Changeset - 6860ba091be1
[Not reviewed]
Merge default
1 2 1
ethanzonca - 16 years ago 2009-10-10 16:35:46

Merge
3 files changed with 138 insertions and 22 deletions:
0 comments (0 inline, 0 general)
etc/Makefile.am
Show inline comments
 
sysconf_DATA = distrenclient.conf distrencommon.conf distrenserver.conf distrenslave.conf
 
sysconf_DATA = distrenclient.conf distrencommon.conf distrendaemon.conf distrenslave.conf
etc/distrendaemon.conf
Show inline comments
 
file renamed from etc/distrenserver.conf to etc/distrendaemon.conf
 
/*
 
 configuration for _this_ server
 
*/
 
me
 
daemon
 
{
 
  datadir = "/var/lib/distren"
 
  render_types = {"povray", "blender", "inkscape", "imagemagick", "dcraw"}
 
}
 

	
 
include("distrencommon.conf")
 

	
 
datadir = "/var/lib/distren"
 

	
 
listen {
 
       type = "unix"
 
       path = "/var/run/distrend.sock"
 
       port = "0770"
 
}
 
\ No newline at end of file
src/server/distrend.c
Show inline comments
 
@@ -68,25 +68,26 @@ struct general_info {
 

	
 
/*
 
  internally defined funcs's prototypes @TODO: Make all functions nice and proper */
 
void distrenjob_remove(struct distrenjob *head, struct distrenjob *bj);
 

	
 
struct distrenjob *distrenjob_get(struct distrenjob *head, jobnum_t jobnum);
 
void distrenjob_enqueue(struct distrenjob *head, struct distrenjob *job);
 
void mortition(struct distrenjob *head, struct distrenjob *job);
 
int makeJobDataXML(struct distrenjob *job);
 
int updateJobListXML(struct distrenjob *head);
 
int createQueueFromXML(struct distrenjob *head);
 
int reCreateQueueFromXML(struct distrenjob *head, xmlDocPtr doc, xmlNodePtr current);
 

	
 
void updateGeneralInfo();
 
void importGeneralInfo();
 

	
 
/* Global Vars, eliminate these */
 
jobnum_t jobnum = 0; // The next job number to create in the queue
 
int hcfjob; // Highest consecutively finished job
 
int highest_jobnum; // The job number of the most recently created job, this is used when creating new jobs
 

	
 

	
 
/* ********************** Functions ************************* */
 

	
 
/** Dumps all data in RAM to an xml file (such as current jobs, etc) which is parsed by start_data. Remember to invoke this before shutting down! */
 
int xml_dump()
 
{
 
@@ -129,87 +130,95 @@ void distrend_unlisten()
 
}
 
/**
 
   This is probably just a placeholder for remotio
 
*/
 
void remotio_send_to_client()
 
{
 
	// I am futile!
 
}
 

	
 
/** Fill variables after crash / shutdown from XML dumps */
 
int start_data(struct distrenjob *head)
 
{
 
  general_info.hibernate = 0;
 
  struct stat buffer;
 
  if(stat(SYSCONFDIR "/data.xml", &buffer) == 0){
 

	
 
    // @TODO: retrieve total_finished_jobs and total_finished_frames from xml file
 
  if(stat("general_info.xml", &buffer) == 0){
 
    importGeneralInfo();
 

	
 
    fprintf(stderr,"Parsing XML files and restoring previous state...\n");
 
    createQueueFromXML(head);
 
    return 1;
 
  }
 
  else{
 
    general_info.total_finished_jobs = 0;
 
    general_info.total_frames_rendered = 0;
 
    general_info.free_clients = 0;
 
    general_info.highest_jobnum = 0;
 
    general_info.jobs_in_queue = 0;
 
    general_info.rendering_clients = 0;
 
    general_info.hibernate = 0;
 
    fprintf(stderr,"Couldn't find XML dump, starting up fresh.\n");
 
    return 2;
 
  }
 
}
 

	
 
/** Finish-Setter: Sets a frame to the "completed" status.*/
 
void finish_frame(struct distrenjob *head, struct distrenjob *distrenjob, int frame)
 
