Skip to content
Percona Software for MySQL Documentation
Rate this page
☆ ☆ ☆ ☆ ☆
Thanks for your feedback
Thank you! The feedback has been submitted.

Get free database assistance or contact our experts for personalized support.

Get help from Percona

Index of files created by Percona XtraDB Cluster

Percona XtraDB Cluster can create the following files:

GRA_*.log

GRA_*.log files contain row-based binary log events for a transaction that failed to apply.

The replica thread does not apply that transaction.

Each file has a matching warning or error in the MySQL error log.

Some errors are harmless. One example is a Data Definition Language (DDL) statement that drops a table that does not exist.

Check these logs to determine what failed.

To decode a GRA log file, add a binary log header. Use an instance with binlog_checksum set to NONE. Extract the first 123 bytes from a binary log file.

The following commands create GRA_HEADER, combine the files, and decode the result:

head -c 123 mysqld-bin.000001 > GRA_HEADER
cat GRA_HEADER > /var/lib/mysql/GRA_1_2-bin.log
cat /var/lib/mysql/GRA_1_2.log >> /var/lib/mysql/GRA_1_2-bin.log
mysqlbinlog -vvv /var/lib/mysql/GRA_1_2-bin.log
mysqlbinlog output 
/*!50530 SET @@SESSION.PSEUDO_SLAVE_MODE=1*/;
/*!50003 SET @OLD_COMPLETION_TYPE=@@COMPLETION_TYPE,COMPLETION_TYPE=0*/;
DELIMITER /*!*/;
# at 4
#160809  16:04:05 server id 3  end_log_pos 123     Start: binlog v 4, server v 8.0-log created 160809 16:04:05 at startup
# Warning: this binlog is either in use or was not closed properly.
ROLLBACK/*!*/;
BINLOG '
nbGpVw8DAAAAdwAAAHsAAAABAAQANS43LjEyLTVyYzEtbG9nAAAAAAAAAAAAAAAAAAAAAAAAAAAA
AAAAAAAAAAAAAAAAAACdsalXEzgNAAgAEgAEBAQEEgAAXwAEGggAAAAICAgCAAAACgoKKioAEjQA
ALfQ8hw=
'/*!*/;
# at 123
#160809  16:05:49 server id 2  end_log_pos 75     Query    thread_id=11    exec_time=0    error_code=0
use `test`/*!*/;
SET TIMESTAMP=1470738949/*!*/;
SET @@session.pseudo_thread_id=11/*!*/;
SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=0, @@session.unique_checks=1, @@session.autocommit=1/*!*/;
SET @@session.sql_mode=1436549152/*!*/;
SET @@session.auto_increment_increment=1, @@session.auto_increment_offset=1/*!*/;
/*!\C utf8 *//*!*/;
SET @@session.character_set_client=33,@@session.collation_connection=33,@@session.collation_server=8/*!*/;
SET @@session.lc_time_names=0/*!*/;
SET @@session.collation_database=DEFAULT/*!*/;
drop table t
/*!*/;
SET @@SESSION.GTID_NEXT= 'AUTOMATIC' /* added by mysqlbinlog */ /*!*/;
DELIMITER ;
# End of log file
/*!50003 SET COMPLETION_TYPE=@OLD_COMPLETION_TYPE*/;
/*!50530 SET @@SESSION.PSEUDO_SLAVE_MODE=0*/;

Use the decoded statement and timestamp to find the matching error in the MySQL error log.

Error message 
160805  9:33:37 8:52:21 [ERROR] Slave SQL: Error 'Unknown table 'test'' on query. Default database: 'test'. Query: 'drop table test', Error_code: 1051
160805  9:33:37 8:52:21 [Warning] WSREP: RBR event 1 Query apply warning: 1, 3

In this example, a DROP TABLE statement ran against a table that does not exist.

gcache.page

See gcache.page_size.

See also

Percona Database Performance Blog: All You Need to Know About GCache (Galera-Cache)

galera.cache

galera.cache is the main store for write sets.

The file is a permanent ring buffer. The node allocates the file on disk at initialization.

The gcache.size variable sets the file size. A larger file caches more write sets. A joining node is more likely to receive Incremental State Transfer (IST) instead of State Snapshot Transfer (SST).

The gcache.name variable sets the file name.

grastate.dat

grastate.dat stores Galera state information.

The file has the following fields:

  • version: Format version of grastate.dat

  • uuid: Unique identifier for the state and the change sequence. See UUID.

  • seqno: Sequence number of the last change. The value is a 64-bit signed integer.

  • cert_index: Unused. Certification index restore is not implemented.

seqno uses the following values:

  • 0: No write sets were generated or applied on that node for the life of the file

  • -1: The server is running, or the last shutdown was not clean

  • A value greater than 0: The last shutdown was clean

While the server runs, Galera writes -1 to grastate.dat. A clean shutdown writes the correct seqno value.

At the next start, -1 means the previous shutdown was not clean. A value greater than 0 means the previous shutdown was clean. The server then writes -1 again. The next shutdown uses the same detection.

The following examples show grastate.dat in three situations.

The node is not running. This state means the node crashed during a transaction:

GALERA saved state
version: 2.1
uuid:    1917033b-7081-11e2-0800-707f5d3b106b
seqno:   -1
cert_index:

The node is not running. This state means the node shut down cleanly:

GALERA saved state
version: 2.1
uuid:    1917033b-7081-11e2-0800-707f5d3b106b
seqno:   5192193423942
cert_index:

The node is not running. This state means the node crashed during a DDL statement:

GALERA saved state
version: 2.1
uuid:    00000000-0000-0000-0000-000000000000
seqno:   -1
cert_index:

The same zero UUID and seqno of -1 also appear after the cluster votes the node out for inconsistent data.

Galera marks the state as corrupt. You can inspect the node while the node still runs.

A corrupt grastate.dat file does not force a full SST at the next start. Restart recovery reads a usable position from InnoDB. The node can rejoin with IST. The local data stays inconsistent.

To force SST after the cluster votes the node out for inconsistent data, use one of the following methods:

If the option is yes, the node deletes the file after Galera marks the state as corrupt. Copy the file first if you still need the file.

gvwstate.dat

The Primary Component is the subset of nodes that holds quorum. Only that subset accepts writes.

Primary Component recovery restores that subset after a full cluster outage. Galera uses gvwstate.dat for this process when pc.recovery is true (the default). After every node from the last saved Primary Component can communicate again, the cluster restores the Primary Component. A manual bootstrap is not required.

A planned full-cluster restart still requires an explicit bootstrap.

Galera creates or updates gvwstate.dat when the Primary Component forms or changes. The file records the latest Primary Component for the node. A clean shutdown deletes the file.

The first part stores the UUID of the node. The second part stores the view. The view is written between #vwbeg and #vwend.

The view record has the following fields:

  • view_id: [view_type] [view_uuid] [view_seq]. view_type is always 3. That value means a primary view. view_uuid and view_seq identify the view.

  • bootstrap: 0 or 1. This value does not change Primary Component recovery.

  • member: UUID and segment of each node in the Primary Component.

Example gvwstate.dat file 
my_uuid: c5d5d990-30ee-11e4-aab1-46d0ed84b408
#vwbeg
view_id: 3 bc85bd53-31ac-11e4-9895-1f2ce13f2542 2 
bootstrap: 0
member: bc85bd53-31ac-11e4-9895-1f2ce13f2542 0
member: c5d5d990-30ee-11e4-aab1-46d0ed84b408 0
#vwend