YggLogControl:
CEDAR DEFINITIONS =
BEGIN
VolumeID: TYPE = YggEnvironment.VolumeID;
FileStore: TYPE = YggEnvironment.FileStore;
TransID: TYPE = YggEnvironment.TransID;
PageCount: TYPE = YggEnvironment.PageCount;
RecordID: TYPE = YggLog.RecordID;
RecordType: TYPE = YggLog.RecordType;
This interface contains procedures for initialization of the log subsystem.
One aspect of initialization is registering procedures to take part in the analysis
pass of recovery and in periodic log checkpoints. The transaction and backup
subsystems are expected to register procedures for this purpose.
All calls to RegisterAnalysisProc and RegisterCheckpointProc must complete before
the analysis pass of recovery begins. To ensure this, a client can be called by
InitProcs.CalledBeforeAnalysisPass, or can call from a PROGRAM proc (called by
STARTing the module).
Format:
PROC [logVolume: VolumeID, logFile, restartFile: YggEnvironment.DID, myFileStore: FileStore];
This call formats the given files to a well-defined initial state that is
acceptable to recovery. This should not be called while the server is running.
Call Format, then call Recover, in order to cold-start a server.
CAUTION: The previous contents of logFile and restartFile are lost!
Recover:
PROC [logVolume: VolumeID, logFile, restartFile: YggEnvironment.DID];
This call performs a warm-start of the log subsystem using the given log.
As a side effect, many procedures registered through the YggLog and YggLogControl
interfaces are called, as well as the procedures defined in InitProcs.
AnalysisProc:
TYPE =
PROC [record: RecordID, type: RecordType, trans: TransID];
Called from Recover during analysis pass over log.
This proc can read record using YggLog.ReadForRecovery (to skip over transaction
portion of log), or LogBasic.GetCurrent (to read entire log record).
RegisterAnalysisProc: PROC [recordType: RecordType, analysisProc: AnalysisProc];
ForkCheckpointProcess:
PROC [myFileStore: FileStore, wakeupPeriodInSeconds:
NAT];
Forks a process that performs periodic log checkpoints. Should not be called
until Recover has returned.
This process is an infinite loop that sleeps for wakeupPeriodInSeconds seconds,
then does the following:
1) calls each subsystem that writes log records. At present the three subsystems
that write log records are the worker, coordinator, and backup. Each subsystem
declares how much log it depends upon, and how much log it must examine at
restart. If a subsystem notices that it is using too much log, it takes action
(probably in another process) to reduce its use. The worker subsystem forces all
dirty FilePageMgr pages to disk and writes workerComplete log records for
transactions whose effects are fully done (or undone).
2) writes a checkpointComplete log record if some value in this record is different
than in the previous checkpointComplete record.
3) initiates an incremental garbage collection in order to keep zones compact.
(it makes sense to do this in the checkpoint process since step 1 often frees storage).
CheckpointProc:
TYPE =
PROC []
RETURNS [keepRecord, startAnalysisRecord: RecordID];
nextRecord is the RecordID of the next log record to be written. This parm
should be used by the CheckpointProc to control its use of the finite log
resource (keeping the log from filling up and keeping the cost of restart down
to a reasonable value).
keepRecord is oldest record in online log required by this component.
startAnalysisRecord is oldest record in online log that must be examined during
the analysis pass of restart.
Making keepRecord and startAnalysisRecord distinct allows the backup system to
require much more online log than other subsystems, yet allow restart to be fast.
RegisterCheckpointProc: PROC [checkpointProc: CheckpointProc];
WordsUsableByClient:
PROC []
RETURNS [words:
INT];
Client should not allow its use of log (RecordID of next AlpineLog.Write - keepRecord)
to exceed this bound.
RecordIDOfNextWrite:
PROC []
RETURNS [RecordID];
RecordID that AlpineLog.Write would have returned as thisRecord, had it been called.
There is no way to guarantee that calling AlpineLog.Write after calling this proc will
generate this RecordID, however.
WriteNoopRecords:
PROC [nPages: PageCount]
RETURNS [followingRecord: RecordID];
For use by a test program that wishes to generate a load on the log.
END.