{
 
  distrenjob->frameset[frame].status = FRAMESETSTATUS_DONE;
 
  distrenjob->total_render_time = distrenjob->total_render_time + (clock() - distrenjob[jobnum].frameset[frame].start_time);
 
  distrenjob->completed_frames++;
 
  distrenjob->assigned_frames--;
 
  general_info.total_frames_rendered++; // Increase total frames var for stats
 

	
 
  if(distrenjob->completed_frames == distrenjob->total_frames)
 
    {
 
      mortition(head, distrenjob);
 
    }
 

	
 
  updateGeneralInfo();
 
}
 

	
 
/** "mortition" check to see if a job is actually done by scanning the folder of the job to make sure all frames are present*/
 
void mortition(struct distrenjob *head, struct distrenjob *job)
 
{
 
  short int isJobDone;
 
  int counter;
 
  char *path_and_number;
 
  struct stat buffer;
 

	
 
  isJobDone = 1; // set isJobDone to true
 
  for(counter = 0; counter < job->total_frames; counter++)
 
    {
 
      _distren_asprintf(&path_and_number, "stor/job%d/out/%d.%s", job->jobnum, job->frameset[counter].num, job->output_format);
 
      if(stat(path_and_number, &buffer) == -1)
 
        {
 
          job->frameset[counter].status = FRAMESETSTATUS_UNASSIGNED;
 
          job->completed_frames--;
 
          general_info.total_frames_rendered--;
 
          isJobDone = 0; // if a missing frame is found, set isJobDone to false
 
        }
 
    }
 

	
 
  if(isJobDone) // if all frames were accounted for
 
    {
 
      distrenjob_remove(head, job);
 
      distrenjob_free(&job);
 
      general_info.jobs_in_queue--;
 
      updateJobListXML(head);
 
    }
 
  else{
 
    job->prev_frame_index = 0; // if the job isn't done, have frame_finder() start from the first frame, allowing it to see the frames that are now unassigned
 
  }
 

	
 
  updateGeneralInfo();
 
}
 

	
 
/** scans the frames of a job to initialize a job after server */
 
/* returns 1 if the job is completely done and there are no missing frames */
 
/* returns 0 if a missing frame is found */
 
int restoreJobState(struct distrenjob *job)
 
{
 
  short int isJobDone;
 
  int counter;
 
  char *path_and_number;
 
  struct stat buffer;
 

	
 
@@ -221,56 +230,68 @@ int restoreJobState(struct distrenjob *j
 
        {
 
          job->frameset[counter].status = FRAMESETSTATUS_ASSIGNED;
 
          job->completed_frames++;
 
        }
 
      else
 
        isJobDone = 0;
 
    }
 

	
 
  return isJobDone;
 
}
 

	
 
/** creates a structure from starting data, then calls another function to actually add that struct to the queue */
 
void prepare_distrenjob(struct distrenjob *head, int type, char *name, char *submitter, char *email, int priority, int start_frame, int end_frame, int width, int height)
 
int prepare_distrenjob(struct distrenjob *head, int type, char *name, char *submitter, char *email, int priority, int start_frame, int end_frame, int width, int height)
 
{
 
  int counter2;
 
  int counter;
 
  int tmp;
 

	
 
  struct distrenjob *distrenjob;
 
  distrenjob_new(&distrenjob);
 
  tmp = distrenjob_new(&distrenjob);
 
  if(tmp)
 
    return 1;
 

	
 
  distrenjob->type = 1;
 
  distrenjob->name = name;
 
  distrenjob->submitter = submitter;
 
  distrenjob->email = email;
 
  distrenjob->priority = priority;
 
  distrenjob->width = width;
 
  distrenjob->height = height;
 
  distrenjob->total_frames = (end_frame - start_frame + 1); // sets the total number of frames in animation for status purposes
 
  distrenjob->total_frames = (end_frame - start_frame + 1); /* sets the total number of frames in animation for status purposes */
 
  distrenjob->frameset = malloc(sizeof(struct frameset) * distrenjob->total_frames);
 
  if(!distrenjob->frameset)
 
    {
 
      distrenjob_free(&distrenjob);
 
      return 1;
 
    }
 

	
 
