int makeflow_clean_file( struct dag *d, struct batch_queue *queue, struct dag_file *f, int silent)
{
if(!f)
return 1;
if(batch_fs_unlink(queue, f->filename) == 0) {
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_DELETE);
debug(D_MAKEFLOW_RUN, "File deleted %s\n", f->filename);
} else if(errno != ENOENT) {
if(f->state == DAG_FILE_STATE_EXPECT || dag_file_should_exist(f))
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_DELETE);
debug(D_MAKEFLOW_RUN, "Makeflow: Couldn't delete %s: %s\n", f->filename, strerror(errno));
return 1;
}
return 0;
}
void makeflow_clean(struct dag *d, struct batch_queue *queue, makeflow_clean_depth clean_depth)
{
struct dag_file *f;
char *name;
hash_table_firstkey(d->files);
while(hash_table_nextkey(d->files, &name, (void **) &f)) {
if(dag_file_is_source(f))
continue;
int silent = 1;
if(dag_file_should_exist(f))
silent = 0;
if(clean_depth == MAKEFLOW_CLEAN_ALL){
makeflow_file_clean(d, queue, f, silent);
} else if(set_lookup(d->outputs, f) && (clean_depth == MAKEFLOW_CLEAN_OUTPUTS)) {
makeflow_file_clean(d, queue, f, silent);
} else if(!set_lookup(d->outputs, f) && (clean_depth == MAKEFLOW_CLEAN_INTERMEDIATES)){
makeflow_file_clean(d, queue, f, silent);
}
}
struct dag_node *n;
for(n = d->nodes; n; n = n->next) {
/* If the node is a Makeflow job, then we should recursively call the *
* clean operation on it. */
if(n->nested_job) {
char *command = xxmalloc(sizeof(char) * (strlen(n->command) + 4));
sprintf(command, "%s -c", n->command);
/* XXX this should use the batch job interface for consistency */
makeflow_node_export_variables(d, n);
system(command);
free(command);
}
}
}
/** The clean_mode variable was added so that we could better print out error messages
* apply in the situation. Currently only used to silence node rerun checking.
*/
void makeflow_log_recover(struct dag *d, const char *filename, int verbose_mode, struct batch_queue *queue, makeflow_clean_depth clean_mode)
{
char *line, *name, file[MAX_BUFFER_SIZE];
int nodeid, state, jobid, file_state;
int first_run = 1;
struct dag_node *n;
struct dag_file *f;
struct stat buf;
timestamp_t previous_completion_time;
d->logfile = fopen(filename, "r");
if(d->logfile) {
int linenum = 0;
first_run = 0;
printf("recovering from log file %s...\n",filename);
while((line = get_line(d->logfile))) {
linenum++;
if(sscanf(line, "# %d %s %" SCNu64 "", &file_state, file, &previous_completion_time) == 3) {
f = dag_file_lookup_or_create(d, file);
f->state = file_state;
if(file_state == DAG_FILE_STATE_EXISTS){
d->completed_files += 1;
f->creation_logged = (time_t) (previous_completion_time / 1000000);
} else if(file_state == DAG_FILE_STATE_DELETE){
d->deleted_files += 1;
}
continue;
}
if(line[0] == '#')
continue;
if(sscanf(line, "%" SCNu64 " %d %d %d", &previous_completion_time, &nodeid, &state, &jobid) == 4) {
n = itable_lookup(d->node_table, nodeid);
if(n) {
n->state = state;
n->jobid = jobid;
/* Log timestamp is in microseconds, we need seconds for diff. */
n->previous_completion = (time_t) (previous_completion_time / 1000000);
continue;
}
}
fprintf(stderr, "makeflow: %s appears to be corrupted on line %d\n", filename, linenum);
exit(1);
}
fclose(d->logfile);
}
d->logfile = fopen(filename, "a");
if(!d->logfile) {
fprintf(stderr, "makeflow: couldn't open logfile %s: %s\n", filename, strerror(errno));
exit(1);
}
if(setvbuf(d->logfile, NULL, _IOLBF, BUFSIZ) != 0) {
fprintf(stderr, "makeflow: couldn't set line buffer on logfile %s: %s\n", filename, strerror(errno));
exit(1);
}
if(first_run && verbose_mode) {
struct dag_file *f;
struct dag_node *p;
for(n = d->nodes; n; n = n->next) {
/* Record node information to log */
fprintf(d->logfile, "# NODE\t%d\t%s\n", n->nodeid, n->command);
/* Record the node category to the log */
fprintf(d->logfile, "# CATEGORY\t%d\t%s\n", n->nodeid, n->category->label);
fprintf(d->logfile, "# SYMBOL\t%d\t%s\n", n->nodeid, n->category->label); /* also write the SYMBOL as alias of CATEGORY, deprecated. */
/* Record node parents to log */
fprintf(d->logfile, "# PARENTS\t%d", n->nodeid);
list_first_item(n->source_files);
