MWFileRC

class MWFileRC : public MWRMComm

A Resource Management and Communication class that uses Condor for underlying support of resource managament

Inheritance:

Public Methods
void CheckLogFilesRunning( ): Always get the last result
int handle_finished_worker( int i ): Handle a message from the worker
NUM_FILE_TYPES MWFileRC( bool val, int id ): Constructor
~MWFileRC(): Destructor

Public

A. Resource Management Routines

int setup( int argc, char *argv[], int *mytid, int *mastertid ): Initialises
void exit( int exitval ): Shutdown
int init_beginning_workers( int *nworkers, MWWorkerID ***workers ): Initialize workers if already some have started up
int start_worker( MWWorkerID *w ): This function is actually a misonomer
int removeWorker( MWWorkerID *w ): This function removes a existing worker
int hostaddlogic( int *w ): Figure out whether or not to generate a new worker depending on whether new requests have been made
int config( int *, int *, MWWorkerID *** ): A dummy function
int read_RMstate( FILE *fp = NULL ): A routine for reading in the MW-File state at the time of checkpointing
int write_RMstate( FILE *fp = NULL ): A routine for writing in the MW-File state at the time of checkpointing

B. Communication Routines

int initsend( int useless = 0 ): Initialize the send buffer
int send( int toWhom, int msgtag ): Send function
int recv( int fromWhom, int msgtag ): Recv function
int bufinfo( int buf_id, int *len, int *tag, int *sending_host ): Get some info about the recv buffer
int pack( char *bytes, int nitem, int stride = 1 ): pack some bytes
int pack( float *f, int nitem, int stride = 1 ): float
int pack( double *d, int nitem, int stride = 1 ): double
int pack( int *i, int nitem, int stride = 1 ): int
int pack( unsigned int *ui, int nitem, int stride = 1 ): unsigned int
int pack( short *sh, int nitem, int stride = 1 ): short
int pack( unsigned short *ush, int nitem, int stride = 1 ): unsigned short
int pack( long *l, int nitem, int stride = 1 ): long
int pack( unsigned long *ul, int nitem, int stride = 1 ): unsigned long
int pack( char *str ): string
int unpack( char *bytes, int nitem, int stride = 1 ): Unpack some bytes
int unpack( float *f, int nitem, int stride = 1 ): float
int unpack( double *d, int nitem, int stride = 1 ): double
int unpack( int *i, int nitem, int stride = 1 ): int
int unpack( unsigned int *ui, int nitem, int stride = 1 ): unsigned int
int unpack( short *sh, int nitem, int stride = 1 ): short
int unpack( unsigned short *ush, int nitem, int stride = 1 ): unsigned short
int unpack( long *l, int nitem, int stride = 1 ): long
int unpack( unsigned long *ul, int nitem, int stride = 1 ): unsigned long
int unpack( char *str ): string

Private Fields
struct FileWorker *fileWorkers: An array of the number of workers
int actual_task: The actual task that is being executed
int CHECKLOG_FREQ: This determines with what frequency should we check the log files
long long checksum: The checksum calculated
char control_directory[256]: The control directory the master reads for all resource management functions
int current_num_workers: The number of workers that were present
int cyclePosition: This is a variable that keeps the cycle in effect
int expected_number: The expected number of the next message
int FileRCID: The value is the id that the worker gets
int* hostadd_reqs: A variable array keeping track of how many have been requested
int** hostaddind_reqs: A variable to keep track of which of each have been requested
char init_file[256]: The init file
char input_directory[256]: The directory where the master has to send the work
bool isMaster: The bool indicates the mode of the RC instance a true value means that it is a master and a false means that it is a worker
int master_expected_number: The expected_number of the next message that the master is expecting
int MasterUp: Internal variable that is init to what the worker has to send to the upper layer in case of a master wakeup
int max_num_workers: An integer corresponding to the number of fileWorkers struct that we had/have as the maximum;
char moment_worker_file[256]: The file in the control directory that will contain the momentary number of workers
int msgTag: The tag of the message that just came in
char output_directory[256]: The directory where all the workers have to send their output
List* recvList: A list of all the items that are received
List* sendList: A list of all the items that are sent in a send after a series of packs
int subId: A variable to keep track of submit files
int submitted_num_workers: The number of submitted workers
int turnNo: This keeps the track of how many cycles were made
int whomRecv: The message came from whom
long worker_timeout: The worker timeout in minutes
int* workerArch: A variable array helping in choosing the next arch to chose from
FileRCEvent* workerEvents: An array of message tags to be sent to the upper layers as to what tag has to be sent with the event associated with this worker

Private Methods
void CheckLogFilesResuscicate( ): Is Called when we are recovering from a crash
int ChoseArchNum(): Get a architecture number
int do_spawn( int numworkers, int arch ): Create a worker
void GetCondorId( char *lgfile, int *cId, int *pId ): Find the condor_ID of the worker with this log file
int GetCounter( char *file1, char *file2 ): Another helping routine
int GetMasterExpectedNumber( char *file ): Another helping routine
int GetWorkerCounter( char *file ): The helping routine for resuscicate
int handle_executing_worker( int i ): Called when the worker starts executing first
int handle_killed_worker( int i ): Is Called when a task is dead
int handle_master_executing(): Is called when the worker detects that the master has come up
int handle_resumed_worker( int i ): Is Called when a host is resumed
int handle_suspended_worker( int i ): Is Called when a host is suspended
int handle_transited_worker( int i ): Called in no_checkpoint mode
int handle_work( int msgtag ): Is Called when a message is received by the worker
void inform_target_num_workers( ): Information by the upper layer
void InitStructures( ): A function that inits some internal structures
bool IsComplete( int i ): Determine whether worker i has sent a message
void killWorker( int i ): Kill a worker
int master_recv( int fromWhom, int msgtag ): Master receive
void resuscicate( ): The main resuscicate function
int setup_notifies( int worker_id ): Notify various events
int worker_recv( int fromWhom, int msgtag ): Worker receive

Private
: Polls the log files to see some changes
: Some private functions

Documentation

A Resource Management and Communication class that uses Condor for underlying support of resource managament. Some crude inter-process communication is provided using the userlog feature of Condor. resource management.

NUM_FILE_TYPES MWFileRC( bool val, int id )

Constructor

~MWFileRC()

Destructor

A. Resource Management Routines

Here we implement the pure virtual functions found in ur parent class, MWRMComm.

int setup( int argc, char *argv[], int *mytid, int *mastertid )

Initialises. Depending on whether it is master or worker instance it initializes all the internal variables.

void exit( int exitval )

Shutdown. Kills all the workers if it is master

int init_beginning_workers( int *nworkers, MWWorkerID ***workers )

Initialize workers if already some have started up

int start_worker( MWWorkerID *w )

This function is actually a misonomer. It DOES NOT spawn a new worker. Rather it just inits the structure that is passed on to it

int removeWorker( MWWorkerID *w )

This function removes a existing worker

int hostaddlogic( int *w )

Figure out whether or not to generate a new worker depending on whether new requests have been made

int config( int *, int *, MWWorkerID *** )

A dummy function. For the sake of conformity with the Pvm

int read_RMstate( FILE *fp = NULL )

A routine for reading in the MW-File state at the time of checkpointing

int write_RMstate( FILE *fp = NULL )

A routine for writing in the MW-File state at the time of checkpointing

B. Communication Routines

Unlike MWPvmRC, the communication routines are non-trivial because Condor provides no inter-process comminucation. Thus we use files for communication. So a send is essentially a file write operation and a recv is a file-read operation. We maintain 2 lists:- The sendList and recvList for taking care of what is to be written/read to/from the files. As in pvm a user beings by calling initsend which creates a new list. Calls to pack insert into the list what is being packed. And finally a send writes the entire thing into a file identified by the destination. Corresponding things happpen in recv.

int initsend( int useless = 0 )

Initialize the send buffer

int send( int toWhom, int msgtag )

Send function

int recv( int fromWhom, int msgtag )

Recv function

int bufinfo( int buf_id, int *len, int *tag, int *sending_host )

Get some info about the recv buffer

int pack( char *bytes, int nitem, int stride = 1 )

pack some bytes

int pack( float *f, int nitem, int stride = 1 )

float

int pack( double *d, int nitem, int stride = 1 )

double

int pack( int *i, int nitem, int stride = 1 )

int

int pack( unsigned int *ui, int nitem, int stride = 1 )

unsigned int

int pack( short *sh, int nitem, int stride = 1 )

short

int pack( unsigned short *ush, int nitem, int stride = 1 )

unsigned short

int pack( long *l, int nitem, int stride = 1 )

long

int pack( unsigned long *ul, int nitem, int stride = 1 )

unsigned long

int pack( char *str )

string

int unpack( char *bytes, int nitem, int stride = 1 )

Unpack some bytes

int unpack( float *f, int nitem, int stride = 1 )

float

int unpack( double *d, int nitem, int stride = 1 )

double

int unpack( int *i, int nitem, int stride = 1 )

int

int unpack( unsigned int *ui, int nitem, int stride = 1 )

unsigned int

int unpack( short *sh, int nitem, int stride = 1 )

short

int unpack( unsigned short *ush, int nitem, int stride = 1 )

unsigned short

int unpack( long *l, int nitem, int stride = 1 )

long

int unpack( unsigned long *ul, int nitem, int stride = 1 )

unsigned long

int unpack( char *str )

string

Some private functions

int handle_finished_worker( int i )

Handle a message from the worker

int handle_resumed_worker( int i )

Is Called when a host is resumed

int handle_suspended_worker( int i )

Is Called when a host is suspended

int handle_killed_worker( int i )

Is Called when a task is dead

int handle_executing_worker( int i )

Called when the worker starts executing first

int handle_transited_worker( int i )

Called in no_checkpoint mode

int handle_master_executing()

Is called when the worker detects that the master has come up

int handle_work( int msgtag )

Is Called when a message is received by the worker

Polls the log files to see some changes

void CheckLogFilesRunning( )

Always get the last result

void CheckLogFilesResuscicate( )

Is Called when we are recovering from a crash

void resuscicate( )

The main resuscicate function

int GetWorkerCounter( char *file )

The helping routine for resuscicate

int GetCounter( char *file1, char *file2 )

Another helping routine

int GetMasterExpectedNumber( char *file )

Another helping routine

void GetCondorId( char *lgfile, int *cId, int *pId )

Find the condor_ID of the worker with this log file

int do_spawn( int numworkers, int arch )

Create a worker

void killWorker( int i )

Kill a worker

int setup_notifies( int worker_id )

Notify various events

int master_recv( int fromWhom, int msgtag )

Master receive

int worker_recv( int fromWhom, int msgtag )

Worker receive

bool IsComplete( int i )

Determine whether worker i has sent a message

int ChoseArchNum()

Get a architecture number

void inform_target_num_workers( )

Information by the upper layer

void InitStructures( )

A function that inits some internal structures

bool isMaster

The bool indicates the mode of the RC instance a true value means that it is a master and a false means that it is a worker. We need to make it a tristate as somtimes an RC can be both a master and a worker.

int FileRCID