  /* prepares all the frames by setting that status to "unassigned" */
 
  counter2 = start_frame;
 
  for(counter = 0; counter <= (end_frame- start_frame + 1); counter++){
 
    distrenjob->frameset[counter].num = counter2;
 
    distrenjob->frameset[counter].status = FRAMESETSTATUS_UNASSIGNED;
 

	
 
    counter2++;
 
  }
 

	
 
  /* add job to queue */
 
  makeJobDataXML(distrenjob);
 
  distrenjob_enqueue(head, distrenjob);
 
  updateJobListXML(head);
 

	
 
  general_info.jobs_in_queue++;
 
  updateJobListXML(head);
 
  updateGeneralInfo();
 

	
 
  return 0;
 
}
 

	
 

	
 
/** distrenjob_enqueue: This function adds the job to the queue based on its priority */
 
void distrenjob_enqueue(struct distrenjob *head, struct distrenjob *job) {
 
  struct distrenjob *prev_job = head; // create pointer to previous job
 
  struct distrenjob *current_job;     // create pointer to current_job (which job is being compared to)
 

	
 
  // iterate through linked list of jobs
 
  for(current_job = head; current_job != NULL; current_job = current_job->next){
 
    if(current_job == NULL){ // if it has reached the end of the list, add job there
 
      current_job = job;
 
@@ -357,35 +378,32 @@ int find_jobframe(struct distrenjob *hea
 
  *job = distrenjob_ptr;
 
  *frame = &distrenjob_ptr->frameset[frame_counter];
 

	
 
  return 0;
 
}
 

	
 
/** Checks for dead, latent, or stale slaves */
 
void frame_watchdog(struct distrenjob *distrenjob_head)
 
{
 
  struct distrenjob *distrenjob_ptr;
 
  unsigned int counter;
 

	
 
  /*watchdog_forgiveness = seconds of forgiveness before frame is re-assigned */
 
  distrenjob_ptr = distrenjob_head;
 

	
 
  for(distrenjob_ptr = distrenjob_head; distrenjob_ptr; distrenjob_ptr = distrenjob_ptr->next)
 
  for(distrenjob_ptr = distrenjob_head->next; distrenjob_ptr; distrenjob_ptr = distrenjob_ptr->next)
 
  /* iterate through jobs */
 

	
 
    /* if the job has been started, checks by seeing if either to first or second frame has been started */
 
    if(distrenjob_ptr->frameset[0].status != FRAMESETSTATUS_UNASSIGNED || distrenjob_ptr->frameset[1].status != FRAMESETSTATUS_UNASSIGNED)
 
      /* iterate through all frames for this job: */
 
      for(counter = 0; counter < distrenjob_ptr->total_frames; counter ++)
 
      /* iterate through all frames for this job*/
 

	
 
	/*watchdog_forgiveness = seconds of forgiveness before frame is re-assigned: */
 
        if((distrenjob_ptr->frameset[counter].start_time + distrenjob_ptr->watchdog_forgiveness) < clock())
 
        /*
 
        If frame is not completed within the number of seconds specified by watchdog_forgiveness
 
        Then change the frame status to unassigned
 
        */
 
          distrenjob_ptr->frameset[counter].status = FRAMESETSTATUS_UNASSIGNED;
 
}
 

	
 
/**
 
   Finds a distrenjob struct based on the jobnum
 
   @arg jobnum job number to search for
 
   @return NULL on job doesn't exist
 
@@ -459,101 +477,199 @@ int distrend_do_config(int argc, char *a
 
   */
 
  tmp = xml2distrenjob(&distrenjob, "distrenjob.xml.example");
 
  if(tmp)
 
    fprintf(stderr, "xml2distrenjob() returned %d. Try to cd to distren/doc if you want to test out the xml2distrenjob() function. (This will only fix this error if the error is due to an inability of the xml library to access distrenjob.xml.example)\n\n", tmp);
 
  else
 
    fprintf(stderr, "using email ``%s'' for user ``%s'' -- reading in XML files and pulling data from them using libxml2+XPath works!!!\n", distrenjob->email, distrenjob->submitter);
 

	
 
  fprintf(stderr, "%s:%d running config\n", __FILE__, __LINE__);
 

	
 
  *config = malloc(sizeof(struct distrend_config));
 
  myopts[1].simple_value = &(*config)->datadir;
 

