Jim Schueler > BerkeleyDB-Locks-0_4 > BerkeleyDB::Locks

Download:
BerkeleyDB-Locks-0_4.tar.gz

Dependencies

Annotate this POD

View/Report Bugs
Module Version: 0.04   Source  

NAME ^

BerkeleyDB::Locks - Perl extension for Berkeley DB 4.1

SYNOPSIS ^

  use BerkeleyDB ;
  use BerkeleyDB::Locks;

  $env = new BerkeleyDB::Env ... ;
  $watch = $env->lockmonitor ;

  while (1) {
        my @released = $watch->monitor ;
        warn "released: " join ' ', @released ;
        sleep (10) ;
        }

DESCRIPTION ^

To demystify the Berkeley locking subsystem, realize that "a lock" in Berkeley-ese refers to a verb, not a noun. A lock is an event that links the locker (a process identifier) and a lock-object which represents a database page or record.

A complete list of lock states follows. An active lock is generally either WAITING or HELD. A FREE lock is not assigned to either a locker or lock object. The other states represent error conditions or are transient.

        ABORTED
        ERROR
        EXPIRED
        FREE
        HELD
        NOTEXIST
        PENDING
        WAITING

In normal operation, a locker operates on a database object without interference. The locker applies a hold lock to the object, performs its operation, and then frees the lock.

Heavy or critical operations on an object are coordinated using the locking subsystem. If an object is unavailable, the locker's request is queued with a wait lock. Each waiting lock request is served in turn, and subsequently upgraded to a hold lock. When the locker operation is complete, the lock is freed.

A lock-up occurs when a hold lock is never released and the queued locks wait indefinitely for their turn.

The monitor() function traverses the table of lockers, locates all of the active locks associated with each locker, and maintains a complete list of waiting conditions.

A wait condition is defined by a waiting lock and a holding lock attached to the same lock object. These transient Lock events are defined by two properties: The lock offset defines the structure location; and the gen value which is incremented for each event.

A wait condition that persists between consecutive calls to monitor() is an indication of a lock-up. monitor() releases the holding locks on the object under contention. monitor() returns the identities of these holding locks and their properties (before release) as name value pairs.

The holding locks are identified by their offset location. Their properties are hashes containing the following values:

        mode:   Creation mode
        status: Current status of lock (listed above)
        gen:    Increments for each lock activation 
        locker: Integer ID of the locker
        obj:    Offset ID of the object

The poll() function also identifies the objects under waiting locks. This function returns the object holders and holder properties similar to the monitor() function. poll() does not release the held locks. Also, poll() reports the holders for all wait conditions. monitor() only reports holders of persistent wait conditions.

CAVEATS ^

This lock monitoring tool does not guarantee "fitness of use". This API is the result of fiddling under the Berkeley hood. These capabilities have not been made public by the Berkeley developers.

The Berkeley locking subsystem is perfectly stable. Every locking fault I've encountered was caused by an implementation bug. Nevertheless, I believe this module can be a very helpful debugging tool- the reporting features are intended to be used. One indispensible enhancement is a function that allows a process to look up its own assigned locker IDs. Thus, a process can determine when it has caused a lock-up and abort its operation before releasing its locks.

I cannot give any details about the consequences of prematurely releasing a lock. My lock-ups are generally caused because a thread dies before releasing its held locks. But since the purpose of locking is to ensure data integrity, assume that careless use of this lock monitor may result in data corruption. For that reason, this module is best suited in production environments where continuous operation has greater priority than data integrity; for example, where the Berkeley system operates on a data copy. Otherwise, use of this lock monitor should be restricted to the bench.

This application has only been tested on Berkeley DB 4.1. The lock architecture changed significantly between Berkeley DB 3 and Berkeley DB 4. Support for Berkeley DB 3 can be considered an enhancement.

EXPORT

None by default.

AUTHOR ^

Jim Schueler, <jschueler@tqis.com<gt>

SEE ALSO ^

perl.

syntax highlighting: