during resyncronization. To check the status of the resyncronization process run:

gpstate -m

Options

-a (do not prompt)

Do not prompt the user for confirmation.

-B parallel_processes

The number of segments to recover in parallel. If not specified, the utility will start up to 16 parallel processes depending on how many segment instances it needs to recover.

-d master_data_directory

Optional. The master host data directory. If not specified, the value set for $MASTER_DATA_DIRECTORY will be used.

-F (full recovery)

Optional. Perform a full copy of the active segment instance in order to recover the failed segment. The default is to only copy over the incremental changes that occurred while the segment was down.

Warning: A full recovery deletes the data directory of the down segment instance before copying the data from the active (current primary) segment instance. Before performing a full recovery, ensure that the segment failure did not cause data corruption and that any host segment disk issues have been fixed.

Also, for a full recovery, the utility does not restore custom files that are stored in the segment instance data directory even if the custom files are also in the active segment instance. You must restore the custom files manually. For example, when using the gpfdists protocol (gpfdist with SSL encryption) to manage external data, client certificate files are required in the segment instance $PGDATA/gpfdists directory. These files are not restored. For information about configuring gpfdists, see Encrypting gpfdist Connections.

Full recovery can take a long time for large databases, so gprecoverseg displays a running estimate of the completion progress of the copy for each segment. Progress for each segment is updated once per second, using ANSI escape codes to update the line for each segment in-place. If you are redirecting the gprecoverseg output to a file, or if the ANSI escape codes do not work correctly on your terminal, you can include the -s option on the command line to omit the ANSI escape codes. This outputs a new line once per second for each segment. Include the --no-progress option to completely disable the progress reports.

-i recover_config_file

Specifies the name of a file with the details about failed segments to recover. Each line in the file is in the following format. The word SPACE indicates the location of a required space. Do not add additional spaces.

<failed_host_address>|<port>|<data_directory>SPACE 
<recovery_host_address>|<port>|<data_directory>

Comments

Lines beginning with # are treated as comments and ignored.

Segments to Recover

Each line after the first specifies a segment to recover. This line can have one of two formats. In the event of in-place recovery, enter one group of colon delimited fields in the line. For example:

failedAddress|failedPort|failedDataDirectory

For recovery to a new location, enter two groups of fields separated by a space in the line. The required space is indicated by SPACE. Do not add additional spaces.

failedAddress|failedPort|failedDataDirectorySPACEnewAddress|
newPort|newDataDirectory

Examples

In-place recovery of a single mirror

sdw1-1|50001|/data1/mirror/gpseg16

Recovery of a single mirror to a new host

sdw1-1|50001|/data1/mirror/gpseg16SPACEsdw4-1|50001|/data1/recover1/gpseg16

Obtaining a Sample File

You can use the -o option to output a sample recovery configuration file to use as a starting point.

-l logfile_directory

The directory to write the log file. Defaults to ~/gpAdminLogs.

-o output_recover_config_file

Specifies a file name and location to output a sample recovery configuration file. The output file lists the currently invalid segments and their default recovery location in the format that is required by the -i option. Use together with the -p option to output a sample file for recovering on a different host. This file can be edited to supply alternate recovery locations if needed.

-p new_recover_host[,…]

Specifies a spare host outside of the currently configured Greenplum Database array on which to recover invalid segments. In the case of multiple failed segment hosts, you can specify a comma-separated list. The spare host must have the Greenplum Database software installed and configured, and have the same hardware and OS configuration as the current segment hosts (same OS version, locales, gpadmin user account, data directory locations created, ssh keys exchanged, number of network interfaces, network interface naming convention, and so on.).

-q (no screen output)

Run in quiet mode. Command output is not displayed on the screen, but is still written to the log file.

-r (rebalance segments)

After a segment recovery, segment instances may not be returned to the preferred role that they were given at system initialization time. This can leave the system in a potentially unbalanced state, as some segment hosts may have more active segments than is optimal for top system performance. This option rebalances primary and mirror segments by returning them to their preferred roles. All segments must be valid and synchronized before running gprecoverseg -r. If there are any in progress queries, they will be cancelled and rolled back.

-s

Show pg_basebackup progress sequentially instead of in-place. Useful when writing to a file, or if a tty does not support escape sequences. The default is to show progress in-place.

--no-progress

Suppresses progress reports from the pg_basebackup utility. The default is to display progress of base backups.

-v (verbose)

Sets logging output to verbose.

--version (version)

Displays the version of this utility.

-? (help)

Displays the online help.

Examples

Recover any failed segment instances in place:

$ gprecoverseg

Rebalance your Greenplum Database system after a recovery by resetting all segments to their preferred role. First check that all segments are up and synchronized.

$ gpstate -m
$ gprecoverseg -r

Recover any failed segment instances to a newly configured spare segment host:

$ gprecoverseg -i recover_config_file

Output the default recovery configuration file:

$ gprecoverseg -o /home/gpadmin/recover_config_file

See Also

gpstart, gpstop

Need help? Visit the Greenplum Database Community