	
 
  options_init(argc, argv, &(*config)->mycfg, myopts, "server", &(*config)->options);
 
  options_init(argc, argv, &(*config)->mycfg, myopts, "daemon", &(*config)->options);
 

	
 
  fprintf(stderr, "using %s as datadir\n", (*config)->datadir);
 

	
 
  xmlcleanup();
 
  return 0;
 
}
 
int distrend_config_free(struct distrend_config *config)
 
{
 
  options_free(config->options);
 
  free(config);
 

	
 
  return 0;
 
}
 
/* ************************** XML Functions ************************* */
 

	
 
// writes the general_info.xml file which is a copy of the general_info structure
 
// except that it doesn't hold free_clients and rendering_clients
 
void updateGeneralInfo()
 
{
 
  xmlTextWriterPtr writer;
 
  char *tmp;
 

	
 
  writer = xmlNewTextWriterFilename("general_info.xml", 0);
 
  xmlTextWriterStartDocument(writer, NULL, "utf-8", NULL);
 

	
 
  xmlTextWriterStartElement(writer, (xmlChar*)"general_info");
 

	
 
  _distren_asprintf(&tmp, "%d", general_info.jobs_in_queue);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"jobs_in_queue", (xmlChar*)tmp);
 
  _distren_asprintf(&tmp, "%d", general_info.total_finished_jobs);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"total_finished_jobs", (xmlChar*)tmp);
 
  _distren_asprintf(&tmp, "%d", general_info.total_frames_rendered);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"total_frames_rendered", (xmlChar*)tmp);
 
  _distren_asprintf(&tmp, "%d", general_info.highest_jobnum);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"highest_jobnum", (xmlChar*)tmp);
 

	
 
  xmlTextWriterEndDocument(writer);
 
  xmlFreeTextWriter(writer);
 
}
 

	
 
// this reads the information from general_info.xml to the general_info structure
 
void importGeneralInfo()
 
{
 
  xmlDocPtr doc;
 
  xmlNodePtr cur;
 

	
 
  doc = xmlParseFile("general_info.xml");
 
  cur = xmlDocGetRootElement(doc);
 
  if (xmlStrcmp(cur->name, (xmlChar*)"general_info"))
 
    {
 
      fprintf(stderr, "xml document is wrong type");
 
      xmlFreeDoc(doc);
 
      return;
 
    }
 

	
 
  cur = cur->xmlChildrenNode;
 
  general_info.jobs_in_queue = atoi((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 
  cur = cur->next;
 
  general_info.total_finished_jobs = atoi((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 
  cur = cur->next;
 
  general_info.total_frames_rendered = atoi((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 
  cur = cur->next;
 
  general_info.highest_jobnum = atoi((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 

	
 
  general_info.hibernate = 0;
 
  general_info.free_clients = 0;
 
  general_info.rendering_clients = 0;
 

	
 
  xmlFreeDoc(doc);
 
}
 

	
 
// returns 1 on successful completion of xml file
 
// creates the xml file that slaves read from, and is used to restart distren
 
int makeJobDataXML(struct distrenjob *job)
 
{
 
  xmlTextWriterPtr writer;
 
  char *tmp; // temporarily holds strings to be given to the xml writer
 

	
 
  _distren_asprintf(&tmp, "stor/job%d/job_info.xml", job->jobnum);
 

	
 
  // create xml document at the location tmp with no compression
 
  writer = xmlNewTextWriterFilename(tmp, 0);
 
  xmlTextWriterStartDocument(writer, NULL, "UTF-8", NULL);
 
  xmlTextWriterStartDocument(writer, NULL, "utf-8", NULL);
 

	
 
  // write resolution element and add its attributes
 
  xmlTextWriterStartElement(writer, (xmlChar*)"distrenjob");
 
  // write distrenjob element and add its attributes
 
  xmlTextWriterStartElement(writer, (xmlChar*)"job");
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"name", (xmlChar*)job->name);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"submitter", (xmlChar*)job->submitter);
 
  _distren_asprintf(&tmp, "%d", job->priority);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"priority", (xmlChar*)tmp);
 

	
 
  // write resolution element and add its attributes
 
  xmlTextWriterStartElement(writer, (xmlChar*)"resolution");
 
