Trace file locks

In order to detect locking problems in multi-user applications, file locks may be traced. In order to trace file locks, the option

• TraceLocks=YES

has to be set to true ( YES ). File locks are written to the directory referenced in the TRACE option;

%TRACE%/ userName _ processID .lck

Since lock trace files may become very large, trace information is stored as binary data with the following structure:

struct _LockTrace {

DateTime timestamp; // lock/unlock time

int64 position; // file position

int32 length; // lock length

char operation; // +: lock, -: unlock

};

In order to keep lock files small, the lock file directory should refer to users home directory.

In order to analyze lock data, a lock trace database may be created by loading single files of all lock trace files ( *.lck ) in a directory (see example below). In order to run the analysis, naming conventions as described above, are required. The procedure creates an import database with two extents:

ObjectLocks: contains locks for each process

int32 pid; // process id

STRING(20) user; // file name without extension

SET<ObjectLock> locks; // locks for this process

ObjectLock contains all locks/unlocks

int32 pid; // process id

STRING(20) user; // file name without extension

DateTime timestamp; // last lock/unlock time for position

int64 position; // file position

int32 length; // lock length

int32 cnt; // number of currently active locks

When importing lock trace files, a lock summary record is created for pid , position and length . The cnt field contains the number of currently active locks for the file lock in the current process.

In order to select locks for a certain time interval and a specific file position, the ObjectLock ::List function may be called, which creates a result file in the TRACE directory (directory defined in the TRACE option).