while( (f = list_next_item(n->source_files)) ) {
p = f->created_by;
if(p)
fprintf(d->logfile, "\t%d", p->nodeid);
}
fputc('\n', d->logfile);
/* Record node inputs to log */
fprintf(d->logfile, "# SOURCES\t%d", n->nodeid);
list_first_item(n->source_files);
while( (f = list_next_item(n->source_files)) ) {
fprintf(d->logfile, "\t%s", f->filename);
}
fputc('\n', d->logfile);
/* Record node outputs to log */
fprintf(d->logfile, "# TARGETS\t%d", n->nodeid);
list_first_item(n->target_files);
while( (f = list_next_item(n->target_files)) ) {
fprintf(d->logfile, "\t%s", f->filename);
}
fputc('\n', d->logfile);
/* Record translated command to log */
fprintf(d->logfile, "# COMMAND\t%d\t%s\n", n->nodeid, n->command);
}
}
dag_count_states(d);
// Check for log consistency
if(!first_run) {
hash_table_firstkey(d->files);
while(hash_table_nextkey(d->files, &name, (void **) &f)) {
if(dag_file_should_exist(f) && !dag_file_is_source(f) && !(batch_fs_stat(queue, f->filename, &buf) >= 0)){
fprintf(stderr, "makeflow: %s is reported as existing, but does not exist.\n", f->filename);
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_UNKNOWN);
continue;
}
if(S_ISDIR(buf.st_mode))
continue;
if(dag_file_should_exist(f) && !dag_file_is_source(f) && difftime(buf.st_mtime, f->creation_logged) > 0) {
fprintf(stderr, "makeflow: %s is reported as existing, but has been modified (%" SCNu64 " ,%" SCNu64 ").\n", f->filename, (uint64_t)buf.st_mtime, (uint64_t)f->creation_logged);
makeflow_clean_file(d, queue, f, 0);
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_UNKNOWN);
}
}
}
int silent = 0;
if(clean_mode != MAKEFLOW_CLEAN_NONE)
silent = 1;
// Decide rerun tasks
if(!first_run) {
struct itable *rerun_table = itable_create(0);
for(n = d->nodes; n; n = n->next) {
makeflow_node_decide_rerun(rerun_table, d, n, silent);
}
itable_delete(rerun_table);
}
//Update file reference counts from nodes in log
for(n = d->nodes; n; n = n->next) {
if(n->state == DAG_NODE_STATE_COMPLETE)
{
struct dag_file *f;
list_first_item(n->source_files);
while((f = list_next_item(n->source_files)))
f->ref_count += -1;
}
}
}
int makeflow_clean(struct dag *d, struct batch_queue *queue, makeflow_clean_depth clean_depth)//, struct makeflow_wrapper *w, struct makeflow_monitor *m)
{
struct dag_file *f;
char *name;
hash_table_firstkey(d->files);
while(hash_table_nextkey(d->files, &name, (void **) &f)) {
int silent = 1;
if(dag_file_should_exist(f))
silent = 0;
/* We have a record of the file, but it is no longer created or used so delete */
if(dag_file_is_source(f) && dag_file_is_sink(f) && !set_lookup(d->inputs, f))
makeflow_clean_file(d, queue, f, silent);
if(dag_file_is_source(f)) {
if(f->source && (clean_depth == MAKEFLOW_CLEAN_CACHE || clean_depth == MAKEFLOW_CLEAN_ALL)) { /* this file is specified in the mountfile */
if(makeflow_clean_mount_target(f->filename)) {
fprintf(stderr, "Failed to remove %s!\n", f->filename);
return -1;
}
}
continue;
}
if(clean_depth == MAKEFLOW_CLEAN_ALL) {
makeflow_clean_file(d, queue, f, silent);
} else if(set_lookup(d->outputs, f) && (clean_depth == MAKEFLOW_CLEAN_OUTPUTS)) {
makeflow_clean_file(d, queue, f, silent);
} else if(!set_lookup(d->outputs, f) && (clean_depth == MAKEFLOW_CLEAN_INTERMEDIATES)){
makeflow_clean_file(d, queue, f, silent);
}
}
/* clean up the cache dir created due to the usage of mountfile */
if(clean_depth == MAKEFLOW_CLEAN_CACHE || clean_depth == MAKEFLOW_CLEAN_ALL) {
if(d->cache_dir && unlink_recursive(d->cache_dir)) {
fprintf(stderr, "Failed to clean up the cache dir (%s) created due to the usage of the mountfile!\n", d->cache_dir);
dag_mount_clean(d);
return -1;
}
dag_mount_clean(d);
}
struct dag_node *n;
for(n = d->nodes; n; n = n->next) {
/* If the node is a Makeflow job, then we should recursively call the *
* clean operation on it. */
if(n->nested_job) {
char *command = xxmalloc(sizeof(char) * (strlen(n->command) + 4));
sprintf(command, "%s -c", n->command);
/* XXX this should use the batch job interface for consistency */
makeflow_node_export_variables(d, n);
system(command);
free(command);
}
}
return 0;
}