  _distren_asprintf(&tmp, "%d", job->width);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"width", (xmlChar*)tmp);
 
  _distren_asprintf(&tmp, "%d", job->height);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"height", (xmlChar*)tmp);
 
  xmlTextWriterEndElement(writer);
 

	
 
  // write video element and its attributes
 
  xmlTextWriterStartElement(writer, (xmlChar*)"video");
 
  _distren_asprintf(&tmp, "%d", job->frameset[0].num);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"start_frame", (xmlChar*)tmp);
 
  _distren_asprintf(&tmp, "%d", job->frameset[(job->total_frames - 1)].num);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"end_fame", (xmlChar*)tmp);
 
  xmlTextWriterWriteAttribute(writer, (xmlChar*)"output_format", (xmlChar*)job->output_format);
 
  xmlTextWriterEndElement(writer);
 

	
 
  // write watchdog forgiveness element
 
  _distren_asprintf(&tmp, "%d", job->watchdog_forgiveness);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"wd_forgiveness", (xmlChar*)tmp);
 

	
 
  // write total_render_time element
 
  _distren_asprintf(&tmp, "%d", job->total_render_time);
 
  xmlTextWriterWriteElement(writer, (xmlChar*)"total_render_time", (xmlChar*)tmp);
 

	
 
  // end document
 
  xmlTextWriterEndDocument(writer);
 

	
 
  // free writer and save xml file to disk
 
  xmlFreeTextWriter(writer);
 

	
 
  return 1;
 
}
 

	
 
// extracts data from the xml created by above function and creates a job from it
 
// it returns a pointer to the created job
 
struct distrenjob *createJobFromXML(int job_number)
 
{
 
  xmlDocPtr doc;
 
  xmlNodePtr cur;
 
  char *file_name;
 
  struct distrenjob *distrenjob;
 
  int start_frame;
 
  int counter;
 
  int counter2;
 

	
 
  distrenjob_new(&distrenjob);
 

	
 
  _distren_asprintf(&file_name, "stor/job%d/job_info.xml", job_number);
 

	
 
  doc = xmlParseFile(file_name);
 
  cur = xmlDocGetRootElement(doc);
 

	
 
  distrenjob->name = (char*)xmlGetProp(cur, (xmlChar*)"name");
 
  distrenjob->submitter = (char*)xmlGetProp(cur, (xmlChar*)"submitter");
 
  distrenjob->priority = atoi((char*)xmlGetProp(cur, (xmlChar*)"priority"));
 

	
 
  cur = cur->xmlChildrenNode;
 
  distrenjob->width = atoi((char*)xmlGetProp(cur, (xmlChar*)"width"));
 
  distrenjob->height = atoi((char*)xmlGetProp(cur, (xmlChar*)"number"));
 

	
 
  cur = cur->next;
 
  start_frame = atoi((char*)xmlGetProp(cur, (xmlChar*)"start_frame"));
 
  distrenjob->total_frames = atoi((char*)xmlGetProp(cur, (xmlChar*)"end_frame")) - start_frame + 1;
 
  distrenjob->output_format = (char*)xmlGetProp(cur, (xmlChar*)"output_format");
 

	
 
  counter2 = start_frame;
 
  for(counter = 0; counter <= distrenjob->total_frames; counter++)
 
    {
 
      distrenjob->frameset[counter].num = counter2;
 
      distrenjob->frameset[counter].status = FRAMESETSTATUS_UNASSIGNED;
 

	
 
      counter2++;
 
    }
 

	
 
  cur = cur->next;
 
  distrenjob->watchdog_forgiveness = atoi((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 

	
 
  cur = cur->next;
 
  distrenjob->total_render_time = (time_t)atol((char*)xmlNodeListGetString(doc, cur->xmlChildrenNode, 1));
 

	
 
  xmlFreeDoc(doc);
 

	
 
  restoreJobState(distrenjob);
 
  return distrenjob;
 
}
 

	
 
// returns 1 if successful
 
// updates job_list.xml which lists all the jobs in the queue
 
int updateJobListXML(struct distrenjob *head)
 
{
 
  struct distrenjob *job;
 
  xmlTextWriterPtr writer;
 
  char *tmp;
 
  char *tmp2;
0 comments (0 inline, 0 general)